Merge branch 'master' into partition-advanced-widgets

Amp-Thread-ID: https://ampcode.com/threads/T-019c3236-1417-74aa-82a3-bcb365fbe9d1

.coderabbit.yaml

@@ -1,127 +0,0 @@
-# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
-language: "en-US"
-early_access: false
-tone_instructions: "Only comment on issues introduced by this PR's changes. Do not flag pre-existing problems in moved, re-indented, or reformatted code."
-
-reviews:
-  profile: "chill"
-  request_changes_workflow: false
-  high_level_summary: false
-  poem: false
-  review_status: false
-  review_details: false
-  commit_status: true
-  collapse_walkthrough: true
-  changed_files_summary: false
-  sequence_diagrams: false
-  estimate_code_review_effort: false
-  assess_linked_issues: false
-  related_issues: false
-  related_prs: false
-  suggested_labels: false
-  auto_apply_labels: false
-  suggested_reviewers: false
-  auto_assign_reviewers: false
-  in_progress_fortune: false
-  enable_prompt_for_ai_agents: true
-
-  path_filters:
-    - "!comfy_api_nodes/apis/**"
-    - "!**/generated/*.pyi"
-    - "!.ci/**"
-    - "!script_examples/**"
-    - "!**/__pycache__/**"
-    - "!**/*.ipynb"
-    - "!**/*.png"
-    - "!**/*.bat"
-
-  path_instructions:
-    - path: "**"
-      instructions: |
-        IMPORTANT: Only comment on issues directly introduced by this PR's code changes.
-        Do NOT flag pre-existing issues in code that was merely moved, re-indented,
-        de-indented, or reformatted without logic changes. If code appears in the diff
-        only due to whitespace or structural reformatting (e.g., removing a `with:` block),
-        treat it as unchanged. Contributors should not feel obligated to address
-        pre-existing issues outside the scope of their contribution.
-    - path: "comfy/**"
-      instructions: |
-        Core ML/diffusion engine. Focus on:
-        - Backward compatibility (breaking changes affect all custom nodes)
-        - Memory management and GPU resource handling
-        - Performance implications in hot paths
-        - Thread safety for concurrent execution
-    - path: "comfy_api_nodes/**"
-      instructions: |
-        Third-party API integration nodes. Focus on:
-        - No hardcoded API keys or secrets
-        - Proper error handling for API failures (timeouts, rate limits, auth errors)
-        - Correct Pydantic model usage
-        - Security of user data passed to external APIs
-    - path: "comfy_extras/**"
-      instructions: |
-        Community-contributed extra nodes. Focus on:
-        - Consistency with node patterns (INPUT_TYPES, RETURN_TYPES, FUNCTION, CATEGORY)
-        - No breaking changes to existing node interfaces
-    - path: "comfy_execution/**"
-      instructions: |
-        Execution engine (graph execution, caching, jobs). Focus on:
-        - Caching correctness
-        - Concurrent execution safety
-        - Graph validation edge cases
-    - path: "nodes.py"
-      instructions: |
-        Core node definitions (2500+ lines). Focus on:
-        - Backward compatibility of NODE_CLASS_MAPPINGS
-        - Consistency of INPUT_TYPES return format
-    - path: "alembic_db/**"
-      instructions: |
-        Database migrations. Focus on:
-        - Migration safety and rollback support
-        - Data preservation during schema changes
-
-  auto_review:
-    enabled: true
-    auto_incremental_review: true
-    drafts: false
-    ignore_title_keywords:
-      - "WIP"
-      - "DO NOT REVIEW"
-      - "DO NOT MERGE"
-
-  finishing_touches:
-    docstrings:
-      enabled: false
-    unit_tests:
-      enabled: false
-
-  tools:
-    ruff:
-      enabled: false
-    pylint:
-      enabled: false
-    flake8:
-      enabled: false
-    gitleaks:
-      enabled: true
-    shellcheck:
-      enabled: false
-    markdownlint:
-      enabled: false
-    yamllint:
-      enabled: false
-    languagetool:
-      enabled: false
-    github-checks:
-      enabled: true
-      timeout_ms: 90000
-    ast-grep:
-      essential_rules: true
-
-chat:
-  auto_reply: true
-
-knowledge_base:
-  opt_out: false
-  learnings:
-    scope: "auto"

README.md

@@ -229,9 +229,9 @@ AMD users can install rocm and pytorch with pip if you don't have it already ins
 
 ```pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.1```
 
-This is the command to install the nightly with ROCm 7.2 which might have some performance improvements:
+This is the command to install the nightly with ROCm 7.1 which might have some performance improvements:
 
-```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.2```
+```pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.1```
 
 
 ### AMD GPUs (Experimental: Windows and Linux), RDNA 3, 3.5 and 4 only.

app/node_replace_manager.py

@@ -46,8 +46,6 @@ class NodeReplaceManager:
         connections: dict[str, list[tuple[str, str, int]]] = {}
         need_replacement: set[str] = set()
         for node_number, node_struct in prompt.items():
-            if "class_type" not in node_struct or "inputs" not in node_struct:
-                continue
             class_type = node_struct["class_type"]
             # need replacement if not in NODE_CLASS_MAPPINGS and has replacement
             if class_type not in nodes.NODE_CLASS_MAPPINGS.keys() and self.has_replacement(class_type):

app/subgraph_manager.py

@@ -53,7 +53,7 @@ class SubgraphManager:
         return entry_id, entry
 
     async def load_entry_data(self, entry: SubgraphEntry):
-        with open(entry['path'], 'r', encoding='utf-8') as f:
+        with open(entry['path'], 'r') as f:
             entry['data'] = f.read()
         return entry
 

blueprints/.glsl/Brightness_and_Contrast_1.frag

@@ -1,44 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0; // Brightness slider -100..100
-uniform float u_float1; // Contrast slider -100..100
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const float MID_GRAY = 0.18;  // 18% reflectance
-
-// sRGB gamma 2.2 approximation
-vec3 srgbToLinear(vec3 c) {
-    return pow(max(c, 0.0), vec3(2.2));
-}
-
-vec3 linearToSrgb(vec3 c) {
-    return pow(max(c, 0.0), vec3(1.0/2.2));
-}
-
-float mapBrightness(float b) {
-    return clamp(b / 100.0, -1.0, 1.0);
-}
-
-float mapContrast(float c) {
-    return clamp(c / 100.0 + 1.0, 0.0, 2.0);
-}
-
-void main() {
-    vec4 orig = texture(u_image0, v_texCoord);
-
-    float brightness = mapBrightness(u_float0);
-    float contrast   = mapContrast(u_float1);
-
-    vec3 lin = srgbToLinear(orig.rgb);
-
-    lin = (lin - MID_GRAY) * contrast + brightness + MID_GRAY;
-
-    // Convert back to sRGB
-    vec3 result = linearToSrgb(clamp(lin, 0.0, 1.0));
-
-    fragColor = vec4(result, orig.a);
-}

blueprints/.glsl/Chromatic_Aberration_16.frag

@@ -1,72 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Mode
-uniform float u_float0;  // Amount (0 to 100)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int MODE_LINEAR   = 0;
-const int MODE_RADIAL   = 1;
-const int MODE_BARREL   = 2;
-const int MODE_SWIRL    = 3;
-const int MODE_DIAGONAL = 4;
-
-const float AMOUNT_SCALE = 0.0005;
-const float RADIAL_MULT = 4.0;
-const float BARREL_MULT = 8.0;
-const float INV_SQRT2 = 0.70710678118;
-
-void main() {
-    vec2 uv = v_texCoord;
-    vec4 original = texture(u_image0, uv);
-
-    float amount = u_float0 * AMOUNT_SCALE;
-
-    if (amount < 0.000001) {
-        fragColor = original;
-        return;
-    }
-
-    // Aspect-corrected coordinates for circular effects
-    float aspect = u_resolution.x / u_resolution.y;
-    vec2 centered = uv - 0.5;
-    vec2 corrected = vec2(centered.x * aspect, centered.y);
-    float r = length(corrected);
-    vec2 dir = r > 0.0001 ? corrected / r : vec2(0.0);
-    vec2 offset = vec2(0.0);
-
-    if (u_int0 == MODE_LINEAR) {
-        // Horizontal shift (no aspect correction needed)
-        offset = vec2(amount, 0.0);
-    }
-    else if (u_int0 == MODE_RADIAL) {
-        // Outward from center, stronger at edges
-        offset = dir * r * amount * RADIAL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_BARREL) {
-        // Lens distortion simulation (r² falloff)
-        offset = dir * r * r * amount * BARREL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_SWIRL) {
-        // Perpendicular to radial (rotational aberration)
-        vec2 perp = vec2(-dir.y, dir.x);
-        offset = perp * r * amount * RADIAL_MULT;
-        offset.x /= aspect;  // Convert back to UV space
-    }
-    else if (u_int0 == MODE_DIAGONAL) {
-        // 45° offset (no aspect correction needed)
-        offset = vec2(amount, amount) * INV_SQRT2;
-    }
-    
-    float red = texture(u_image0, uv + offset).r;
-    float green = original.g;
-    float blue = texture(u_image0, uv - offset).b;
-    
-    fragColor = vec4(red, green, blue, original.a);
-}

blueprints/.glsl/Color_Adjustment_15.frag

@@ -1,78 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0; // temperature (-100 to 100)
-uniform float u_float1; // tint (-100 to 100)
-uniform float u_float2; // vibrance (-100 to 100)
-uniform float u_float3; // saturation (-100 to 100)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const float INPUT_SCALE = 0.01;
-const float TEMP_TINT_PRIMARY = 0.3;
-const float TEMP_TINT_SECONDARY = 0.15;
-const float VIBRANCE_BOOST = 2.0;
-const float SATURATION_BOOST = 2.0;
-const float SKIN_PROTECTION = 0.5;
-const float EPSILON = 0.001;
-const vec3 LUMA_WEIGHTS = vec3(0.299, 0.587, 0.114);
-
-void main() {
-    vec4 tex = texture(u_image0, v_texCoord);
-    vec3 color = tex.rgb;
-    
-    // Scale inputs: -100/100 → -1/1
-    float temperature = u_float0 * INPUT_SCALE;
-    float tint = u_float1 * INPUT_SCALE;
-    float vibrance = u_float2 * INPUT_SCALE;
-    float saturation = u_float3 * INPUT_SCALE;
-    
-    // Temperature (warm/cool): positive = warm, negative = cool
-    color.r += temperature * TEMP_TINT_PRIMARY;
-    color.b -= temperature * TEMP_TINT_PRIMARY;
-    
-    // Tint (green/magenta): positive = green, negative = magenta
-    color.g += tint * TEMP_TINT_PRIMARY;
-    color.r -= tint * TEMP_TINT_SECONDARY;
-    color.b -= tint * TEMP_TINT_SECONDARY;
-    
-    // Single clamp after temperature/tint
-    color = clamp(color, 0.0, 1.0);
-    
-    // Vibrance with skin protection
-    if (vibrance != 0.0) {
-        float maxC = max(color.r, max(color.g, color.b));
-        float minC = min(color.r, min(color.g, color.b));
-        float sat = maxC - minC;
-        float gray = dot(color, LUMA_WEIGHTS);
-        
-        if (vibrance < 0.0) {
-            // Desaturate: -100 → gray
-            color = mix(vec3(gray), color, 1.0 + vibrance);
-        } else {
-            // Boost less saturated colors more
-            float vibranceAmt = vibrance * (1.0 - sat);
-            
-            // Branchless skin tone protection
-            float isWarmTone = step(color.b, color.g) * step(color.g, color.r);
-            float warmth = (color.r - color.b) / max(maxC, EPSILON);
-            float skinTone = isWarmTone * warmth * sat * (1.0 - sat);
-            vibranceAmt *= (1.0 - skinTone * SKIN_PROTECTION);
-            
-            color = mix(vec3(gray), color, 1.0 + vibranceAmt * VIBRANCE_BOOST);
-        }
-    }
-    
-    // Saturation
-    if (saturation != 0.0) {
-        float gray = dot(color, LUMA_WEIGHTS);
-        float satMix = saturation < 0.0
-            ? 1.0 + saturation                      // -100 → gray
-            : 1.0 + saturation * SATURATION_BOOST;  // +100 → 3x boost
-        color = mix(vec3(gray), color, satMix);
-    }
-    
-    fragColor = vec4(clamp(color, 0.0, 1.0), tex.a);
-}

blueprints/.glsl/Edge-Preserving_Blur_128.frag

@@ -1,94 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform float u_float0;   // Blur radius (0–20, default ~5)
-uniform float u_float1;   // Edge threshold (0–100, default ~30)
-uniform int u_int0;       // Step size (0/1 = every pixel, 2+ = skip pixels)
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int MAX_RADIUS = 20;
-const float EPSILON = 0.0001;
-
-// Perceptual luminance
-float getLuminance(vec3 rgb) {
-    return dot(rgb, vec3(0.299, 0.587, 0.114));
-}
-
-vec4 bilateralFilter(vec2 uv, vec2 texelSize, int radius,
-                     float sigmaSpatial, float sigmaColor)
-{
-    vec4 center = texture(u_image0, uv);
-    vec3 centerRGB = center.rgb;
-
-    float invSpatial2 = -0.5 / (sigmaSpatial * sigmaSpatial);
-    float invColor2   = -0.5 / (sigmaColor * sigmaColor + EPSILON);
-
-    vec3 sumRGB = vec3(0.0);
-    float sumWeight = 0.0;
-
-    int step = max(u_int0, 1);
-    float radius2 = float(radius * radius);
-
-    for (int dy = -MAX_RADIUS; dy <= MAX_RADIUS; dy++) {
-        if (dy < -radius || dy > radius) continue;
-        if (abs(dy) % step != 0) continue;
-
-        for (int dx = -MAX_RADIUS; dx <= MAX_RADIUS; dx++) {
-            if (dx < -radius || dx > radius) continue;
-            if (abs(dx) % step != 0) continue;
-
-            vec2 offset = vec2(float(dx), float(dy));
-            float dist2 = dot(offset, offset);
-            if (dist2 > radius2) continue;
-
-            vec3 sampleRGB = texture(u_image0, uv + offset * texelSize).rgb;
-
-            // Spatial Gaussian
-            float spatialWeight = exp(dist2 * invSpatial2);
-
-            // Perceptual color distance (weighted RGB)
-            vec3 diff = sampleRGB - centerRGB;
-            float colorDist = dot(diff * diff, vec3(0.299, 0.587, 0.114));
-            float colorWeight = exp(colorDist * invColor2);
-
-            float w = spatialWeight * colorWeight;
-            sumRGB += sampleRGB * w;
-            sumWeight += w;
-        }
-    }
-
-    vec3 resultRGB = sumRGB / max(sumWeight, EPSILON);
-    return vec4(resultRGB, center.a); // preserve center alpha
-}
-
-void main() {
-    vec2 texelSize = 1.0 / vec2(textureSize(u_image0, 0));
-
-    float radiusF = clamp(u_float0, 0.0, float(MAX_RADIUS));
-    int radius = int(radiusF + 0.5);
-
-    if (radius == 0) {
-        fragColor = texture(u_image0, v_texCoord);
-        return;
-    }
-
-    // Edge threshold → color sigma
-    // Squared curve for better low-end control
-    float t = clamp(u_float1, 0.0, 100.0) / 100.0;
-    t *= t;
-    float sigmaColor = mix(0.01, 0.5, t);
-
-    // Spatial sigma tied to radius
-    float sigmaSpatial = max(radiusF * 0.75, 0.5);
-
-    fragColor = bilateralFilter(
-        v_texCoord,
-        texelSize,
-        radius,
-        sigmaSpatial,
-        sigmaColor
-    );
-}

blueprints/.glsl/Film_Grain_15.frag

@@ -1,124 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0; // grain amount      [0.0 – 1.0]   typical: 0.2–0.8
-uniform float u_float1; // grain size        [0.3 – 3.0]   lower = finer grain
-uniform float u_float2; // color amount      [0.0 – 1.0]   0 = monochrome, 1 = RGB grain
-uniform float u_float3; // luminance bias    [0.0 – 1.0]   0 = uniform, 1 = shadows only
-uniform int   u_int0;   // noise mode        [0 or 1]      0 = smooth, 1 = grainy
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-// High-quality integer hash (pcg-like)
-uint pcg(uint v) {
-    uint state = v * 747796405u + 2891336453u;
-    uint word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;
-    return (word >> 22u) ^ word;
-}
-
-// 2D -> 1D hash input
-uint hash2d(uvec2 p) {
-    return pcg(p.x + pcg(p.y));
-}
-
-// Hash to float [0, 1]
-float hashf(uvec2 p) {
-    return float(hash2d(p)) / float(0xffffffffu);
-}
-
-// Hash to float with offset (for RGB channels)
-float hashf(uvec2 p, uint offset) {
-    return float(pcg(hash2d(p) + offset)) / float(0xffffffffu);
-}
-
-// Convert uniform [0,1] to roughly Gaussian distribution
-// Using simple approximation: average of multiple samples
-float toGaussian(uvec2 p) {
-    float sum = hashf(p, 0u) + hashf(p, 1u) + hashf(p, 2u) + hashf(p, 3u);
-    return (sum - 2.0) * 0.7;  // Centered, scaled
-}
-
-float toGaussian(uvec2 p, uint offset) {
-    float sum = hashf(p, offset) + hashf(p, offset + 1u) 
-              + hashf(p, offset + 2u) + hashf(p, offset + 3u);
-    return (sum - 2.0) * 0.7;
-}
-
-// Smooth noise with better interpolation
-float smoothNoise(vec2 p) {
-    vec2 i = floor(p);
-    vec2 f = fract(p);
-    
-    // Quintic interpolation (less banding than cubic)
-    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);
-    
-    uvec2 ui = uvec2(i);
-    float a = toGaussian(ui);
-    float b = toGaussian(ui + uvec2(1u, 0u));
-    float c = toGaussian(ui + uvec2(0u, 1u));
-    float d = toGaussian(ui + uvec2(1u, 1u));
-    
-    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);
-}
-
-float smoothNoise(vec2 p, uint offset) {
-    vec2 i = floor(p);
-    vec2 f = fract(p);
-    
-    f = f * f * f * (f * (f * 6.0 - 15.0) + 10.0);
-    
-    uvec2 ui = uvec2(i);
-    float a = toGaussian(ui, offset);
-    float b = toGaussian(ui + uvec2(1u, 0u), offset);
-    float c = toGaussian(ui + uvec2(0u, 1u), offset);
-    float d = toGaussian(ui + uvec2(1u, 1u), offset);
-    
-    return mix(mix(a, b, f.x), mix(c, d, f.x), f.y);
-}
-
-void main() {
-    vec4 color = texture(u_image0, v_texCoord);
-    
-    // Luminance (Rec.709)
-    float luma = dot(color.rgb, vec3(0.2126, 0.7152, 0.0722));
-    
-    // Grain UV (resolution-independent)
-    vec2 grainUV = v_texCoord * u_resolution / max(u_float1, 0.01);
-    uvec2 grainPixel = uvec2(grainUV);
-    
-    float g;
-    vec3 grainRGB;
-    
-    if (u_int0 == 1) {
-        // Grainy mode: pure hash noise (no interpolation = no banding)
-        g = toGaussian(grainPixel);
-        grainRGB = vec3(
-            toGaussian(grainPixel, 100u),
-            toGaussian(grainPixel, 200u),
-            toGaussian(grainPixel, 300u)
-        );
-    } else {
-        // Smooth mode: interpolated with quintic curve
-        g = smoothNoise(grainUV);
-        grainRGB = vec3(
-            smoothNoise(grainUV, 100u),
-            smoothNoise(grainUV, 200u),
-            smoothNoise(grainUV, 300u)
-        );
-    }
-    
-    // Luminance weighting (less grain in highlights)
-    float lumWeight = mix(1.0, 1.0 - luma, clamp(u_float3, 0.0, 1.0));
-    
-    // Strength
-    float strength = u_float0 * 0.15;
-    
-    // Color vs monochrome grain
-    vec3 grainColor = mix(vec3(g), grainRGB, clamp(u_float2, 0.0, 1.0));
-    
-    color.rgb += grainColor * strength * lumWeight;
-    fragColor0 = vec4(clamp(color.rgb, 0.0, 1.0), color.a);
-}

blueprints/.glsl/Glow_30.frag

@@ -1,133 +0,0 @@
-#version 300 es
-precision mediump float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Blend mode
-uniform int u_int1;      // Color tint
-uniform float u_float0;  // Intensity
-uniform float u_float1;  // Radius
-uniform float u_float2;  // Threshold
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-const int BLEND_ADD      = 0;
-const int BLEND_SCREEN   = 1;
-const int BLEND_SOFT     = 2;
-const int BLEND_OVERLAY  = 3;
-const int BLEND_LIGHTEN  = 4;
-
-const float GOLDEN_ANGLE = 2.39996323;
-const int MAX_SAMPLES = 48;
-const vec3 LUMA = vec3(0.299, 0.587, 0.114);
-
-float hash(vec2 p) {
-    p = fract(p * vec2(123.34, 456.21));
-    p += dot(p, p + 45.32);
-    return fract(p.x * p.y);
-}
-
-vec3 hexToRgb(int h) {
-    return vec3(
-        float((h >> 16) & 255),
-        float((h >> 8) & 255),
-        float(h & 255)
-    ) * (1.0 / 255.0);
-}
-
-vec3 blend(vec3 base, vec3 glow, int mode) {
-    if (mode == BLEND_SCREEN) {
-        return 1.0 - (1.0 - base) * (1.0 - glow);
-    }
-    if (mode == BLEND_SOFT) {
-        return mix(
-            base - (1.0 - 2.0 * glow) * base * (1.0 - base),
-            base + (2.0 * glow - 1.0) * (sqrt(base) - base),
-            step(0.5, glow)
-        );
-    }
-    if (mode == BLEND_OVERLAY) {
-        return mix(
-            2.0 * base * glow,
-            1.0 - 2.0 * (1.0 - base) * (1.0 - glow),
-            step(0.5, base)
-        );
-    }
-    if (mode == BLEND_LIGHTEN) {
-        return max(base, glow);
-    }
-    return base + glow;
-}
-
-void main() {
-    vec4 original = texture(u_image0, v_texCoord);
-    
-    float intensity = u_float0 * 0.05;
-    float radius = u_float1 * u_float1 * 0.012;
-    
-    if (intensity < 0.001 || radius < 0.1) {
-        fragColor = original;
-        return;
-    }
-    
-    float threshold = 1.0 - u_float2 * 0.01;
-    float t0 = threshold - 0.15;
-    float t1 = threshold + 0.15;
-    
-    vec2 texelSize = 1.0 / u_resolution;
-    float radius2 = radius * radius;
-    
-    float sampleScale = clamp(radius * 0.75, 0.35, 1.0);
-    int samples = int(float(MAX_SAMPLES) * sampleScale);
-    
-    float noise = hash(gl_FragCoord.xy);
-    float angleOffset = noise * GOLDEN_ANGLE;
-    float radiusJitter = 0.85 + noise * 0.3;
-    
-    float ca = cos(GOLDEN_ANGLE);
-    float sa = sin(GOLDEN_ANGLE);
-    vec2 dir = vec2(cos(angleOffset), sin(angleOffset));
-    
-    vec3 glow = vec3(0.0);
-    float totalWeight = 0.0;
-    
-    // Center tap
-    float centerMask = smoothstep(t0, t1, dot(original.rgb, LUMA));
-    glow += original.rgb * centerMask * 2.0;
-    totalWeight += 2.0;
-    
-    for (int i = 1; i < MAX_SAMPLES; i++) {
-        if (i >= samples) break;
-        
-        float fi = float(i);
-        float dist = sqrt(fi / float(samples)) * radius * radiusJitter;
-        
-        vec2 offset = dir * dist * texelSize;
-        vec3 c = texture(u_image0, v_texCoord + offset).rgb;
-        float mask = smoothstep(t0, t1, dot(c, LUMA));
-        
-        float w = 1.0 - (dist * dist) / (radius2 * 1.5);
-        w = max(w, 0.0);
-        w *= w;
-        
-        glow += c * mask * w;
-        totalWeight += w;
-        
-        dir = vec2(
-            dir.x * ca - dir.y * sa,
-            dir.x * sa + dir.y * ca
-        );
-    }
-    
-    glow *= intensity / max(totalWeight, 0.001);
-    
-    if (u_int1 > 0) {
-        glow *= hexToRgb(u_int1);
-    }
-    
-    vec3 result = blend(original.rgb, glow, u_int0);
-    result += (noise - 0.5) * (1.0 / 255.0);
-    
-    fragColor = vec4(clamp(result, 0.0, 1.0), original.a);
-}

blueprints/.glsl/Hue_and_Saturation_1.frag

@@ -1,222 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform int u_int0;      // Mode: 0=Master, 1=Reds, 2=Yellows, 3=Greens, 4=Cyans, 5=Blues, 6=Magentas, 7=Colorize
-uniform int u_int1;      // Color Space: 0=HSL, 1=HSB/HSV
-uniform float u_float0;  // Hue (-180 to 180)
-uniform float u_float1;  // Saturation (-100 to 100)
-uniform float u_float2;  // Lightness/Brightness (-100 to 100)
-uniform float u_float3;  // Overlap (0 to 100) - feathering between adjacent color ranges
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-// Color range modes
-const int MODE_MASTER   = 0;
-const int MODE_RED      = 1;
-const int MODE_YELLOW   = 2;
-const int MODE_GREEN    = 3;
-const int MODE_CYAN     = 4;
-const int MODE_BLUE     = 5;
-const int MODE_MAGENTA  = 6;
-const int MODE_COLORIZE = 7;
-
-// Color space modes
-const int COLORSPACE_HSL = 0;
-const int COLORSPACE_HSB = 1;
-
-const float EPSILON = 0.0001;
-
-//=============================================================================
-// RGB <-> HSL Conversions
-//=============================================================================
-
-vec3 rgb2hsl(vec3 c) {
-    float maxC = max(max(c.r, c.g), c.b);
-    float minC = min(min(c.r, c.g), c.b);
-    float delta = maxC - minC;
-
-    float h = 0.0;
-    float s = 0.0;
-    float l = (maxC + minC) * 0.5;
-
-    if (delta > EPSILON) {
-        s = l < 0.5
-            ? delta / (maxC + minC)
-            : delta / (2.0 - maxC - minC);
-
-        if (maxC == c.r) {
-            h = (c.g - c.b) / delta + (c.g < c.b ? 6.0 : 0.0);
-        } else if (maxC == c.g) {
-            h = (c.b - c.r) / delta + 2.0;
-        } else {
-            h = (c.r - c.g) / delta + 4.0;
-        }
-        h /= 6.0;
-    }
-
-    return vec3(h, s, l);
-}
-
-float hue2rgb(float p, float q, float t) {
-    t = fract(t);
-    if (t < 1.0/6.0) return p + (q - p) * 6.0 * t;
-    if (t < 0.5)       return q;
-    if (t < 2.0/3.0)   return p + (q - p) * (2.0/3.0 - t) * 6.0;
-    return p;
-}
-
-vec3 hsl2rgb(vec3 hsl) {
-    if (hsl.y < EPSILON) return vec3(hsl.z);
-
-    float q = hsl.z < 0.5
-        ? hsl.z * (1.0 + hsl.y)
-        : hsl.z + hsl.y - hsl.z * hsl.y;
-    float p = 2.0 * hsl.z - q;
-
-    return vec3(
-        hue2rgb(p, q, hsl.x + 1.0/3.0),
-        hue2rgb(p, q, hsl.x),
-        hue2rgb(p, q, hsl.x - 1.0/3.0)
-    );
-}
-
-vec3 rgb2hsb(vec3 c) {
-    float maxC = max(max(c.r, c.g), c.b);
-    float minC = min(min(c.r, c.g), c.b);
-    float delta = maxC - minC;
-
-    float h = 0.0;
-    float s = (maxC > EPSILON) ? delta / maxC : 0.0;
-    float b = maxC;
-
-    if (delta > EPSILON) {
-        if (maxC == c.r) {
-            h = (c.g - c.b) / delta + (c.g < c.b ? 6.0 : 0.0);
-        } else if (maxC == c.g) {
-            h = (c.b - c.r) / delta + 2.0;
-        } else {
-            h = (c.r - c.g) / delta + 4.0;
-        }
-        h /= 6.0;
-    }
-
-    return vec3(h, s, b);
-}
-
-vec3 hsb2rgb(vec3 hsb) {
-    vec3 rgb = clamp(abs(mod(hsb.x * 6.0 + vec3(0.0, 4.0, 2.0), 6.0) - 3.0) - 1.0, 0.0, 1.0);
-    return hsb.z * mix(vec3(1.0), rgb, hsb.y);
-}
-
-//=============================================================================
-// Color Range Weight Calculation
-//=============================================================================
-
-float hueDistance(float a, float b) {
-    float d = abs(a - b);
-    return min(d, 1.0 - d);
-}
-
-float getHueWeight(float hue, float center, float overlap) {
-    float baseWidth = 1.0 / 6.0;
-    float feather = baseWidth * overlap;
-
-    float d = hueDistance(hue, center);
-
-    float inner = baseWidth * 0.5;
-    float outer = inner + feather;
-
-    return 1.0 - smoothstep(inner, outer, d);
-}
-
-float getModeWeight(float hue, int mode, float overlap) {
-    if (mode == MODE_MASTER || mode == MODE_COLORIZE) return 1.0;
-
-    if (mode == MODE_RED) {
-        return max(
-            getHueWeight(hue, 0.0, overlap),
-            getHueWeight(hue, 1.0, overlap)
-        );
-    }
-
-    float center = float(mode - 1) / 6.0;
-    return getHueWeight(hue, center, overlap);
-}
-
-//=============================================================================
-// Adjustment Functions
-//=============================================================================
-
-float adjustLightness(float l, float amount) {
-    return amount > 0.0
-        ? l + (1.0 - l) * amount
-        : l + l * amount;
-}
-
-float adjustBrightness(float b, float amount) {
-    return clamp(b + amount, 0.0, 1.0);
-}
-
-float adjustSaturation(float s, float amount) {
-    return amount > 0.0
-        ? s + (1.0 - s) * amount
-        : s + s * amount;
-}
-
-vec3 colorize(vec3 rgb, float hue, float sat, float light) {
-    float lum = dot(rgb, vec3(0.299, 0.587, 0.114));
-    float l = adjustLightness(lum, light);
-
-    vec3 hsl = vec3(fract(hue), clamp(sat, 0.0, 1.0), clamp(l, 0.0, 1.0));
-    return hsl2rgb(hsl);
-}
-
-//=============================================================================
-// Main
-//=============================================================================
-
-void main() {
-    vec4 original = texture(u_image0, v_texCoord);
-
-    float hueShift   = u_float0 / 360.0;   // -180..180 -> -0.5..0.5
-    float satAmount  = u_float1 / 100.0;   // -100..100 -> -1..1
-    float lightAmount= u_float2 / 100.0;   // -100..100 -> -1..1
-    float overlap    = u_float3 / 100.0;   // 0..100 -> 0..1
-
-    vec3 result;
-
-    if (u_int0 == MODE_COLORIZE) {
-        result = colorize(original.rgb, hueShift, satAmount, lightAmount);
-        fragColor = vec4(result, original.a);
-        return;
-    }
-
-    vec3 hsx = (u_int1 == COLORSPACE_HSL)
-        ? rgb2hsl(original.rgb)
-        : rgb2hsb(original.rgb);
-
-    float weight = getModeWeight(hsx.x, u_int0, overlap);
-
-    if (u_int0 != MODE_MASTER && hsx.y < EPSILON) {
-        weight = 0.0;
-    }
-
-    if (weight > EPSILON) {
-        float h = fract(hsx.x + hueShift * weight);
-        float s = clamp(adjustSaturation(hsx.y, satAmount * weight), 0.0, 1.0);
-        float v = (u_int1 == COLORSPACE_HSL)
-            ? clamp(adjustLightness(hsx.z, lightAmount * weight), 0.0, 1.0)
-            : clamp(adjustBrightness(hsx.z, lightAmount * weight), 0.0, 1.0);
-
-        vec3 adjusted = vec3(h, s, v);
-        result = (u_int1 == COLORSPACE_HSL)
-            ? hsl2rgb(adjusted)
-            : hsb2rgb(adjusted);
-    } else {
-        result = original.rgb;
-    }
-
-    fragColor = vec4(result, original.a);
-}

blueprints/.glsl/Image_Blur_1.frag

@@ -1,111 +0,0 @@
-#version 300 es
-#pragma passes 2
-precision highp float;
-
-// Blur type constants
-const int BLUR_GAUSSIAN = 0;
-const int BLUR_BOX = 1;
-const int BLUR_RADIAL = 2;
-
-// Radial blur config
-const int RADIAL_SAMPLES = 12;
-const float RADIAL_STRENGTH = 0.0003;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform int u_int0;      // Blur type (BLUR_GAUSSIAN, BLUR_BOX, BLUR_RADIAL)
-uniform float u_float0;  // Blur radius/amount
-uniform int u_pass;      // Pass index (0 = horizontal, 1 = vertical)
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-float gaussian(float x, float sigma) {
-    return exp(-(x * x) / (2.0 * sigma * sigma));
-}
-
-void main() {
-    vec2 texelSize = 1.0 / u_resolution;
-    float radius = max(u_float0, 0.0);
-
-    // Radial (angular) blur - single pass, doesn't use separable
-    if (u_int0 == BLUR_RADIAL) {
-        // Only execute on first pass
-        if (u_pass > 0) {
-            fragColor0 = texture(u_image0, v_texCoord);
-            return;
-        }
-
-        vec2 center = vec2(0.5);
-        vec2 dir = v_texCoord - center;
-        float dist = length(dir);
-
-        if (dist < 1e-4) {
-            fragColor0 = texture(u_image0, v_texCoord);
-            return;
-        }
-
-        vec4 sum = vec4(0.0);
-        float totalWeight = 0.0;
-        float angleStep = radius * RADIAL_STRENGTH;
-
-        dir /= dist;
-
-        float cosStep = cos(angleStep);
-        float sinStep = sin(angleStep);
-
-        float negAngle = -float(RADIAL_SAMPLES) * angleStep;
-        vec2 rotDir = vec2(
-            dir.x * cos(negAngle) - dir.y * sin(negAngle),
-            dir.x * sin(negAngle) + dir.y * cos(negAngle)
-        );
-
-        for (int i = -RADIAL_SAMPLES; i <= RADIAL_SAMPLES; i++) {
-            vec2 uv = center + rotDir * dist;
-            float w = 1.0 - abs(float(i)) / float(RADIAL_SAMPLES);
-            sum += texture(u_image0, uv) * w;
-            totalWeight += w;
-
-            rotDir = vec2(
-                rotDir.x * cosStep - rotDir.y * sinStep,
-                rotDir.x * sinStep + rotDir.y * cosStep
-            );
-        }
-
-        fragColor0 = sum / max(totalWeight, 0.001);
-        return;
-    }
-
-    // Separable Gaussian / Box blur
-    int samples = int(ceil(radius));
-
-    if (samples == 0) {
-        fragColor0 = texture(u_image0, v_texCoord);
-        return;
-    }
-
-    // Direction: pass 0 = horizontal, pass 1 = vertical
-    vec2 dir = (u_pass == 0) ? vec2(1.0, 0.0) : vec2(0.0, 1.0);
-
-    vec4 color = vec4(0.0);
-    float totalWeight = 0.0;
-    float sigma = radius / 2.0;
-
-    for (int i = -samples; i <= samples; i++) {
-        vec2 offset = dir * float(i) * texelSize;
-        vec4 sample_color = texture(u_image0, v_texCoord + offset);
-
-        float weight;
-        if (u_int0 == BLUR_GAUSSIAN) {
-            weight = gaussian(float(i), sigma);
-        } else {
-            // BLUR_BOX
-            weight = 1.0;
-        }
-
-        color += sample_color * weight;
-        totalWeight += weight;
-    }
-
-    fragColor0 = color / totalWeight;
-}

blueprints/.glsl/Image_Channels_23.frag

@@ -1,19 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-layout(location = 1) out vec4 fragColor1;
-layout(location = 2) out vec4 fragColor2;
-layout(location = 3) out vec4 fragColor3;
-
-void main() {
-  vec4 color = texture(u_image0, v_texCoord);
-  // Output each channel as grayscale to separate render targets
-  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel
-  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel
-  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel
-  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel
-}

blueprints/.glsl/Image_Levels_1.frag

@@ -1,71 +0,0 @@
-#version 300 es
-precision highp float;
-
-// Levels Adjustment
-// u_int0:   channel      (0=RGB, 1=R, 2=G, 3=B)         default: 0
-// u_float0: input black  (0-255)                        default: 0
-// u_float1: input white  (0-255)                        default: 255
-// u_float2: gamma        (0.01-9.99)                    default: 1.0
-// u_float3: output black (0-255)                        default: 0
-// u_float4: output white (0-255)                        default: 255
-
-uniform sampler2D u_image0;
-uniform int u_int0;
-uniform float u_float0;
-uniform float u_float1;
-uniform float u_float2;
-uniform float u_float3;
-uniform float u_float4;
-
-in vec2 v_texCoord;
-out vec4 fragColor;
-
-vec3 applyLevels(vec3 color, float inBlack, float inWhite, float gamma, float outBlack, float outWhite) {
-    float inRange = max(inWhite - inBlack, 0.0001);
-    vec3 result = clamp((color - inBlack) / inRange, 0.0, 1.0);
-    result = pow(result, vec3(1.0 / gamma));
-    result = mix(vec3(outBlack), vec3(outWhite), result);
-    return result;
-}
-
-float applySingleChannel(float value, float inBlack, float inWhite, float gamma, float outBlack, float outWhite) {
-    float inRange = max(inWhite - inBlack, 0.0001);
-    float result = clamp((value - inBlack) / inRange, 0.0, 1.0);
-    result = pow(result, 1.0 / gamma);
-    result = mix(outBlack, outWhite, result);
-    return result;
-}
-
-void main() {
-    vec4 texColor = texture(u_image0, v_texCoord);
-    vec3 color = texColor.rgb;
-    
-    float inBlack = u_float0 / 255.0;
-    float inWhite = u_float1 / 255.0;
-    float gamma = u_float2;
-    float outBlack = u_float3 / 255.0;
-    float outWhite = u_float4 / 255.0;
-    
-    vec3 result;
-    
-    if (u_int0 == 0) {
-        result = applyLevels(color, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 1) {
-        result = color;
-        result.r = applySingleChannel(color.r, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 2) {
-        result = color;
-        result.g = applySingleChannel(color.g, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else if (u_int0 == 3) {
-        result = color;
-        result.b = applySingleChannel(color.b, inBlack, inWhite, gamma, outBlack, outWhite);
-    }
-    else {
-        result = color;
-    }
-    
-    fragColor = vec4(result, texColor.a);
-}

blueprints/.glsl/README.md

@@ -1,28 +0,0 @@
-# GLSL Shader Sources
-
-This folder contains the GLSL fragment shaders extracted from blueprint JSON files for easier editing and version control.
-
-## File Naming Convention
-
-`{Blueprint_Name}_{node_id}.frag`
-
-- **Blueprint_Name**: The JSON filename with spaces/special chars replaced by underscores
-- **node_id**: The GLSLShader node ID within the subgraph
-
-## Usage
-
-```bash
-# Extract shaders from blueprint JSONs to this folder
-python update_blueprints.py extract
-
-# Patch edited shaders back into blueprint JSONs
-python update_blueprints.py patch
-```
-
-## Workflow
-
-1. Run `extract` to pull current shaders from JSONs
-2. Edit `.frag` files
-3. Run `patch` to update the blueprint JSONs
-4. Test
-5. Commit both `.frag` files and updated JSONs

blueprints/.glsl/Sharpen_23.frag

@@ -1,28 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0;  // strength [0.0 – 2.0] typical: 0.3–1.0
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-void main() {
-    vec2 texel = 1.0 / u_resolution;
-    
-    // Sample center and neighbors
-    vec4 center = texture(u_image0, v_texCoord);
-    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));
-    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));
-    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));
-    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));
-    
-    // Edge enhancement (Laplacian)
-    vec4 edges = center * 4.0 - top - bottom - left - right;
-    
-    // Add edges back scaled by strength
-    vec4 sharpened = center + edges * u_float0;
-    
-    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);
-}

blueprints/.glsl/Unsharp_Mask_26.frag

@@ -1,61 +0,0 @@
-#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-uniform float u_float0;  // amount    [0.0 - 3.0]  typical: 0.5-1.5
-uniform float u_float1;  // radius    [0.5 - 10.0] blur radius in pixels
-uniform float u_float2;  // threshold [0.0 - 0.1]  min difference to sharpen
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-float gaussian(float x, float sigma) {
-    return exp(-(x * x) / (2.0 * sigma * sigma));
-}
-
-float getLuminance(vec3 color) {
-    return dot(color, vec3(0.2126, 0.7152, 0.0722));
-}
-
-void main() {
-    vec2 texel = 1.0 / u_resolution;
-    float radius = max(u_float1, 0.5);
-    float amount = u_float0;
-    float threshold = u_float2;
-
-    vec4 original = texture(u_image0, v_texCoord);
-
-    // Gaussian blur for the "unsharp" mask
-    int samples = int(ceil(radius));
-    float sigma = radius / 2.0;
-
-    vec4 blurred = vec4(0.0);
-    float totalWeight = 0.0;
-
-    for (int x = -samples; x <= samples; x++) {
-        for (int y = -samples; y <= samples; y++) {
-            vec2 offset = vec2(float(x), float(y)) * texel;
-            vec4 sample_color = texture(u_image0, v_texCoord + offset);
-
-            float dist = length(vec2(float(x), float(y)));
-            float weight = gaussian(dist, sigma);
-            blurred += sample_color * weight;
-            totalWeight += weight;
-        }
-    }
-    blurred /= totalWeight;
-
-    // Unsharp mask = original - blurred
-    vec3 mask = original.rgb - blurred.rgb;
-
-    // Luminance-based threshold with smooth falloff
-    float lumaDelta = abs(getLuminance(original.rgb) - getLuminance(blurred.rgb));
-    float thresholdScale = smoothstep(0.0, threshold, lumaDelta);
-    mask *= thresholdScale;
-
-    // Sharpen: original + mask * amount
-    vec3 sharpened = original.rgb + mask * amount;
-
-    fragColor0 = vec4(clamp(sharpened, 0.0, 1.0), original.a);
-}

blueprints/.glsl/update_blueprints.py

@@ -1,159 +0,0 @@
-#!/usr/bin/env python3
-"""
-Shader Blueprint Updater
-
-Syncs GLSL shader files between this folder and blueprint JSON files.
-
-File naming convention:
-    {Blueprint Name}_{node_id}.frag
-
-Usage:
-    python update_blueprints.py extract   # Extract shaders from JSONs to here
-    python update_blueprints.py patch     # Patch shaders back into JSONs
-    python update_blueprints.py           # Same as patch (default)
-"""
-
-import json
-import logging
-import sys
-import re
-from pathlib import Path
-
-logging.basicConfig(level=logging.INFO, format='%(message)s')
-logger = logging.getLogger(__name__)
-
-GLSL_DIR = Path(__file__).parent
-BLUEPRINTS_DIR = GLSL_DIR.parent
-
-
-def get_blueprint_files():
-    """Get all blueprint JSON files."""
-    return sorted(BLUEPRINTS_DIR.glob("*.json"))
-
-
-def sanitize_filename(name):
-    """Convert blueprint name to safe filename."""
-    return re.sub(r'[^\w\-]', '_', name)
-
-
-def extract_shaders():
-    """Extract all shaders from blueprint JSONs to this folder."""
-    extracted = 0
-    for json_path in get_blueprint_files():
-        blueprint_name = json_path.stem
-
-        try:
-            with open(json_path, 'r') as f:
-                data = json.load(f)
-        except (json.JSONDecodeError, IOError) as e:
-            logger.warning("Skipping %s: %s", json_path.name, e)
-            continue
-
-        # Find GLSLShader nodes in subgraphs
-        for subgraph in data.get('definitions', {}).get('subgraphs', []):
-            for node in subgraph.get('nodes', []):
-                if node.get('type') == 'GLSLShader':
-                    node_id = node.get('id')
-                    widgets = node.get('widgets_values', [])
-
-                    # Find shader code (first string that looks like GLSL)
-                    for widget in widgets:
-                        if isinstance(widget, str) and widget.startswith('#version'):
-                            safe_name = sanitize_filename(blueprint_name)
-                            frag_name = f"{safe_name}_{node_id}.frag"
-                            frag_path = GLSL_DIR / frag_name
-
-                            with open(frag_path, 'w') as f:
-                                f.write(widget)
-
-                            logger.info("  Extracted: %s", frag_name)
-                            extracted += 1
-                            break
-
-    logger.info("\nExtracted %d shader(s)", extracted)
-
-
-def patch_shaders():
-    """Patch shaders from this folder back into blueprint JSONs."""
-    # Build lookup: blueprint_name -> [(node_id, shader_code), ...]
-    shader_updates = {}
-
-    for frag_path in sorted(GLSL_DIR.glob("*.frag")):
-        # Parse filename: {blueprint_name}_{node_id}.frag
-        parts = frag_path.stem.rsplit('_', 1)
-        if len(parts) != 2:
-            logger.warning("Skipping %s: invalid filename format", frag_path.name)
-            continue
-
-        blueprint_name, node_id_str = parts
-
-        try:
-            node_id = int(node_id_str)
-        except ValueError:
-            logger.warning("Skipping %s: invalid node_id", frag_path.name)
-            continue
-
-        with open(frag_path, 'r') as f:
-            shader_code = f.read()
-
-        if blueprint_name not in shader_updates:
-            shader_updates[blueprint_name] = []
-        shader_updates[blueprint_name].append((node_id, shader_code))
-
-    # Apply updates to JSON files
-    patched = 0
-    for json_path in get_blueprint_files():
-        blueprint_name = sanitize_filename(json_path.stem)
-
-        if blueprint_name not in shader_updates:
-            continue
-
-        try:
-            with open(json_path, 'r') as f:
-                data = json.load(f)
-        except (json.JSONDecodeError, IOError) as e:
-            logger.error("Error reading %s: %s", json_path.name, e)
-            continue
-
-        modified = False
-        for node_id, shader_code in shader_updates[blueprint_name]:
-            # Find the node and update
-            for subgraph in data.get('definitions', {}).get('subgraphs', []):
-                for node in subgraph.get('nodes', []):
-                    if node.get('id') == node_id and node.get('type') == 'GLSLShader':
-                        widgets = node.get('widgets_values', [])
-                        if len(widgets) > 0 and widgets[0] != shader_code:
-                            widgets[0] = shader_code
-                            modified = True
-                            logger.info("  Patched: %s (node %d)", json_path.name, node_id)
-                            patched += 1
-
-        if modified:
-            with open(json_path, 'w') as f:
-                json.dump(data, f)
-
-    if patched == 0:
-        logger.info("No changes to apply.")
-    else:
-        logger.info("\nPatched %d shader(s)", patched)
-
-
-def main():
-    if len(sys.argv) < 2:
-        command = "patch"
-    else:
-        command = sys.argv[1].lower()
-
-    if command == "extract":
-        logger.info("Extracting shaders from blueprints...")
-        extract_shaders()
-    elif command in ("patch", "update", "apply"):
-        logger.info("Patching shaders into blueprints...")
-        patch_shaders()
-    else:
-        logger.info(__doc__)
-        sys.exit(1)
-
-
-if __name__ == "__main__":
-    main()

blueprints/Image Channels.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 29, "last_link_id": 0, "nodes": [{"id": 29, "type": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "pos": [1970, -230], "size": [180, 86], "flags": {}, "order": 5, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": []}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": []}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": []}], "title": "Image Channels", "properties": {"proxyWidgets": []}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "4c9d6ea4-b912-40e5-8766-6793a9758c53", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 28, "lastLinkId": 39, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Image Channels", "inputNode": {"id": -10, "bounding": [1820, -185, 120, 60]}, "outputNode": {"id": -20, "bounding": [2460, -215, 120, 120]}, "inputs": [{"id": "3522932b-2d86-4a1f-a02a-cb29f3a9d7fe", "name": "images.image0", "type": "IMAGE", "linkIds": [39], "localized_name": "images.image0", "label": "image", "pos": [1920, -165]}], "outputs": [{"id": "605cb9c3-b065-4d9b-81d2-3ec331889b2b", "name": "IMAGE0", "type": "IMAGE", "linkIds": [26], "localized_name": "IMAGE0", "label": "R", "pos": [2480, -195]}, {"id": "fb44a77e-0522-43e9-9527-82e7465b3596", "name": "IMAGE1", "type": "IMAGE", "linkIds": [27], "localized_name": "IMAGE1", "label": "G", "pos": [2480, -175]}, {"id": "81460ee6-0131-402a-874f-6bf3001fc4ff", "name": "IMAGE2", "type": "IMAGE", "linkIds": [28], "localized_name": "IMAGE2", "label": "B", "pos": [2480, -155]}, {"id": "ae690246-80d4-4951-b1d9-9306d8a77417", "name": "IMAGE3", "type": "IMAGE", "linkIds": [29], "localized_name": "IMAGE3", "label": "A", "pos": [2480, -135]}], "widgets": [], "nodes": [{"id": 23, "type": "GLSLShader", "pos": [2000, -330], "size": [400, 172], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 39}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}], "outputs": [{"label": "R", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [26]}, {"label": "G", "localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": [27]}, {"label": "B", "localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": [28]}, {"label": "A", "localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": [29]}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\nlayout(location = 1) out vec4 fragColor1;\nlayout(location = 2) out vec4 fragColor2;\nlayout(location = 3) out vec4 fragColor3;\n\nvoid main() {\n  vec4 color = texture(u_image0, v_texCoord);\n  // Output each channel as grayscale to separate render targets\n  fragColor0 = vec4(vec3(color.r), 1.0);  // Red channel\n  fragColor1 = vec4(vec3(color.g), 1.0);  // Green channel\n  fragColor2 = vec4(vec3(color.b), 1.0);  // Blue channel\n  fragColor3 = vec4(vec3(color.a), 1.0);  // Alpha channel\n}\n", "from_input"]}], "groups": [], "links": [{"id": 39, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 26, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}, {"id": 27, "origin_id": 23, "origin_slot": 1, "target_id": -20, "target_slot": 1, "type": "IMAGE"}, {"id": 28, "origin_id": 23, "origin_slot": 2, "target_id": -20, "target_slot": 2, "type": "IMAGE"}, {"id": 29, "origin_id": 23, "origin_slot": 3, "target_id": -20, "target_slot": 3, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Color adjust"}]}}

blueprints/Prompt Enhance.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 15, "last_link_id": 0, "nodes": [{"id": 15, "type": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "pos": [-1490, 2040], "size": [400, 260], "flags": {}, "order": 0, "mode": 0, "inputs": [{"name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": null}, {"label": "reference images", "name": "images", "type": "IMAGE", "link": null}], "outputs": [{"name": "STRING", "type": "STRING", "links": null}], "title": "Prompt Enhance", "properties": {"proxyWidgets": [["-1", "prompt"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": [""]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "24d8bbfd-39d4-4774-bff0-3de40cc7a471", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 15, "lastLinkId": 14, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Prompt Enhance", "inputNode": {"id": -10, "bounding": [-2170, 2110, 138.876953125, 80]}, "outputNode": {"id": -20, "bounding": [-640, 2110, 120, 60]}, "inputs": [{"id": "aeab7216-00e0-4528-a09b-bba50845c5a6", "name": "prompt", "type": "STRING", "linkIds": [11], "pos": [-2051.123046875, 2130]}, {"id": "7b73fd36-aa31-4771-9066-f6c83879994b", "name": "images", "type": "IMAGE", "linkIds": [14], "label": "reference images", "pos": [-2051.123046875, 2150]}], "outputs": [{"id": "c7b0d930-68a1-48d1-b496-0519e5837064", "name": "STRING", "type": "STRING", "linkIds": [13], "pos": [-620, 2130]}], "widgets": [], "nodes": [{"id": 11, "type": "GeminiNode", "pos": [-1560, 1990], "size": [470, 470], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "shape": 7, "type": "IMAGE", "link": 14}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": null}, {"localized_name": "video", "name": "video", "shape": 7, "type": "VIDEO", "link": null}, {"localized_name": "files", "name": "files", "shape": 7, "type": "GEMINI_INPUT_FILES", "link": null}, {"localized_name": "prompt", "name": "prompt", "type": "STRING", "widget": {"name": "prompt"}, "link": 11}, {"localized_name": "model", "name": "model", "type": "COMBO", "widget": {"name": "model"}, "link": null}, {"localized_name": "seed", "name": "seed", "type": "INT", "widget": {"name": "seed"}, "link": null}, {"localized_name": "system_prompt", "name": "system_prompt", "shape": 7, "type": "STRING", "widget": {"name": "system_prompt"}, "link": null}], "outputs": [{"localized_name": "STRING", "name": "STRING", "type": "STRING", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.14.1", "Node name for S&R": "GeminiNode"}, "widgets_values": ["", "gemini-3-pro-preview", 42, "randomize", "You are an expert in prompt writing.\nBased on the input, rewrite the user's input into a detailed prompt.\nincluding camera settings, lighting, composition, and style.\nReturn the prompt only"], "color": "#432", "bgcolor": "#653"}], "groups": [], "links": [{"id": 11, "origin_id": -10, "origin_slot": 0, "target_id": 11, "target_slot": 4, "type": "STRING"}, {"id": 13, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "STRING"}, {"id": 14, "origin_id": -10, "origin_slot": 1, "target_id": 11, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Text generation/Prompt enhance"}]}, "extra": {}}

blueprints/Sharpen.json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 25, "last_link_id": 0, "nodes": [{"id": 25, "type": "621ba4e2-22a8-482d-a369-023753198b7b", "pos": [4610, -790], "size": [230, 58], "flags": {}, "order": 4, "mode": 0, "inputs": [{"label": "image", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": null}], "outputs": [{"label": "IMAGE", "localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": []}], "title": "Sharpen", "properties": {"proxyWidgets": [["24", "value"]]}, "widgets_values": []}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "621ba4e2-22a8-482d-a369-023753198b7b", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 24, "lastLinkId": 36, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Sharpen", "inputNode": {"id": -10, "bounding": [4090, -825, 120, 60]}, "outputNode": {"id": -20, "bounding": [5150, -825, 120, 60]}, "inputs": [{"id": "37011fb7-14b7-4e0e-b1a0-6a02e8da1fd7", "name": "images.image0", "type": "IMAGE", "linkIds": [34], "localized_name": "images.image0", "label": "image", "pos": [4190, -805]}], "outputs": [{"id": "e9182b3f-635c-4cd4-a152-4b4be17ae4b9", "name": "IMAGE0", "type": "IMAGE", "linkIds": [35], "localized_name": "IMAGE0", "label": "IMAGE", "pos": [5170, -805]}], "widgets": [], "nodes": [{"id": 24, "type": "PrimitiveFloat", "pos": [4280, -1240], "size": [270, 58], "flags": {}, "order": 0, "mode": 0, "inputs": [{"label": "strength", "localized_name": "value", "name": "value", "type": "FLOAT", "widget": {"name": "value"}, "link": null}], "outputs": [{"localized_name": "FLOAT", "name": "FLOAT", "type": "FLOAT", "links": [36]}], "properties": {"Node name for S&R": "PrimitiveFloat", "min": 0, "max": 3, "precision": 2, "step": 0.05}, "widgets_values": [0.5]}, {"id": 23, "type": "GLSLShader", "pos": [4570, -1240], "size": [370, 192], "flags": {}, "order": 1, "mode": 0, "inputs": [{"label": "image0", "localized_name": "images.image0", "name": "images.image0", "type": "IMAGE", "link": 34}, {"label": "image1", "localized_name": "images.image1", "name": "images.image1", "shape": 7, "type": "IMAGE", "link": null}, {"label": "u_float0", "localized_name": "floats.u_float0", "name": "floats.u_float0", "shape": 7, "type": "FLOAT", "link": 36}, {"label": "u_float1", "localized_name": "floats.u_float1", "name": "floats.u_float1", "shape": 7, "type": "FLOAT", "link": null}, {"label": "u_int0", "localized_name": "ints.u_int0", "name": "ints.u_int0", "shape": 7, "type": "INT", "link": null}, {"localized_name": "fragment_shader", "name": "fragment_shader", "type": "STRING", "widget": {"name": "fragment_shader"}, "link": null}, {"localized_name": "size_mode", "name": "size_mode", "type": "COMFY_DYNAMICCOMBO_V3", "widget": {"name": "size_mode"}, "link": null}], "outputs": [{"localized_name": "IMAGE0", "name": "IMAGE0", "type": "IMAGE", "links": [35]}, {"localized_name": "IMAGE1", "name": "IMAGE1", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE2", "name": "IMAGE2", "type": "IMAGE", "links": null}, {"localized_name": "IMAGE3", "name": "IMAGE3", "type": "IMAGE", "links": null}], "properties": {"Node name for S&R": "GLSLShader"}, "widgets_values": ["#version 300 es\nprecision highp float;\n\nuniform sampler2D u_image0;\nuniform vec2 u_resolution;\nuniform float u_float0;  // strength [0.0 – 2.0] typical: 0.3–1.0\n\nin vec2 v_texCoord;\nlayout(location = 0) out vec4 fragColor0;\n\nvoid main() {\n    vec2 texel = 1.0 / u_resolution;\n    \n    // Sample center and neighbors\n    vec4 center = texture(u_image0, v_texCoord);\n    vec4 top    = texture(u_image0, v_texCoord + vec2( 0.0, -texel.y));\n    vec4 bottom = texture(u_image0, v_texCoord + vec2( 0.0,  texel.y));\n    vec4 left   = texture(u_image0, v_texCoord + vec2(-texel.x,  0.0));\n    vec4 right  = texture(u_image0, v_texCoord + vec2( texel.x,  0.0));\n    \n    // Edge enhancement (Laplacian)\n    vec4 edges = center * 4.0 - top - bottom - left - right;\n    \n    // Add edges back scaled by strength\n    vec4 sharpened = center + edges * u_float0;\n    \n    fragColor0 = vec4(clamp(sharpened.rgb, 0.0, 1.0), center.a);\n}", "from_input"]}], "groups": [], "links": [{"id": 36, "origin_id": 24, "origin_slot": 0, "target_id": 23, "target_slot": 2, "type": "FLOAT"}, {"id": 34, "origin_id": -10, "origin_slot": 0, "target_id": 23, "target_slot": 0, "type": "IMAGE"}, {"id": 35, "origin_id": 23, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "IMAGE"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Image Tools/Sharpen"}]}}

blueprints/Video Upscale(GAN x4).json

@@ -1 +0,0 @@
-{"revision": 0, "last_node_id": 13, "last_link_id": 0, "nodes": [{"id": 13, "type": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "pos": [1120, 330], "size": [240, 58], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": null}, {"name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": null}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": []}], "title": "Video Upscale(GAN x4)", "properties": {"proxyWidgets": [["-1", "model_name"]], "cnr_id": "comfy-core", "ver": "0.14.1"}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "links": [], "version": 0.4, "definitions": {"subgraphs": [{"id": "cf95b747-3e17-46cb-8097-cac60ff9b2e1", "version": 1, "state": {"lastGroupId": 0, "lastNodeId": 13, "lastLinkId": 19, "lastRerouteId": 0}, "revision": 0, "config": {}, "name": "Video Upscale(GAN x4)", "inputNode": {"id": -10, "bounding": [550, 460, 120, 80]}, "outputNode": {"id": -20, "bounding": [1490, 460, 120, 60]}, "inputs": [{"id": "666d633e-93e7-42dc-8d11-2b7b99b0f2a6", "name": "video", "type": "VIDEO", "linkIds": [10], "localized_name": "video", "pos": [650, 480]}, {"id": "2e23a087-caa8-4d65-99e6-662761aa905a", "name": "model_name", "type": "COMBO", "linkIds": [19], "pos": [650, 500]}], "outputs": [{"id": "0c1768ea-3ec2-412f-9af6-8e0fa36dae70", "name": "VIDEO", "type": "VIDEO", "linkIds": [15], "localized_name": "VIDEO", "pos": [1510, 480]}], "widgets": [], "nodes": [{"id": 2, "type": "ImageUpscaleWithModel", "pos": [1110, 450], "size": [320, 46], "flags": {}, "order": 1, "mode": 0, "inputs": [{"localized_name": "upscale_model", "name": "upscale_model", "type": "UPSCALE_MODEL", "link": 1}, {"localized_name": "image", "name": "image", "type": "IMAGE", "link": 14}], "outputs": [{"localized_name": "IMAGE", "name": "IMAGE", "type": "IMAGE", "links": [13]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "ImageUpscaleWithModel"}}, {"id": 11, "type": "CreateVideo", "pos": [1110, 550], "size": [320, 78], "flags": {}, "order": 3, "mode": 0, "inputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "link": 13}, {"localized_name": "audio", "name": "audio", "shape": 7, "type": "AUDIO", "link": 16}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "widget": {"name": "fps"}, "link": 12}], "outputs": [{"localized_name": "VIDEO", "name": "VIDEO", "type": "VIDEO", "links": [15]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "CreateVideo"}, "widgets_values": [30]}, {"id": 10, "type": "GetVideoComponents", "pos": [1110, 330], "size": [320, 70], "flags": {}, "order": 2, "mode": 0, "inputs": [{"localized_name": "video", "name": "video", "type": "VIDEO", "link": 10}], "outputs": [{"localized_name": "images", "name": "images", "type": "IMAGE", "links": [14]}, {"localized_name": "audio", "name": "audio", "type": "AUDIO", "links": [16]}, {"localized_name": "fps", "name": "fps", "type": "FLOAT", "links": [12]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "GetVideoComponents"}}, {"id": 1, "type": "UpscaleModelLoader", "pos": [750, 450], "size": [280, 60], "flags": {}, "order": 0, "mode": 0, "inputs": [{"localized_name": "model_name", "name": "model_name", "type": "COMBO", "widget": {"name": "model_name"}, "link": 19}], "outputs": [{"localized_name": "UPSCALE_MODEL", "name": "UPSCALE_MODEL", "type": "UPSCALE_MODEL", "links": [1]}], "properties": {"cnr_id": "comfy-core", "ver": "0.10.0", "Node name for S&R": "UpscaleModelLoader", "models": [{"name": "RealESRGAN_x4plus.safetensors", "url": "https://huggingface.co/Comfy-Org/Real-ESRGAN_repackaged/resolve/main/RealESRGAN_x4plus.safetensors", "directory": "upscale_models"}]}, "widgets_values": ["RealESRGAN_x4plus.safetensors"]}], "groups": [], "links": [{"id": 1, "origin_id": 1, "origin_slot": 0, "target_id": 2, "target_slot": 0, "type": "UPSCALE_MODEL"}, {"id": 14, "origin_id": 10, "origin_slot": 0, "target_id": 2, "target_slot": 1, "type": "IMAGE"}, {"id": 13, "origin_id": 2, "origin_slot": 0, "target_id": 11, "target_slot": 0, "type": "IMAGE"}, {"id": 16, "origin_id": 10, "origin_slot": 1, "target_id": 11, "target_slot": 1, "type": "AUDIO"}, {"id": 12, "origin_id": 10, "origin_slot": 2, "target_id": 11, "target_slot": 2, "type": "FLOAT"}, {"id": 10, "origin_id": -10, "origin_slot": 0, "target_id": 10, "target_slot": 0, "type": "VIDEO"}, {"id": 15, "origin_id": 11, "origin_slot": 0, "target_id": -20, "target_slot": 0, "type": "VIDEO"}, {"id": 19, "origin_id": -10, "origin_slot": 1, "target_id": 1, "target_slot": 0, "type": "COMBO"}], "extra": {"workflowRendererVersion": "LG"}, "category": "Video generation and editing/Enhance video"}]}, "extra": {}}

comfy/comfy_types/node_typing.py

@@ -176,8 +176,6 @@ class InputTypeOptions(TypedDict):
     """COMBO type only. Specifies the configuration for a multi-select widget.
     Available after ComfyUI frontend v1.13.4
     https://github.com/Comfy-Org/ComfyUI_frontend/pull/2987"""
-    gradient_stops: NotRequired[list[list[float]]]
-    """Gradient color stops for gradientslider display mode. Each stop is [offset, r, g, b] (``FLOAT``)."""
 
 
 class HiddenInputTypeDict(TypedDict):

comfy/ldm/lightricks/av_model.py

@@ -9,7 +9,6 @@ from comfy.ldm.lightricks.model import (
     LTXVModel,
 )
 from comfy.ldm.lightricks.symmetric_patchifier import AudioPatchifier
-from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector
 import comfy.ldm.common_dit
 
 class CompressedTimestep:
@@ -451,29 +450,6 @@ class LTXAVModel(LTXVModel):
             operations=self.operations,
         )
 
-        self.audio_embeddings_connector = Embeddings1DConnector(
-            split_rope=True,
-            double_precision_rope=True,
-            dtype=dtype,
-            device=device,
-            operations=self.operations,
-        )
-
-        self.video_embeddings_connector = Embeddings1DConnector(
-            split_rope=True,
-            double_precision_rope=True,
-            dtype=dtype,
-            device=device,
-            operations=self.operations,
-        )
-
-    def preprocess_text_embeds(self, context):
-        if context.shape[-1] == self.caption_channels * 2:
-            return context
-        out_vid = self.video_embeddings_connector(context)[0]
-        out_audio = self.audio_embeddings_connector(context)[0]
-        return torch.concat((out_vid, out_audio), dim=-1)
-
     def _init_transformer_blocks(self, device, dtype, **kwargs):
         """Initialize transformer blocks for LTXAV."""
         self.transformer_blocks = nn.ModuleList(

comfy/ldm/lightricks/embeddings_connector.py

@@ -157,9 +157,11 @@ class Embeddings1DConnector(nn.Module):
         self.num_learnable_registers = num_learnable_registers
         if self.num_learnable_registers:
             self.learnable_registers = nn.Parameter(
-                torch.empty(
+                torch.rand(
                     self.num_learnable_registers, inner_dim, dtype=dtype, device=device
                 )
+                * 2.0
+                - 1.0
             )
 
     def get_fractional_positions(self, indices_grid):
@@ -232,7 +234,7 @@ class Embeddings1DConnector(nn.Module):
 
         return indices
 
-    def precompute_freqs_cis(self, indices_grid, spacing="exp", out_dtype=None):
+    def precompute_freqs_cis(self, indices_grid, spacing="exp"):
         dim = self.inner_dim
         n_elem = 2  # 2 because of cos and sin
         freqs = self.precompute_freqs(indices_grid, spacing)
@@ -245,7 +247,7 @@ class Embeddings1DConnector(nn.Module):
             )
         else:
             cos_freq, sin_freq = interleaved_freqs_cis(freqs, dim % n_elem)
-        return cos_freq.to(dtype=out_dtype), sin_freq.to(dtype=out_dtype), self.split_rope
+        return cos_freq.to(self.dtype), sin_freq.to(self.dtype), self.split_rope
 
     def forward(
         self,
@@ -286,7 +288,7 @@ class Embeddings1DConnector(nn.Module):
             hidden_states.shape[1], dtype=torch.float32, device=hidden_states.device
         )
         indices_grid = indices_grid[None, None, :]
-        freqs_cis = self.precompute_freqs_cis(indices_grid, out_dtype=hidden_states.dtype)
+        freqs_cis = self.precompute_freqs_cis(indices_grid)
 
         # 2. Blocks
         for block_idx, block in enumerate(self.transformer_1d_blocks):

comfy/memory_management.py

@@ -78,4 +78,4 @@ def interpret_gathered_like(tensors, gathered):
 
     return dest_views
 
-aimdo_enabled = False
+aimdo_allocator = None

comfy/model_base.py

@@ -988,14 +988,10 @@ class LTXAV(BaseModel):
     def extra_conds(self, **kwargs):
         out = super().extra_conds(**kwargs)
         attention_mask = kwargs.get("attention_mask", None)
-        device = kwargs["device"]
-
         if attention_mask is not None:
             out['attention_mask'] = comfy.conds.CONDRegular(attention_mask)
         cross_attn = kwargs.get("cross_attn", None)
         if cross_attn is not None:
-            if hasattr(self.diffusion_model, "preprocess_text_embeds"):
-                cross_attn = self.diffusion_model.preprocess_text_embeds(cross_attn.to(device=device, dtype=self.get_dtype_inference()))
             out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
 
         out['frame_rate'] = comfy.conds.CONDConstant(kwargs.get("frame_rate", 25))

comfy/model_management.py

@@ -836,7 +836,7 @@ def unet_inital_load_device(parameters, dtype):
 
     mem_dev = get_free_memory(torch_dev)
     mem_cpu = get_free_memory(cpu_dev)
-    if mem_dev > mem_cpu and model_size < mem_dev and comfy.memory_management.aimdo_enabled:
+    if mem_dev > mem_cpu and model_size < mem_dev and comfy.memory_management.aimdo_allocator is None:
         return torch_dev
     else:
         return cpu_dev
@@ -1121,6 +1121,7 @@ def get_cast_buffer(offload_stream, device, size, ref):
             synchronize()
             del STREAM_CAST_BUFFERS[offload_stream]
             del cast_buffer
+            #FIXME: This doesn't work in Aimdo because mempool cant clear cache
             soft_empty_cache()
         with wf_context:
             cast_buffer = torch.empty((size), dtype=torch.int8, device=device)

comfy/model_patcher.py

@@ -271,7 +271,6 @@ class ModelPatcher:
         self.is_clip = False
         self.hook_mode = comfy.hooks.EnumHookMode.MaxSpeed
 
-        self.cached_patcher_init: tuple[Callable, tuple] | None = None
         if not hasattr(self.model, 'model_loaded_weight_memory'):
             self.model.model_loaded_weight_memory = 0
 
@@ -308,15 +307,8 @@ class ModelPatcher:
     def get_free_memory(self, device):
         return comfy.model_management.get_free_memory(device)
 
-    def clone(self, disable_dynamic=False):
-        class_ = self.__class__
-        model = self.model
-        if self.is_dynamic() and disable_dynamic:
-            class_ = ModelPatcher
-            temp_model_patcher = self.cached_patcher_init[0](*self.cached_patcher_init[1], disable_dynamic=True)
-            model = temp_model_patcher.model
-
-        n = class_(model, self.load_device, self.offload_device, self.model_size(), weight_inplace_update=self.weight_inplace_update)
+    def clone(self):
+        n = self.__class__(self.model, self.load_device, self.offload_device, self.model_size(), weight_inplace_update=self.weight_inplace_update)
         n.patches = {}
         for k in self.patches:
             n.patches[k] = self.patches[k][:]
@@ -370,8 +362,6 @@ class ModelPatcher:
         n.is_clip = self.is_clip
         n.hook_mode = self.hook_mode
 
-        n.cached_patcher_init = self.cached_patcher_init
-
         for callback in self.get_all_callbacks(CallbacksMP.ON_CLONE):
             callback(self, n)
         return n

comfy/ops.py

@@ -19,7 +19,7 @@
 import torch
 import logging
 import comfy.model_management
-from comfy.cli_args import args, PerformanceFeature
+from comfy.cli_args import args, PerformanceFeature, enables_dynamic_vram
 import comfy.float
 import json
 import comfy.memory_management
@@ -296,7 +296,7 @@ class disable_weight_init:
     class Linear(torch.nn.Linear, CastWeightBiasOp):
 
         def __init__(self, in_features, out_features, bias=True, device=None, dtype=None):
-            if not comfy.model_management.WINDOWS or not comfy.memory_management.aimdo_enabled:
+            if not comfy.model_management.WINDOWS or not enables_dynamic_vram():
                 super().__init__(in_features, out_features, bias, device, dtype)
                 return
 
@@ -317,7 +317,7 @@ class disable_weight_init:
         def _load_from_state_dict(self, state_dict, prefix, local_metadata,
                                 strict, missing_keys, unexpected_keys, error_msgs):
 
-            if not comfy.model_management.WINDOWS or not comfy.memory_management.aimdo_enabled:
+            if not comfy.model_management.WINDOWS or not enables_dynamic_vram():
                 return super()._load_from_state_dict(state_dict, prefix, local_metadata, strict,
                                                      missing_keys, unexpected_keys, error_msgs)
             assign_to_params_buffers = local_metadata.get("assign_to_params_buffers", False)
@@ -827,10 +827,6 @@ def mixed_precision_ops(quant_config={}, compute_dtype=torch.bfloat16, full_prec
                 else:
                     sd = {}
 
-                if not hasattr(self, 'weight'):
-                    logging.warning("Warning: state dict on uninitialized op {}".format(prefix))
-                    return sd
-
                 if self.bias is not None:
                     sd["{}bias".format(prefix)] = self.bias
 

comfy/sd.py

@@ -426,8 +426,10 @@ class CLIP:
     def generate(self, tokens, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.95, min_p=0.0, repetition_penalty=1.0, seed=None):
         self.cond_stage_model.reset_clip_options()
 
+        if self.layer_idx is not None:
+            self.cond_stage_model.set_clip_options({"layer": self.layer_idx})
+
         self.load_model()
-        self.cond_stage_model.set_clip_options({"layer": None})
         self.cond_stage_model.set_clip_options({"execution_device": self.patcher.load_device})
         return self.cond_stage_model.generate(tokens, do_sample=do_sample, max_length=max_length, temperature=temperature, top_k=top_k, top_p=top_p, min_p=min_p, repetition_penalty=repetition_penalty, seed=seed)
 
@@ -1530,24 +1532,14 @@ def load_checkpoint(config_path=None, ckpt_path=None, output_vae=True, output_cl
 
     return (model, clip, vae)
 
-def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}, disable_dynamic=False):
+def load_checkpoint_guess_config(ckpt_path, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}):
     sd, metadata = comfy.utils.load_torch_file(ckpt_path, return_metadata=True)
-    out = load_state_dict_guess_config(sd, output_vae, output_clip, output_clipvision, embedding_directory, output_model, model_options, te_model_options=te_model_options, metadata=metadata, disable_dynamic=disable_dynamic)
+    out = load_state_dict_guess_config(sd, output_vae, output_clip, output_clipvision, embedding_directory, output_model, model_options, te_model_options=te_model_options, metadata=metadata)
     if out is None:
         raise RuntimeError("ERROR: Could not detect model type of: {}\n{}".format(ckpt_path, model_detection_error_hint(ckpt_path, sd)))
-    if output_model:
-        out[0].cached_patcher_init = (load_checkpoint_guess_config_model_only, (ckpt_path, embedding_directory, model_options, te_model_options))
     return out
 
-def load_checkpoint_guess_config_model_only(ckpt_path, embedding_directory=None, model_options={}, te_model_options={}, disable_dynamic=False):
-    model, *_ = load_checkpoint_guess_config(ckpt_path, False, False, False,
-            embedding_directory=embedding_directory,
-            model_options=model_options,
-            te_model_options=te_model_options,
-            disable_dynamic=disable_dynamic)
-    return model
-
-def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}, metadata=None, disable_dynamic=False):
+def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_clipvision=False, embedding_directory=None, output_model=True, model_options={}, te_model_options={}, metadata=None):
     clip = None
     clipvision = None
     vae = None
@@ -1596,8 +1588,7 @@ def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_c
     if output_model:
         inital_load_device = model_management.unet_inital_load_device(parameters, unet_dtype)
         model = model_config.get_model(sd, diffusion_model_prefix, device=inital_load_device)
-        ModelPatcher = comfy.model_patcher.ModelPatcher if disable_dynamic else comfy.model_patcher.CoreModelPatcher
-        model_patcher = ModelPatcher(model, load_device=load_device, offload_device=model_management.unet_offload_device())
+        model_patcher = comfy.model_patcher.CoreModelPatcher(model, load_device=load_device, offload_device=model_management.unet_offload_device())
         model.load_model_weights(sd, diffusion_model_prefix, assign=model_patcher.is_dynamic())
 
     if output_vae:
@@ -1648,7 +1639,7 @@ def load_state_dict_guess_config(sd, output_vae=True, output_clip=True, output_c
     return (model_patcher, clip, vae, clipvision)
 
 
-def load_diffusion_model_state_dict(sd, model_options={}, metadata=None, disable_dynamic=False):
+def load_diffusion_model_state_dict(sd, model_options={}, metadata=None):
     """
     Loads a UNet diffusion model from a state dictionary, supporting both diffusers and regular formats.
 
@@ -1732,8 +1723,7 @@ def load_diffusion_model_state_dict(sd, model_options={}, metadata=None, disable
         model_config.optimizations["fp8"] = True
 
     model = model_config.get_model(new_sd, "")
-    ModelPatcher = comfy.model_patcher.ModelPatcher if disable_dynamic else comfy.model_patcher.CoreModelPatcher
-    model_patcher = ModelPatcher(model, load_device=load_device, offload_device=offload_device)
+    model_patcher = comfy.model_patcher.CoreModelPatcher(model, load_device=load_device, offload_device=offload_device)
     if not model_management.is_device_cpu(offload_device):
         model.to(offload_device)
     model.load_model_weights(new_sd, "", assign=model_patcher.is_dynamic())
@@ -1742,13 +1732,12 @@ def load_diffusion_model_state_dict(sd, model_options={}, metadata=None, disable
         logging.info("left over keys in diffusion model: {}".format(left_over))
     return model_patcher
 
-def load_diffusion_model(unet_path, model_options={}, disable_dynamic=False):
+def load_diffusion_model(unet_path, model_options={}):
     sd, metadata = comfy.utils.load_torch_file(unet_path, return_metadata=True)
-    model = load_diffusion_model_state_dict(sd, model_options=model_options, metadata=metadata, disable_dynamic=disable_dynamic)
+    model = load_diffusion_model_state_dict(sd, model_options=model_options, metadata=metadata)
     if model is None:
         logging.error("ERROR UNSUPPORTED DIFFUSION MODEL {}".format(unet_path))
         raise RuntimeError("ERROR: Could not detect model type of: {}\n{}".format(unet_path, model_detection_error_hint(unet_path, sd)))
-    model.cached_patcher_init = (load_diffusion_model, (unet_path, model_options))
     return model
 
 def load_unet(unet_path, dtype=None):

comfy/sd1_clip.py

@@ -308,14 +308,14 @@ class SDClipModel(torch.nn.Module, ClipTokenWeightEncoder):
     def load_sd(self, sd):
         return self.transformer.load_state_dict(sd, strict=False, assign=getattr(self, "can_assign_sd", False))
 
-    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+    def generate(self, tokens, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[]):
         if isinstance(tokens, dict):
             tokens_only = next(iter(tokens.values())) # todo: get this better?
         else:
             tokens_only = tokens
         tokens_only = [[t[0] for t in b] for b in tokens_only]
         embeds = self.process_tokens(tokens_only, device=self.execution_device)[0]
-        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed)
+        return self.transformer.generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens)
 
 def parse_parentheses(string):
     result = []
@@ -573,8 +573,6 @@ class SDTokenizer:
         min_length = tokenizer_options.get("{}_min_length".format(self.embedding_key), self.min_length)
         min_padding = tokenizer_options.get("{}_min_padding".format(self.embedding_key), self.min_padding)
 
-        min_length = kwargs.get("min_length", min_length)
-
         text = escape_important(text)
         if kwargs.get("disable_weights", self.disable_weights):
             parsed_weights = [(text, 1.0)]

comfy/text_encoders/anima.py

@@ -33,8 +33,6 @@ class AnimaTokenizer:
     def state_dict(self):
         return {}
 
-    def decode(self, token_ids, **kwargs):
-        return self.qwen3_06b.decode(token_ids, **kwargs)
 
 class Qwen3_06BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="last", layer_idx=None, dtype=None, attention_mask=True, model_options={}):

comfy/text_encoders/llama.py

@@ -105,7 +105,6 @@ class Qwen3_06BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_06B_ACE15_Config:
@@ -129,7 +128,6 @@ class Qwen3_06B_ACE15_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_2B_ACE15_lm_Config:
@@ -153,7 +151,6 @@ class Qwen3_2B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4B_ACE15_lm_Config:
@@ -177,7 +174,6 @@ class Qwen3_4B_ACE15_lm_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_4BConfig:
@@ -201,7 +197,6 @@ class Qwen3_4BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Qwen3_8BConfig:
@@ -225,7 +220,6 @@ class Qwen3_8BConfig:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [151643, 151645]
 
 @dataclass
 class Ovis25_2BConfig:
@@ -296,7 +290,6 @@ class Gemma2_2B_Config:
     rope_scale = None
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [1]
 
 @dataclass
 class Gemma3_4B_Config:
@@ -321,7 +314,6 @@ class Gemma3_4B_Config:
     rope_scale = [8.0, 1.0]
     final_norm: bool = True
     lm_head: bool = False
-    stop_tokens = [1, 106]
 
 GEMMA3_VISION_CONFIG = {"num_channels": 3, "hidden_act": "gelu_pytorch_tanh", "hidden_size": 1152, "image_size": 896, "intermediate_size": 4304, "model_type": "siglip_vision_model", "num_attention_heads": 16, "num_hidden_layers": 27, "patch_size": 14}
 
@@ -355,7 +347,6 @@ class Gemma3_12B_Config:
     lm_head: bool = False
     vision_config = GEMMA3_VISION_CONFIG
     mm_tokens_per_image = 256
-    stop_tokens = [1, 106]
 
 class RMSNorm(nn.Module):
     def __init__(self, dim: int, eps: float = 1e-5, add=False, device=None, dtype=None):
@@ -812,13 +803,10 @@ class BaseGenerate:
         comfy.ops.uncast_bias_weight(module, weight, None, offload_stream)
         return x
 
-    def generate(self, embeds=None, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.9, min_p=0.0, repetition_penalty=1.0, seed=42, stop_tokens=None, initial_tokens=[], execution_dtype=None, min_tokens=0):
+    def generate(self, embeds=None, do_sample=True, max_length=256, temperature=1.0, top_k=50, top_p=0.9, min_p=0.0, repetition_penalty=1.0, seed=42, stop_tokens=[], initial_tokens=[], execution_dtype=None, min_tokens=0):
         device = embeds.device
         model_config = self.model.config
 
-        if stop_tokens is None:
-            stop_tokens = self.model.config.stop_tokens
-
         if execution_dtype is None:
             if comfy.model_management.should_use_bf16(device):
                 execution_dtype = torch.bfloat16
@@ -937,7 +925,7 @@ class Qwen25_3B(BaseLlama, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_06B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_06B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_06BConfig(**config_dict)
@@ -964,7 +952,7 @@ class Qwen3_2B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_4B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_4B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_4BConfig(**config_dict)
@@ -982,7 +970,7 @@ class Qwen3_4B_ACE15_lm(BaseLlama, BaseQwen3, torch.nn.Module):
         self.model = Llama2_(config, device=device, dtype=dtype, ops=operations)
         self.dtype = dtype
 
-class Qwen3_8B(BaseLlama, BaseQwen3, BaseGenerate, torch.nn.Module):
+class Qwen3_8B(BaseLlama, BaseQwen3, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Qwen3_8BConfig(**config_dict)
@@ -1046,7 +1034,7 @@ class Qwen25_7BVLI(BaseLlama, BaseGenerate, torch.nn.Module):
 
         return super().forward(x, attention_mask=attention_mask, embeds=embeds, num_tokens=num_tokens, intermediate_output=intermediate_output, final_layer_norm_intermediate=final_layer_norm_intermediate, dtype=dtype, position_ids=position_ids)
 
-class Gemma2_2B(BaseLlama, BaseGenerate, torch.nn.Module):
+class Gemma2_2B(BaseLlama, torch.nn.Module):
     def __init__(self, config_dict, dtype, device, operations):
         super().__init__()
         config = Gemma2_2B_Config(**config_dict)

comfy/text_encoders/lt.py

@@ -3,6 +3,7 @@ import os
 from transformers import T5TokenizerFast
 from .spiece_tokenizer import SPieceTokenizer
 import comfy.text_encoders.genmo
+from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector
 import torch
 import comfy.utils
 import math
@@ -101,7 +102,6 @@ class LTXAVTEModel(torch.nn.Module):
         super().__init__()
         self.dtypes = set()
         self.dtypes.add(dtype)
-        self.compat_mode = False
 
         self.gemma3_12b = Gemma3_12BModel(device=device, dtype=dtype_llama, model_options=model_options, layer="all", layer_idx=None)
         self.dtypes.add(dtype_llama)
@@ -109,11 +109,6 @@ class LTXAVTEModel(torch.nn.Module):
         operations = self.gemma3_12b.operations # TODO
         self.text_embedding_projection = operations.Linear(3840 * 49, 3840, bias=False, dtype=dtype, device=device)
 
-    def enable_compat_mode(self):  # TODO: remove
-        from comfy.ldm.lightricks.embeddings_connector import Embeddings1DConnector
-        operations = self.gemma3_12b.operations
-        dtype = self.text_embedding_projection.weight.dtype
-        device = self.text_embedding_projection.weight.device
         self.audio_embeddings_connector = Embeddings1DConnector(
             split_rope=True,
             double_precision_rope=True,
@@ -129,7 +124,6 @@ class LTXAVTEModel(torch.nn.Module):
             device=device,
             operations=operations,
         )
-        self.compat_mode = True
 
     def set_clip_options(self, options):
         self.execution_device = options.get("execution_device", self.execution_device)
@@ -152,11 +146,9 @@ class LTXAVTEModel(torch.nn.Module):
         out = out.reshape((out.shape[0], out.shape[1], -1))
         out = self.text_embedding_projection(out)
         out = out.float()
-
-        if self.compat_mode:
-            out_vid = self.video_embeddings_connector(out)[0]
-            out_audio = self.audio_embeddings_connector(out)[0]
-            out = torch.concat((out_vid, out_audio), dim=-1)
+        out_vid = self.video_embeddings_connector(out)[0]
+        out_audio = self.audio_embeddings_connector(out)[0]
+        out = torch.concat((out_vid, out_audio), dim=-1)
 
         return out.to(out_device), pooled
 
@@ -167,30 +159,20 @@ class LTXAVTEModel(torch.nn.Module):
         if "model.layers.47.self_attn.q_norm.weight" in sd:
             return self.gemma3_12b.load_sd(sd)
         else:
-            sdo = comfy.utils.state_dict_prefix_replace(sd, {"text_embedding_projection.aggregate_embed.weight": "text_embedding_projection.weight"}, filter_keys=True)
+            sdo = comfy.utils.state_dict_prefix_replace(sd, {"text_embedding_projection.aggregate_embed.weight": "text_embedding_projection.weight", "model.diffusion_model.video_embeddings_connector.": "video_embeddings_connector.", "model.diffusion_model.audio_embeddings_connector.": "audio_embeddings_connector."}, filter_keys=True)
             if len(sdo) == 0:
                 sdo = sd
 
             missing_all = []
             unexpected_all = []
 
-            for prefix, component in [("text_embedding_projection.", self.text_embedding_projection)]:
+            for prefix, component in [("text_embedding_projection.", self.text_embedding_projection), ("video_embeddings_connector.", self.video_embeddings_connector), ("audio_embeddings_connector.", self.audio_embeddings_connector)]:
                 component_sd = {k.replace(prefix, ""): v for k, v in sdo.items() if k.startswith(prefix)}
                 if component_sd:
                     missing, unexpected = component.load_state_dict(component_sd, strict=False, assign=getattr(self, "can_assign_sd", False))
                     missing_all.extend([f"{prefix}{k}" for k in missing])
                     unexpected_all.extend([f"{prefix}{k}" for k in unexpected])
 
-            if "model.diffusion_model.audio_embeddings_connector.transformer_1d_blocks.2.attn1.to_q.bias" not in sd:  # TODO: remove
-                ww = sd.get("model.diffusion_model.audio_embeddings_connector.transformer_1d_blocks.0.attn1.to_q.bias", None)
-                if ww is not None:
-                    if ww.shape[0] == 3840:
-                        self.enable_compat_mode()
-                        sdv = comfy.utils.state_dict_prefix_replace(sd, {"model.diffusion_model.video_embeddings_connector.": ""}, filter_keys=True)
-                        self.video_embeddings_connector.load_state_dict(sdv, strict=False, assign=getattr(self, "can_assign_sd", False))
-                        sda = comfy.utils.state_dict_prefix_replace(sd, {"model.diffusion_model.audio_embeddings_connector.": ""}, filter_keys=True)
-                        self.audio_embeddings_connector.load_state_dict(sda, strict=False, assign=getattr(self, "can_assign_sd", False))
-
             return (missing_all, unexpected_all)
 
     def memory_estimation_function(self, token_weight_pairs, device=None):

comfy/text_encoders/lumina2.py

@@ -31,6 +31,9 @@ class Gemma2_2BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
         super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma2_2B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
+    def generate(self, embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+        return super().generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[107])
+
 class Gemma3_4BModel(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
         llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)
@@ -40,6 +43,9 @@ class Gemma3_4BModel(sd1_clip.SDClipModel):
 
         super().__init__(device=device, layer=layer, layer_idx=layer_idx, textmodel_json_config={}, dtype=dtype, special_tokens={"start": 2, "pad": 0}, layer_norm_hidden_state=False, model_class=comfy.text_encoders.llama.Gemma3_4B, enable_attention_masks=attention_mask, return_attention_masks=attention_mask, model_options=model_options)
 
+    def generate(self, embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed):
+        return super().generate(embeds, do_sample, max_length, temperature, top_k, top_p, min_p, repetition_penalty, seed, stop_tokens=[106])
+
 class Gemma3_4B_Vision_Model(sd1_clip.SDClipModel):
     def __init__(self, device="cpu", layer="hidden", layer_idx=-2, dtype=None, attention_mask=True, model_options={}):
         llama_quantization_metadata = model_options.get("llama_quantization_metadata", None)

comfy/utils.py

@@ -29,7 +29,7 @@ import itertools
 from torch.nn.functional import interpolate
 from tqdm.auto import trange
 from einops import rearrange
-from comfy.cli_args import args
+from comfy.cli_args import args, enables_dynamic_vram
 import json
 import time
 import mmap
@@ -113,7 +113,7 @@ def load_torch_file(ckpt, safe_load=False, device=None, return_metadata=False):
     metadata = None
     if ckpt.lower().endswith(".safetensors") or ckpt.lower().endswith(".sft"):
         try:
-            if comfy.memory_management.aimdo_enabled:
+            if enables_dynamic_vram():
                 sd, metadata = load_safetensors(ckpt)
                 if not return_metadata:
                     metadata = None
@@ -1154,7 +1154,7 @@ def tiled_scale(samples, function, tile_x=64, tile_y=64, overlap = 8, upscale_am
     return tiled_scale_multidim(samples, function, (tile_y, tile_x), overlap=overlap, upscale_amount=upscale_amount, out_channels=out_channels, output_device=output_device, pbar=pbar)
 
 def model_trange(*args, **kwargs):
-    if not comfy.memory_management.aimdo_enabled:
+    if comfy.memory_management.aimdo_allocator is None:
         return trange(*args, **kwargs)
 
     pbar = trange(*args, **kwargs, smoothing=1.0)

comfy_api/latest/_input_impl/video_types.py

@@ -444,7 +444,7 @@ class VideoFromComponents(VideoInput):
             output.mux(packet)
 
             if audio_stream and self.__components.audio:
-                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().contiguous().numpy(), format='fltp', layout=layout)
+                frame = av.AudioFrame.from_ndarray(waveform.float().cpu().numpy(), format='fltp', layout=layout)
                 frame.sample_rate = audio_sample_rate
                 frame.pts = 0
                 output.mux(audio_stream.encode(frame))

comfy_api/latest/_io.py

@@ -73,7 +73,6 @@ class RemoteOptions:
 class NumberDisplay(str, Enum):
     number = "number"
     slider = "slider"
-    gradient_slider = "gradientslider"
 
 
 class ControlAfterGenerate(str, Enum):
@@ -297,15 +296,13 @@ class Float(ComfyTypeIO):
         '''Float input.'''
         def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None, lazy: bool=None,
                     default: float=None, min: float=None, max: float=None, step: float=None, round: float=None,
-                    display_mode: NumberDisplay=None, gradient_stops: list[list[float]]=None,
-                    socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
+                    display_mode: NumberDisplay=None, socketless: bool=None, force_input: bool=None, extra_dict=None, raw_link: bool=None, advanced: bool=None):
             super().__init__(id, display_name, optional, tooltip, lazy, default, socketless, None, force_input, extra_dict, raw_link, advanced)
             self.min = min
             self.max = max
             self.step = step
             self.round = round
             self.display_mode = display_mode
-            self.gradient_stops = gradient_stops
             self.default: float
 
         def as_dict(self):
@@ -315,7 +312,6 @@ class Float(ComfyTypeIO):
                 "step": self.step,
                 "round": self.round,
                 "display": self.display_mode,
-                "gradient_stops": self.gradient_stops,
             })
 
 @comfytype(io_type="STRING")

comfy_api_nodes/apis/bytedance.py

@@ -27,7 +27,6 @@ class Seedream4TaskCreationRequest(BaseModel):
     sequential_image_generation: str = Field("disabled")
     sequential_image_generation_options: Seedream4Options = Field(Seedream4Options(max_images=15))
     watermark: bool = Field(False)
-    output_format: str | None = None
 
 
 class ImageTaskCreationResponse(BaseModel):
@@ -107,7 +106,6 @@ RECOMMENDED_PRESETS_SEEDREAM_4 = [
     ("2496x1664 (3:2)", 2496, 1664),
     ("1664x2496 (2:3)", 1664, 2496),
     ("3024x1296 (21:9)", 3024, 1296),
-    ("3072x3072 (1:1)", 3072, 3072),
     ("4096x4096 (1:1)", 4096, 4096),
     ("Custom", None, None),
 ]

comfy_api_nodes/apis/elevenlabs.py

@@ -1,88 +0,0 @@
-from pydantic import BaseModel, Field
-
-
-class SpeechToTextRequest(BaseModel):
-    model_id: str = Field(...)
-    cloud_storage_url: str = Field(...)
-    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
-    tag_audio_events: bool | None = Field(None, description="Annotate sounds like (laughter) in transcript")
-    num_speakers: int | None = Field(None, description="Max speakers predicted")
-    timestamps_granularity: str = Field(default="word", description="Timing precision: none, word, or character")
-    diarize: bool | None = Field(None, description="Annotate which speaker is talking")
-    diarization_threshold: float | None = Field(None, description="Speaker separation sensitivity")
-    temperature: float | None = Field(None, description="Randomness control")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-
-
-class SpeechToTextWord(BaseModel):
-    text: str = Field(..., description="The word text")
-    type: str = Field(default="word", description="Type of text element (word, spacing, etc.)")
-    start: float | None = Field(None, description="Start time in seconds (when timestamps enabled)")
-    end: float | None = Field(None, description="End time in seconds (when timestamps enabled)")
-    speaker_id: str | None = Field(None, description="Speaker identifier when diarization is enabled")
-    logprob: float | None = Field(None, description="Log probability of the word")
-
-
-class SpeechToTextResponse(BaseModel):
-    language_code: str = Field(..., description="Detected or specified language code")
-    language_probability: float | None = Field(None, description="Confidence of language detection")
-    text: str = Field(..., description="Full transcript text")
-    words: list[SpeechToTextWord] | None = Field(None, description="Word-level timing information")
-
-
-class TextToSpeechVoiceSettings(BaseModel):
-    stability: float | None = Field(None, description="Voice stability")
-    similarity_boost: float | None = Field(None, description="Similarity boost")
-    style: float | None = Field(None, description="Style exaggeration")
-    use_speaker_boost: bool | None = Field(None, description="Boost similarity to original speaker")
-    speed: float | None = Field(None, description="Speech speed")
-
-
-class TextToSpeechRequest(BaseModel):
-    text: str = Field(..., description="Text to convert to speech")
-    model_id: str = Field(..., description="Model ID for TTS")
-    language_code: str | None = Field(None, description="ISO-639-1 or ISO-639-3 language code")
-    voice_settings: TextToSpeechVoiceSettings | None = Field(None, description="Voice settings")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")
-
-
-class TextToSoundEffectsRequest(BaseModel):
-    text: str = Field(..., description="Text prompt to convert into a sound effect")
-    duration_seconds: float = Field(..., description="Duration of generated sound in seconds")
-    prompt_influence: float = Field(..., description="How closely generation follows the prompt")
-    loop: bool | None = Field(None, description="Whether to create a smoothly looping sound effect")
-
-
-class AddVoiceRequest(BaseModel):
-    name: str = Field(..., description="Name that identifies the voice")
-    remove_background_noise: bool = Field(..., description="Remove background noise from voice samples")
-
-
-class AddVoiceResponse(BaseModel):
-    voice_id: str = Field(..., description="The newly created voice's unique identifier")
-
-
-class SpeechToSpeechRequest(BaseModel):
-    model_id: str = Field(..., description="Model ID for speech-to-speech")
-    voice_settings: str = Field(..., description="JSON string of voice settings")
-    seed: int = Field(..., description="Seed for deterministic sampling")
-    remove_background_noise: bool = Field(..., description="Remove background noise from input audio")
-
-
-class DialogueInput(BaseModel):
-    text: str = Field(..., description="Text content to convert to speech")
-    voice_id: str = Field(..., description="Voice identifier for this dialogue segment")
-
-
-class DialogueSettings(BaseModel):
-    stability: float | None = Field(None, description="Voice stability (0-1)")
-
-
-class TextToDialogueRequest(BaseModel):
-    inputs: list[DialogueInput] = Field(..., description="List of dialogue segments")
-    model_id: str = Field(..., description="Model ID for dialogue generation")
-    language_code: str | None = Field(None, description="ISO-639-1 language code")
-    settings: DialogueSettings | None = Field(None, description="Voice settings")
-    seed: int | None = Field(None, description="Seed for deterministic sampling")
-    apply_text_normalization: str | None = Field(None, description="Text normalization mode: auto, on, off")

comfy_api_nodes/apis/kling.py

@@ -134,13 +134,6 @@ class ImageToVideoWithAudioRequest(BaseModel):
     shot_type: str | None = Field(None)
 
 
-class KlingAvatarRequest(BaseModel):
-    image: str = Field(...)
-    sound_file: str = Field(...)
-    prompt: str | None = Field(None)
-    mode: str = Field(...)
-
-
 class MotionControlRequest(BaseModel):
     prompt: str = Field(...)
     image_url: str = Field(...)

comfy_api_nodes/nodes_bytedance.py

@@ -37,12 +37,6 @@ from comfy_api_nodes.util import (
 
 BYTEPLUS_IMAGE_ENDPOINT = "/proxy/byteplus/api/v3/images/generations"
 
-SEEDREAM_MODELS = {
-    "seedream 5.0 lite": "seedream-5-0-260128",
-    "seedream-4-5-251128": "seedream-4-5-251128",
-    "seedream-4-0-250828": "seedream-4-0-250828",
-}
-
 # Long-running tasks endpoints(e.g., video)
 BYTEPLUS_TASK_ENDPOINT = "/proxy/byteplus/api/v3/contents/generations/tasks"
 BYTEPLUS_TASK_STATUS_ENDPOINT = "/proxy/byteplus/api/v3/contents/generations/tasks"  # + /{task_id}
@@ -186,13 +180,14 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ByteDanceSeedreamNode",
-            display_name="ByteDance Seedream 5.0",
+            display_name="ByteDance Seedream 4.5",
             category="api node/image/ByteDance",
             description="Unified text-to-image generation and precise single-sentence editing at up to 4K resolution.",
             inputs=[
                 IO.Combo.Input(
                     "model",
-                    options=list(SEEDREAM_MODELS.keys()),
+                    options=["seedream-4-5-251128", "seedream-4-0-250828"],
+                    tooltip="Model name",
                 ),
                 IO.String.Input(
                     "prompt",
@@ -203,7 +198,7 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                 IO.Image.Input(
                     "image",
                     tooltip="Input image(s) for image-to-image generation. "
-                    "Reference image(s) for single or multi-reference generation.",
+                    "List of 1-10 images for single or multi-reference generation.",
                     optional=True,
                 ),
                 IO.Combo.Input(
@@ -215,8 +210,8 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                     "width",
                     default=2048,
                     min=1024,
-                    max=6240,
-                    step=2,
+                    max=4096,
+                    step=8,
                     tooltip="Custom width for image. Value is working only if `size_preset` is set to `Custom`",
                     optional=True,
                 ),
@@ -224,8 +219,8 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                     "height",
                     default=2048,
                     min=1024,
-                    max=4992,
-                    step=2,
+                    max=4096,
+                    step=8,
                     tooltip="Custom height for image. Value is working only if `size_preset` is set to `Custom`",
                     optional=True,
                 ),
@@ -288,8 +283,7 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                 depends_on=IO.PriceBadgeDepends(widgets=["model"]),
                 expr="""
                 (
-                  $price := $contains(widgets.model, "5.0 lite") ? 0.035 :
-                            $contains(widgets.model, "4-5") ? 0.04 : 0.03;
+                  $price := $contains(widgets.model, "seedream-4-5-251128") ? 0.04 : 0.03;
                   {
                     "type":"usd",
                     "usd": $price,
@@ -315,7 +309,6 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
         watermark: bool = False,
         fail_on_partial: bool = True,
     ) -> IO.NodeOutput:
-        model = SEEDREAM_MODELS[model]
         validate_string(prompt, strip_whitespace=True, min_length=1)
         w = h = None
         for label, tw, th in RECOMMENDED_PRESETS_SEEDREAM_4:
@@ -325,12 +318,15 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
 
         if w is None or h is None:
             w, h = width, height
-
+            if not (1024 <= w <= 4096) or not (1024 <= h <= 4096):
+                raise ValueError(
+                    f"Custom size out of range: {w}x{h}. " "Both width and height must be between 1024 and 4096 pixels."
+                )
         out_num_pixels = w * h
         mp_provided = out_num_pixels / 1_000_000.0
-        if ("seedream-4-5" in model or "seedream-5-0" in model) and out_num_pixels < 3686400:
+        if "seedream-4-5" in model and out_num_pixels < 3686400:
             raise ValueError(
-                f"Minimum image resolution for the selected model is 3.68MP, "
+                f"Minimum image resolution that Seedream 4.5 can generate is 3.68MP, "
                 f"but {mp_provided:.2f}MP provided."
             )
         if "seedream-4-0" in model and out_num_pixels < 921600:
@@ -338,18 +334,9 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                 f"Minimum image resolution that the selected model can generate is 0.92MP, "
                 f"but {mp_provided:.2f}MP provided."
             )
-        max_pixels = 10_404_496 if "seedream-5-0" in model else 16_777_216
-        if out_num_pixels > max_pixels:
-            raise ValueError(
-                f"Maximum image resolution for the selected model is {max_pixels / 1_000_000:.2f}MP, "
-                f"but {mp_provided:.2f}MP provided."
-            )
         n_input_images = get_number_of_images(image) if image is not None else 0
-        max_num_of_images = 14 if model == "seedream-5-0-260128" else 10
-        if n_input_images > max_num_of_images:
-            raise ValueError(
-                f"Maximum of {max_num_of_images} reference images are supported, but {n_input_images} received."
-            )
+        if n_input_images > 10:
+            raise ValueError(f"Maximum of 10 reference images are supported, but {n_input_images} received.")
         if sequential_image_generation == "auto" and n_input_images + max_images > 15:
             raise ValueError(
                 "The maximum number of generated images plus the number of reference images cannot exceed 15."
@@ -377,7 +364,6 @@ class ByteDanceSeedreamNode(IO.ComfyNode):
                 sequential_image_generation=sequential_image_generation,
                 sequential_image_generation_options=Seedream4Options(max_images=max_images),
                 watermark=watermark,
-                output_format="png" if model == "seedream-5-0-260128" else None,
             ),
         )
         if len(response.data) == 1:

comfy_api_nodes/nodes_elevenlabs.py

@@ -1,924 +0,0 @@
-import json
-import uuid
-
-from typing_extensions import override
-
-from comfy_api.latest import IO, ComfyExtension, Input
-from comfy_api_nodes.apis.elevenlabs import (
-    AddVoiceRequest,
-    AddVoiceResponse,
-    DialogueInput,
-    DialogueSettings,
-    SpeechToSpeechRequest,
-    SpeechToTextRequest,
-    SpeechToTextResponse,
-    TextToDialogueRequest,
-    TextToSoundEffectsRequest,
-    TextToSpeechRequest,
-    TextToSpeechVoiceSettings,
-)
-from comfy_api_nodes.util import (
-    ApiEndpoint,
-    audio_bytes_to_audio_input,
-    audio_ndarray_to_bytesio,
-    audio_tensor_to_contiguous_ndarray,
-    sync_op,
-    sync_op_raw,
-    upload_audio_to_comfyapi,
-    validate_string,
-)
-
-ELEVENLABS_MUSIC_SECTIONS = "ELEVENLABS_MUSIC_SECTIONS"  # Custom type for music sections
-ELEVENLABS_COMPOSITION_PLAN = "ELEVENLABS_COMPOSITION_PLAN"  # Custom type for composition plan
-ELEVENLABS_VOICE = "ELEVENLABS_VOICE"  # Custom type for voice selection
-
-# Predefined ElevenLabs voices: (voice_id, display_name, gender, accent)
-ELEVENLABS_VOICES = [
-    ("CwhRBWXzGAHq8TQ4Fs17", "Roger", "male", "american"),
-    ("EXAVITQu4vr4xnSDxMaL", "Sarah", "female", "american"),
-    ("FGY2WhTYpPnrIDTdsKH5", "Laura", "female", "american"),
-    ("IKne3meq5aSn9XLyUdCD", "Charlie", "male", "australian"),
-    ("JBFqnCBsd6RMkjVDRZzb", "George", "male", "british"),
-    ("N2lVS1w4EtoT3dr4eOWO", "Callum", "male", "american"),
-    ("SAz9YHcvj6GT2YYXdXww", "River", "neutral", "american"),
-    ("SOYHLrjzK2X1ezoPC6cr", "Harry", "male", "american"),
-    ("TX3LPaxmHKxFdv7VOQHJ", "Liam", "male", "american"),
-    ("Xb7hH8MSUJpSbSDYk0k2", "Alice", "female", "british"),
-    ("XrExE9yKIg1WjnnlVkGX", "Matilda", "female", "american"),
-    ("bIHbv24MWmeRgasZH58o", "Will", "male", "american"),
-    ("cgSgspJ2msm6clMCkdW9", "Jessica", "female", "american"),
-    ("cjVigY5qzO86Huf0OWal", "Eric", "male", "american"),
-    ("hpp4J3VqNfWAUOO0d1Us", "Bella", "female", "american"),
-    ("iP95p4xoKVk53GoZ742B", "Chris", "male", "american"),
-    ("nPczCjzI2devNBz1zQrb", "Brian", "male", "american"),
-    ("onwK4e9ZLuTAKqWW03F9", "Daniel", "male", "british"),
-    ("pFZP5JQG7iQjIQuC4Bku", "Lily", "female", "british"),
-    ("pNInz6obpgDQGcFmaJgB", "Adam", "male", "american"),
-    ("pqHfZKP75CvOlQylNhV4", "Bill", "male", "american"),
-]
-
-ELEVENLABS_VOICE_OPTIONS = [f"{name} ({gender}, {accent})" for _, name, gender, accent in ELEVENLABS_VOICES]
-ELEVENLABS_VOICE_MAP = {
-    f"{name} ({gender}, {accent})": voice_id for voice_id, name, gender, accent in ELEVENLABS_VOICES
-}
-
-
-class ElevenLabsSpeechToText(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsSpeechToText",
-            display_name="ElevenLabs Speech to Text",
-            category="api node/audio/ElevenLabs",
-            description="Transcribe audio to text. "
-            "Supports automatic language detection, speaker diarization, and audio event tagging.",
-            inputs=[
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Audio to transcribe.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "scribe_v2",
-                            [
-                                IO.Boolean.Input(
-                                    "tag_audio_events",
-                                    default=False,
-                                    tooltip="Annotate sounds like (laughter), (music), etc. in transcript.",
-                                ),
-                                IO.Boolean.Input(
-                                    "diarize",
-                                    default=False,
-                                    tooltip="Annotate which speaker is talking.",
-                                ),
-                                IO.Float.Input(
-                                    "diarization_threshold",
-                                    default=0.22,
-                                    min=0.1,
-                                    max=0.4,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speaker separation sensitivity. "
-                                    "Lower values are more sensitive to speaker changes.",
-                                ),
-                                IO.Float.Input(
-                                    "temperature",
-                                    default=0.0,
-                                    min=0.0,
-                                    max=2.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Randomness control. "
-                                    "0.0 uses model default. Higher values increase randomness.",
-                                ),
-                                IO.Combo.Input(
-                                    "timestamps_granularity",
-                                    options=["word", "character", "none"],
-                                    default="word",
-                                    tooltip="Timing precision for transcript words.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for transcription.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "num_speakers",
-                    default=0,
-                    min=0,
-                    max=32,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Maximum number of speakers to predict. Set to 0 for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=2147483647,
-                    tooltip="Seed for reproducibility (determinism not guaranteed).",
-                ),
-            ],
-            outputs=[
-                IO.String.Output(display_name="text"),
-                IO.String.Output(display_name="language_code"),
-                IO.String.Output(display_name="words_json"),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.0073,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        audio: Input.Audio,
-        model: dict,
-        language_code: str,
-        num_speakers: int,
-        seed: int,
-    ) -> IO.NodeOutput:
-        if model["diarize"] and num_speakers:
-            raise ValueError(
-                "Number of speakers cannot be specified when diarization is enabled. "
-                "Either disable diarization or set num_speakers to 0."
-            )
-        request = SpeechToTextRequest(
-            model_id=model["model"],
-            cloud_storage_url=await upload_audio_to_comfyapi(
-                cls, audio, container_format="mp4", codec_name="aac", mime_type="audio/mp4"
-            ),
-            language_code=language_code if language_code.strip() else None,
-            tag_audio_events=model["tag_audio_events"],
-            num_speakers=num_speakers if num_speakers > 0 else None,
-            timestamps_granularity=model["timestamps_granularity"],
-            diarize=model["diarize"],
-            diarization_threshold=model["diarization_threshold"] if model["diarize"] else None,
-            seed=seed,
-            temperature=model["temperature"],
-        )
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/speech-to-text", method="POST"),
-            response_model=SpeechToTextResponse,
-            data=request,
-            content_type="multipart/form-data",
-        )
-        words_json = json.dumps(
-            [w.model_dump(exclude_none=True) for w in response.words] if response.words else [],
-            indent=2,
-        )
-        return IO.NodeOutput(response.text, response.language_code, words_json)
-
-
-class ElevenLabsVoiceSelector(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsVoiceSelector",
-            display_name="ElevenLabs Voice Selector",
-            category="api node/audio/ElevenLabs",
-            description="Select a predefined ElevenLabs voice for text-to-speech generation.",
-            inputs=[
-                IO.Combo.Input(
-                    "voice",
-                    options=ELEVENLABS_VOICE_OPTIONS,
-                    tooltip="Choose a voice from the predefined ElevenLabs voices.",
-                ),
-            ],
-            outputs=[
-                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
-            ],
-            is_api_node=False,
-        )
-
-    @classmethod
-    def execute(cls, voice: str) -> IO.NodeOutput:
-        voice_id = ELEVENLABS_VOICE_MAP.get(voice)
-        if not voice_id:
-            raise ValueError(f"Unknown voice: {voice}")
-        return IO.NodeOutput(voice_id)
-
-
-class ElevenLabsTextToSpeech(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToSpeech",
-            display_name="ElevenLabs Text to Speech",
-            category="api node/audio/ElevenLabs",
-            description="Convert text to speech.",
-            inputs=[
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    "voice",
-                    tooltip="Voice to use for speech synthesis. Connect from Voice Selector or Instant Voice Clone.",
-                ),
-                IO.String.Input(
-                    "text",
-                    multiline=True,
-                    default="",
-                    tooltip="The text to convert to speech.",
-                ),
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.01,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.Combo.Input(
-                    "apply_text_normalization",
-                    options=["auto", "on", "off"],
-                    tooltip="Text normalization mode. 'auto' lets the system decide, "
-                    "'on' always applies normalization, 'off' skips it.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_multilingual_v2",
-                            [
-                                IO.Float.Input(
-                                    "speed",
-                                    default=1.0,
-                                    min=0.7,
-                                    max=1.3,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-                                ),
-                                IO.Float.Input(
-                                    "similarity_boost",
-                                    default=0.75,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-                                ),
-                                IO.Boolean.Input(
-                                    "use_speaker_boost",
-                                    default=False,
-                                    tooltip="Boost similarity to the original speaker voice.",
-                                ),
-                                IO.Float.Input(
-                                    "style",
-                                    default=0.0,
-                                    min=0.0,
-                                    max=0.2,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Style exaggeration. Higher values increase stylistic expression "
-                                    "but may reduce stability.",
-                                ),
-                            ],
-                        ),
-                        IO.DynamicCombo.Option(
-                            "eleven_v3",
-                            [
-                                IO.Float.Input(
-                                    "speed",
-                                    default=1.0,
-                                    min=0.7,
-                                    max=1.3,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-                                ),
-                                IO.Float.Input(
-                                    "similarity_boost",
-                                    default=0.75,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for text-to-speech.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=2147483647,
-                    tooltip="Seed for reproducibility (determinism not guaranteed).",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        voice: str,
-        text: str,
-        stability: float,
-        apply_text_normalization: str,
-        model: dict,
-        language_code: str,
-        seed: int,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        validate_string(text, min_length=1)
-        request = TextToSpeechRequest(
-            text=text,
-            model_id=model["model"],
-            language_code=language_code if language_code.strip() else None,
-            voice_settings=TextToSpeechVoiceSettings(
-                stability=stability,
-                similarity_boost=model["similarity_boost"],
-                speed=model["speed"],
-                use_speaker_boost=model.get("use_speaker_boost", None),
-                style=model.get("style", None),
-            ),
-            seed=seed,
-            apply_text_normalization=apply_text_normalization,
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path=f"/proxy/elevenlabs/v1/text-to-speech/{voice}",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=request,
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsAudioIsolation(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsAudioIsolation",
-            display_name="ElevenLabs Voice Isolation",
-            category="api node/audio/ElevenLabs",
-            description="Remove background noise from audio, isolating vocals or speech.",
-            inputs=[
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Audio to process for background noise removal.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        audio: Input.Audio,
-    ) -> IO.NodeOutput:
-        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
-        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/audio-isolation", method="POST"),
-            files={"audio": ("audio.mp4", audio_bytes_io, "audio/mp4")},
-            content_type="multipart/form-data",
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsTextToSoundEffects(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToSoundEffects",
-            display_name="ElevenLabs Text to Sound Effects",
-            category="api node/audio/ElevenLabs",
-            description="Generate sound effects from text descriptions.",
-            inputs=[
-                IO.String.Input(
-                    "text",
-                    multiline=True,
-                    default="",
-                    tooltip="Text description of the sound effect to generate.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_sfx_v2",
-                            [
-                                IO.Float.Input(
-                                    "duration",
-                                    default=5.0,
-                                    min=0.5,
-                                    max=30.0,
-                                    step=0.1,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="Duration of generated sound in seconds.",
-                                ),
-                                IO.Boolean.Input(
-                                    "loop",
-                                    default=False,
-                                    tooltip="Create a smoothly looping sound effect.",
-                                ),
-                                IO.Float.Input(
-                                    "prompt_influence",
-                                    default=0.3,
-                                    min=0.0,
-                                    max=1.0,
-                                    step=0.01,
-                                    display_mode=IO.NumberDisplay.slider,
-                                    tooltip="How closely generation follows the prompt. "
-                                    "Higher values make the sound follow the text more closely.",
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Model to use for sound effect generation.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.14,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        text: str,
-        model: dict,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        validate_string(text, min_length=1)
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path="/proxy/elevenlabs/v1/sound-generation",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=TextToSoundEffectsRequest(
-                text=text,
-                duration_seconds=model["duration"],
-                prompt_influence=model["prompt_influence"],
-                loop=model.get("loop", None),
-            ),
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsInstantVoiceClone(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsInstantVoiceClone",
-            display_name="ElevenLabs Instant Voice Clone",
-            category="api node/audio/ElevenLabs",
-            description="Create a cloned voice from audio samples. "
-            "Provide 1-8 audio recordings of the voice to clone.",
-            inputs=[
-                IO.Autogrow.Input(
-                    "files",
-                    template=IO.Autogrow.TemplatePrefix(
-                        IO.Audio.Input("audio"),
-                        prefix="audio",
-                        min=1,
-                        max=8,
-                    ),
-                    tooltip="Audio recordings for voice cloning.",
-                ),
-                IO.Boolean.Input(
-                    "remove_background_noise",
-                    default=False,
-                    tooltip="Remove background noise from voice samples using audio isolation.",
-                ),
-            ],
-            outputs=[
-                IO.Custom(ELEVENLABS_VOICE).Output(display_name="voice"),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(expr="""{"type":"usd","usd":0.15}"""),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        files: IO.Autogrow.Type,
-        remove_background_noise: bool,
-    ) -> IO.NodeOutput:
-        file_tuples: list[tuple[str, tuple[str, bytes, str]]] = []
-        for key in files:
-            audio = files[key]
-            sample_rate: int = audio["sample_rate"]
-            waveform = audio["waveform"]
-            audio_data_np = audio_tensor_to_contiguous_ndarray(waveform)
-            audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, sample_rate, "mp4", "aac")
-            file_tuples.append(("files", (f"{key}.mp4", audio_bytes_io.getvalue(), "audio/mp4")))
-
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/elevenlabs/v1/voices/add", method="POST"),
-            response_model=AddVoiceResponse,
-            data=AddVoiceRequest(
-                name=str(uuid.uuid4()),
-                remove_background_noise=remove_background_noise,
-            ),
-            files=file_tuples,
-            content_type="multipart/form-data",
-        )
-        return IO.NodeOutput(response.voice_id)
-
-
-ELEVENLABS_STS_VOICE_SETTINGS = [
-    IO.Float.Input(
-        "speed",
-        default=1.0,
-        min=0.7,
-        max=1.3,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Speech speed. 1.0 is normal, <1.0 slower, >1.0 faster.",
-    ),
-    IO.Float.Input(
-        "similarity_boost",
-        default=0.75,
-        min=0.0,
-        max=1.0,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Similarity boost. Higher values make the voice more similar to the original.",
-    ),
-    IO.Boolean.Input(
-        "use_speaker_boost",
-        default=False,
-        tooltip="Boost similarity to the original speaker voice.",
-    ),
-    IO.Float.Input(
-        "style",
-        default=0.0,
-        min=0.0,
-        max=0.2,
-        step=0.01,
-        display_mode=IO.NumberDisplay.slider,
-        tooltip="Style exaggeration. Higher values increase stylistic expression but may reduce stability.",
-    ),
-]
-
-
-class ElevenLabsSpeechToSpeech(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsSpeechToSpeech",
-            display_name="ElevenLabs Speech to Speech",
-            category="api node/audio/ElevenLabs",
-            description="Transform speech from one voice to another while preserving the original content and emotion.",
-            inputs=[
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    "voice",
-                    tooltip="Target voice for the transformation. "
-                    "Connect from Voice Selector or Instant Voice Clone.",
-                ),
-                IO.Audio.Input(
-                    "audio",
-                    tooltip="Source audio to transform.",
-                ),
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.01,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.DynamicCombo.Input(
-                    "model",
-                    options=[
-                        IO.DynamicCombo.Option(
-                            "eleven_multilingual_sts_v2",
-                            ELEVENLABS_STS_VOICE_SETTINGS,
-                        ),
-                        IO.DynamicCombo.Option(
-                            "eleven_english_sts_v2",
-                            ELEVENLABS_STS_VOICE_SETTINGS,
-                        ),
-                    ],
-                    tooltip="Model to use for speech-to-speech transformation.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=0,
-                    min=0,
-                    max=4294967295,
-                    tooltip="Seed for reproducibility.",
-                ),
-                IO.Boolean.Input(
-                    "remove_background_noise",
-                    default=False,
-                    tooltip="Remove background noise from input audio using audio isolation.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/minute"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        voice: str,
-        audio: Input.Audio,
-        stability: float,
-        model: dict,
-        output_format: str,
-        seed: int,
-        remove_background_noise: bool,
-    ) -> IO.NodeOutput:
-        audio_data_np = audio_tensor_to_contiguous_ndarray(audio["waveform"])
-        audio_bytes_io = audio_ndarray_to_bytesio(audio_data_np, audio["sample_rate"], "mp4", "aac")
-        voice_settings = TextToSpeechVoiceSettings(
-            stability=stability,
-            similarity_boost=model["similarity_boost"],
-            style=model["style"],
-            use_speaker_boost=model["use_speaker_boost"],
-            speed=model["speed"],
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path=f"/proxy/elevenlabs/v1/speech-to-speech/{voice}",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=SpeechToSpeechRequest(
-                model_id=model["model"],
-                voice_settings=voice_settings.model_dump_json(exclude_none=True),
-                seed=seed,
-                remove_background_noise=remove_background_noise,
-            ),
-            files={"audio": ("audio.mp4", audio_bytes_io.getvalue(), "audio/mp4")},
-            content_type="multipart/form-data",
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-def _generate_dialogue_inputs(count: int) -> list:
-    """Generate input widgets for a given number of dialogue entries."""
-    inputs = []
-    for i in range(1, count + 1):
-        inputs.extend(
-            [
-                IO.String.Input(
-                    f"text{i}",
-                    multiline=True,
-                    default="",
-                    tooltip=f"Text content for dialogue entry {i}.",
-                ),
-                IO.Custom(ELEVENLABS_VOICE).Input(
-                    f"voice{i}",
-                    tooltip=f"Voice for dialogue entry {i}. Connect from Voice Selector or Instant Voice Clone.",
-                ),
-            ]
-        )
-    return inputs
-
-
-class ElevenLabsTextToDialogue(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="ElevenLabsTextToDialogue",
-            display_name="ElevenLabs Text to Dialogue",
-            category="api node/audio/ElevenLabs",
-            description="Generate multi-speaker dialogue from text. Each dialogue entry has its own text and voice.",
-            inputs=[
-                IO.Float.Input(
-                    "stability",
-                    default=0.5,
-                    min=0.0,
-                    max=1.0,
-                    step=0.5,
-                    display_mode=IO.NumberDisplay.slider,
-                    tooltip="Voice stability. Lower values give broader emotional range, "
-                    "higher values produce more consistent but potentially monotonous speech.",
-                ),
-                IO.Combo.Input(
-                    "apply_text_normalization",
-                    options=["auto", "on", "off"],
-                    tooltip="Text normalization mode. 'auto' lets the system decide, "
-                    "'on' always applies normalization, 'off' skips it.",
-                ),
-                IO.Combo.Input(
-                    "model",
-                    options=["eleven_v3"],
-                    tooltip="Model to use for dialogue generation.",
-                ),
-                IO.DynamicCombo.Input(
-                    "inputs",
-                    options=[
-                        IO.DynamicCombo.Option("1", _generate_dialogue_inputs(1)),
-                        IO.DynamicCombo.Option("2", _generate_dialogue_inputs(2)),
-                        IO.DynamicCombo.Option("3", _generate_dialogue_inputs(3)),
-                        IO.DynamicCombo.Option("4", _generate_dialogue_inputs(4)),
-                        IO.DynamicCombo.Option("5", _generate_dialogue_inputs(5)),
-                        IO.DynamicCombo.Option("6", _generate_dialogue_inputs(6)),
-                        IO.DynamicCombo.Option("7", _generate_dialogue_inputs(7)),
-                        IO.DynamicCombo.Option("8", _generate_dialogue_inputs(8)),
-                        IO.DynamicCombo.Option("9", _generate_dialogue_inputs(9)),
-                        IO.DynamicCombo.Option("10", _generate_dialogue_inputs(10)),
-                    ],
-                    tooltip="Number of dialogue entries.",
-                ),
-                IO.String.Input(
-                    "language_code",
-                    default="",
-                    tooltip="ISO-639-1 or ISO-639-3 language code (e.g., 'en', 'es', 'fra'). "
-                    "Leave empty for automatic detection.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=1,
-                    min=0,
-                    max=4294967295,
-                    tooltip="Seed for reproducibility.",
-                ),
-                IO.Combo.Input(
-                    "output_format",
-                    options=["mp3_44100_192", "opus_48000_192"],
-                    tooltip="Audio output format.",
-                ),
-            ],
-            outputs=[
-                IO.Audio.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                expr="""{"type":"usd","usd":0.24,"format":{"approximate":true,"suffix":"/1K chars"}}""",
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        stability: float,
-        apply_text_normalization: str,
-        model: str,
-        inputs: dict,
-        language_code: str,
-        seed: int,
-        output_format: str,
-    ) -> IO.NodeOutput:
-        num_entries = int(inputs["inputs"])
-        dialogue_inputs: list[DialogueInput] = []
-        for i in range(1, num_entries + 1):
-            text = inputs[f"text{i}"]
-            voice_id = inputs[f"voice{i}"]
-            validate_string(text, min_length=1)
-            dialogue_inputs.append(DialogueInput(text=text, voice_id=voice_id))
-        request = TextToDialogueRequest(
-            inputs=dialogue_inputs,
-            model_id=model,
-            language_code=language_code if language_code.strip() else None,
-            settings=DialogueSettings(stability=stability),
-            seed=seed,
-            apply_text_normalization=apply_text_normalization,
-        )
-        response = await sync_op_raw(
-            cls,
-            ApiEndpoint(
-                path="/proxy/elevenlabs/v1/text-to-dialogue",
-                method="POST",
-                query_params={"output_format": output_format},
-            ),
-            data=request,
-            as_binary=True,
-        )
-        return IO.NodeOutput(audio_bytes_to_audio_input(response))
-
-
-class ElevenLabsExtension(ComfyExtension):
-    @override
-    async def get_node_list(self) -> list[type[IO.ComfyNode]]:
-        return [
-            ElevenLabsSpeechToText,
-            ElevenLabsVoiceSelector,
-            ElevenLabsTextToSpeech,
-            ElevenLabsAudioIsolation,
-            ElevenLabsTextToSoundEffects,
-            ElevenLabsInstantVoiceClone,
-            ElevenLabsSpeechToSpeech,
-            ElevenLabsTextToDialogue,
-        ]
-
-
-async def comfy_entrypoint() -> ElevenLabsExtension:
-    return ElevenLabsExtension()

comfy_api_nodes/nodes_kling.py

@@ -50,7 +50,6 @@ from comfy_api_nodes.apis import (
 )
 from comfy_api_nodes.apis.kling import (
     ImageToVideoWithAudioRequest,
-    KlingAvatarRequest,
     MotionControlRequest,
     MultiPromptEntry,
     OmniImageParamImage,
@@ -75,7 +74,6 @@ from comfy_api_nodes.util import (
     upload_image_to_comfyapi,
     upload_images_to_comfyapi,
     upload_video_to_comfyapi,
-    validate_audio_duration,
     validate_image_aspect_ratio,
     validate_image_dimensions,
     validate_string,
@@ -3141,103 +3139,6 @@ class KlingFirstLastFrameNode(IO.ComfyNode):
         return IO.NodeOutput(await download_url_to_video_output(final_response.data.task_result.videos[0].url))
 
 
-class KlingAvatarNode(IO.ComfyNode):
-
-    @classmethod
-    def define_schema(cls) -> IO.Schema:
-        return IO.Schema(
-            node_id="KlingAvatarNode",
-            display_name="Kling Avatar 2.0",
-            category="api node/video/Kling",
-            description="Generate broadcast-style digital human videos from a single photo and an audio file.",
-            inputs=[
-                IO.Image.Input(
-                    "image",
-                    tooltip="Avatar reference image. "
-                    "Width and height must be at least 300px. Aspect ratio must be between 1:2.5 and 2.5:1.",
-                ),
-                IO.Audio.Input(
-                    "sound_file",
-                    tooltip="Audio input. Must be between 2 and 300 seconds in duration.",
-                ),
-                IO.Combo.Input("mode", options=["std", "pro"]),
-                IO.String.Input(
-                    "prompt",
-                    multiline=True,
-                    default="",
-                    optional=True,
-                    tooltip="Optional prompt to define avatar actions, emotions, and camera movements.",
-                ),
-                IO.Int.Input(
-                    "seed",
-                    default=0,
-                    min=0,
-                    max=2147483647,
-                    display_mode=IO.NumberDisplay.number,
-                    control_after_generate=True,
-                    tooltip="Seed controls whether the node should re-run; "
-                    "results are non-deterministic regardless of seed.",
-                ),
-            ],
-            outputs=[
-                IO.Video.Output(),
-            ],
-            hidden=[
-                IO.Hidden.auth_token_comfy_org,
-                IO.Hidden.api_key_comfy_org,
-                IO.Hidden.unique_id,
-            ],
-            is_api_node=True,
-            price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["mode"]),
-                expr="""
-                (
-                  $prices := {"std": 0.056, "pro": 0.112};
-                  {"type":"usd","usd": $lookup($prices, widgets.mode), "format":{"suffix":"/second"}}
-                )
-                """,
-            ),
-        )
-
-    @classmethod
-    async def execute(
-        cls,
-        image: Input.Image,
-        sound_file: Input.Audio,
-        mode: str,
-        seed: int,
-        prompt: str = "",
-    ) -> IO.NodeOutput:
-        validate_image_dimensions(image, min_width=300, min_height=300)
-        validate_image_aspect_ratio(image, (1, 2.5), (2.5, 1))
-        validate_audio_duration(sound_file, min_duration=2, max_duration=300)
-        response = await sync_op(
-            cls,
-            ApiEndpoint(path="/proxy/kling/v1/videos/avatar/image2video", method="POST"),
-            response_model=TaskStatusResponse,
-            data=KlingAvatarRequest(
-                image=await upload_image_to_comfyapi(cls, image),
-                sound_file=await upload_audio_to_comfyapi(
-                    cls, sound_file, container_format="mp3", codec_name="libmp3lame", mime_type="audio/mpeg"
-                ),
-                prompt=prompt or None,
-                mode=mode,
-            ),
-        )
-        if response.code:
-            raise RuntimeError(
-                f"Kling request failed. Code: {response.code}, Message: {response.message}, Data: {response.data}"
-            )
-        final_response = await poll_op(
-            cls,
-            ApiEndpoint(path=f"/proxy/kling/v1/videos/avatar/image2video/{response.data.task_id}"),
-            response_model=TaskStatusResponse,
-            status_extractor=lambda r: (r.data.task_status if r.data else None),
-            max_poll_attempts=800,
-        )
-        return IO.NodeOutput(await download_url_to_video_output(final_response.data.task_result.videos[0].url))
-
-
 class KlingExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -3266,7 +3167,6 @@ class KlingExtension(ComfyExtension):
             MotionControl,
             KlingVideoNode,
             KlingFirstLastFrameNode,
-            KlingAvatarNode,
         ]
 
 

comfy_api_nodes/util/__init__.py

@@ -9,8 +9,6 @@ from .client import (
 from .conversions import (
     audio_bytes_to_audio_input,
     audio_input_to_mp3,
-    audio_ndarray_to_bytesio,
-    audio_tensor_to_contiguous_ndarray,
     audio_to_base64_string,
     bytesio_to_image_tensor,
     convert_mask_to_image,
@@ -80,8 +78,6 @@ __all__ = [
     # Conversions
     "audio_bytes_to_audio_input",
     "audio_input_to_mp3",
-    "audio_ndarray_to_bytesio",
-    "audio_tensor_to_contiguous_ndarray",
     "audio_to_base64_string",
     "bytesio_to_image_tensor",
     "convert_mask_to_image",

comfy_extras/nodes_canny.py

@@ -10,7 +10,6 @@ class Canny(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="Canny",
-            display_name="Canny",
             search_aliases=["edge detection", "outline", "contour detection", "line art"],
             category="image/preprocessors",
             essentials_category="Image Tools",

comfy_extras/nodes_glsl.py

@@ -1,895 +0,0 @@
-import os
-import sys
-import re
-import logging
-import ctypes.util
-import importlib.util
-from typing import TypedDict
-
-import numpy as np
-import torch
-
-import nodes
-from comfy_api.latest import ComfyExtension, io, ui
-from typing_extensions import override
-from utils.install_util import get_missing_requirements_message
-
-logger = logging.getLogger(__name__)
-
-
-def _check_opengl_availability():
-    """Early check for OpenGL availability. Raises RuntimeError if unlikely to work."""
-    logger.debug("_check_opengl_availability: starting")
-    missing = []
-
-    # Check Python packages (using find_spec to avoid importing)
-    logger.debug("_check_opengl_availability: checking for glfw package")
-    if importlib.util.find_spec("glfw") is None:
-        missing.append("glfw")
-
-    logger.debug("_check_opengl_availability: checking for OpenGL package")
-    if importlib.util.find_spec("OpenGL") is None:
-        missing.append("PyOpenGL")
-
-    if missing:
-        raise RuntimeError(
-            f"OpenGL dependencies not available.\n{get_missing_requirements_message()}\n"
-        )
-
-    # On Linux without display, check if headless backends are available
-    logger.debug(f"_check_opengl_availability: platform={sys.platform}")
-    if sys.platform.startswith("linux"):
-        has_display = os.environ.get("DISPLAY") or os.environ.get("WAYLAND_DISPLAY")
-        logger.debug(f"_check_opengl_availability: has_display={bool(has_display)}")
-        if not has_display:
-            # Check for EGL or OSMesa libraries
-            logger.debug("_check_opengl_availability: checking for EGL library")
-            has_egl = ctypes.util.find_library("EGL")
-            logger.debug("_check_opengl_availability: checking for OSMesa library")
-            has_osmesa = ctypes.util.find_library("OSMesa")
-
-            # Error disabled for CI as it fails this check
-            # if not has_egl and not has_osmesa:
-            #     raise RuntimeError(
-            #         "GLSL Shader node: No display and no headless backend (EGL/OSMesa) found.\n"
-            #         "See error below for installation instructions."
-            #     )
-            logger.debug(f"Headless mode: EGL={'yes' if has_egl else 'no'}, OSMesa={'yes' if has_osmesa else 'no'}")
-
-    logger.debug("_check_opengl_availability: completed")
-
-
-# Run early check at import time
-logger.debug("nodes_glsl: running _check_opengl_availability at import time")
-_check_opengl_availability()
-
-# OpenGL modules - initialized lazily when context is created
-gl = None
-glfw = None
-EGL = None
-
-
-def _import_opengl():
-    """Import OpenGL module. Called after context is created."""
-    global gl
-    if gl is None:
-        logger.debug("_import_opengl: importing OpenGL.GL")
-        import OpenGL.GL as _gl
-        gl = _gl
-        logger.debug("_import_opengl: import completed")
-    return gl
-
-
-class SizeModeInput(TypedDict):
-    size_mode: str
-    width: int
-    height: int
-
-
-MAX_IMAGES = 5      # u_image0-4
-MAX_UNIFORMS = 5    # u_float0-4, u_int0-4
-MAX_OUTPUTS = 4     # fragColor0-3 (MRT)
-
-# Vertex shader using gl_VertexID trick - no VBO needed.
-# Draws a single triangle that covers the entire screen:
-#
-#     (-1,3)
-#       /|
-#      / |  <- visible area is the unit square from (-1,-1) to (1,1)
-#     /  |     parts outside get clipped away
-# (-1,-1)---(3,-1)
-#
-# v_texCoord is computed from clip space: * 0.5 + 0.5 maps (-1,1) -> (0,1)
-VERTEX_SHADER = """#version 330 core
-out vec2 v_texCoord;
-void main() {
-    vec2 verts[3] = vec2[](vec2(-1, -1), vec2(3, -1), vec2(-1, 3));
-    v_texCoord = verts[gl_VertexID] * 0.5 + 0.5;
-    gl_Position = vec4(verts[gl_VertexID], 0, 1);
-}
-"""
-
-DEFAULT_FRAGMENT_SHADER = """#version 300 es
-precision highp float;
-
-uniform sampler2D u_image0;
-uniform vec2 u_resolution;
-
-in vec2 v_texCoord;
-layout(location = 0) out vec4 fragColor0;
-
-void main() {
-    fragColor0 = texture(u_image0, v_texCoord);
-}
-"""
-
-
-def _convert_es_to_desktop(source: str) -> str:
-    """Convert GLSL ES (WebGL) shader source to desktop GLSL 330 core."""
-    # Remove any existing #version directive
-    source = re.sub(r"#version\s+\d+(\s+es)?\s*\n?", "", source, flags=re.IGNORECASE)
-    # Remove precision qualifiers (not needed in desktop GLSL)
-    source = re.sub(r"precision\s+(lowp|mediump|highp)\s+\w+\s*;\s*\n?", "", source)
-    # Prepend desktop GLSL version
-    return "#version 330 core\n" + source
-
-
-def _detect_output_count(source: str) -> int:
-    """Detect how many fragColor outputs are used in the shader.
-
-    Returns the count of outputs needed (1 to MAX_OUTPUTS).
-    """
-    matches = re.findall(r"fragColor(\d+)", source)
-    if not matches:
-        return 1  # Default to 1 output if none found
-    max_index = max(int(m) for m in matches)
-    return min(max_index + 1, MAX_OUTPUTS)
-
-
-def _detect_pass_count(source: str) -> int:
-    """Detect multi-pass rendering from #pragma passes N directive.
-
-    Returns the number of passes (1 if not specified).
-    """
-    match = re.search(r'#pragma\s+passes\s+(\d+)', source)
-    if match:
-        return max(1, int(match.group(1)))
-    return 1
-
-
-def _init_glfw():
-    """Initialize GLFW. Returns (window, glfw_module). Raises RuntimeError on failure."""
-    logger.debug("_init_glfw: starting")
-    # On macOS, glfw.init() must be called from main thread or it hangs forever
-    if sys.platform == "darwin":
-        logger.debug("_init_glfw: skipping on macOS")
-        raise RuntimeError("GLFW backend not supported on macOS")
-
-    logger.debug("_init_glfw: importing glfw module")
-    import glfw as _glfw
-
-    logger.debug("_init_glfw: calling glfw.init()")
-    if not _glfw.init():
-        raise RuntimeError("glfw.init() failed")
-
-    try:
-        logger.debug("_init_glfw: setting window hints")
-        _glfw.window_hint(_glfw.VISIBLE, _glfw.FALSE)
-        _glfw.window_hint(_glfw.CONTEXT_VERSION_MAJOR, 3)
-        _glfw.window_hint(_glfw.CONTEXT_VERSION_MINOR, 3)
-        _glfw.window_hint(_glfw.OPENGL_PROFILE, _glfw.OPENGL_CORE_PROFILE)
-
-        logger.debug("_init_glfw: calling create_window()")
-        window = _glfw.create_window(64, 64, "ComfyUI GLSL", None, None)
-        if not window:
-            raise RuntimeError("glfw.create_window() failed")
-
-        logger.debug("_init_glfw: calling make_context_current()")
-        _glfw.make_context_current(window)
-        logger.debug("_init_glfw: completed successfully")
-        return window, _glfw
-    except Exception:
-        logger.debug("_init_glfw: failed, terminating glfw")
-        _glfw.terminate()
-        raise
-
-
-def _init_egl():
-    """Initialize EGL for headless rendering. Returns (display, context, surface, EGL_module). Raises RuntimeError on failure."""
-    logger.debug("_init_egl: starting")
-    from OpenGL import EGL as _EGL
-    from OpenGL.EGL import (
-        eglGetDisplay, eglInitialize, eglChooseConfig, eglCreateContext,
-        eglMakeCurrent, eglCreatePbufferSurface, eglBindAPI,
-        eglTerminate, eglDestroyContext, eglDestroySurface,
-        EGL_DEFAULT_DISPLAY, EGL_NO_CONTEXT, EGL_NONE,
-        EGL_SURFACE_TYPE, EGL_PBUFFER_BIT, EGL_RENDERABLE_TYPE, EGL_OPENGL_BIT,
-        EGL_RED_SIZE, EGL_GREEN_SIZE, EGL_BLUE_SIZE, EGL_ALPHA_SIZE, EGL_DEPTH_SIZE,
-        EGL_WIDTH, EGL_HEIGHT, EGL_OPENGL_API,
-    )
-    logger.debug("_init_egl: imports completed")
-
-    display = None
-    context = None
-    surface = None
-
-    try:
-        logger.debug("_init_egl: calling eglGetDisplay()")
-        display = eglGetDisplay(EGL_DEFAULT_DISPLAY)
-        if display == _EGL.EGL_NO_DISPLAY:
-            raise RuntimeError("eglGetDisplay() failed")
-
-        logger.debug("_init_egl: calling eglInitialize()")
-        major, minor = _EGL.EGLint(), _EGL.EGLint()
-        if not eglInitialize(display, major, minor):
-            display = None  # Not initialized, don't terminate
-            raise RuntimeError("eglInitialize() failed")
-        logger.debug(f"_init_egl: EGL version {major.value}.{minor.value}")
-
-        config_attribs = [
-            EGL_SURFACE_TYPE, EGL_PBUFFER_BIT,
-            EGL_RENDERABLE_TYPE, EGL_OPENGL_BIT,
-            EGL_RED_SIZE, 8, EGL_GREEN_SIZE, 8, EGL_BLUE_SIZE, 8, EGL_ALPHA_SIZE, 8,
-            EGL_DEPTH_SIZE, 0, EGL_NONE
-        ]
-        configs = (_EGL.EGLConfig * 1)()
-        num_configs = _EGL.EGLint()
-        if not eglChooseConfig(display, config_attribs, configs, 1, num_configs) or num_configs.value == 0:
-            raise RuntimeError("eglChooseConfig() failed")
-        config = configs[0]
-        logger.debug(f"_init_egl: config chosen, num_configs={num_configs.value}")
-
-        if not eglBindAPI(EGL_OPENGL_API):
-            raise RuntimeError("eglBindAPI() failed")
-
-        logger.debug("_init_egl: calling eglCreateContext()")
-        context_attribs = [
-            _EGL.EGL_CONTEXT_MAJOR_VERSION, 3,
-            _EGL.EGL_CONTEXT_MINOR_VERSION, 3,
-            _EGL.EGL_CONTEXT_OPENGL_PROFILE_MASK, _EGL.EGL_CONTEXT_OPENGL_CORE_PROFILE_BIT,
-            EGL_NONE
-        ]
-        context = eglCreateContext(display, config, EGL_NO_CONTEXT, context_attribs)
-        if context == EGL_NO_CONTEXT:
-            raise RuntimeError("eglCreateContext() failed")
-
-        logger.debug("_init_egl: calling eglCreatePbufferSurface()")
-        pbuffer_attribs = [EGL_WIDTH, 64, EGL_HEIGHT, 64, EGL_NONE]
-        surface = eglCreatePbufferSurface(display, config, pbuffer_attribs)
-        if surface == _EGL.EGL_NO_SURFACE:
-            raise RuntimeError("eglCreatePbufferSurface() failed")
-
-        logger.debug("_init_egl: calling eglMakeCurrent()")
-        if not eglMakeCurrent(display, surface, surface, context):
-            raise RuntimeError("eglMakeCurrent() failed")
-
-        logger.debug("_init_egl: completed successfully")
-        return display, context, surface, _EGL
-
-    except Exception:
-        logger.debug("_init_egl: failed, cleaning up")
-        # Clean up any resources on failure
-        if surface is not None:
-            eglDestroySurface(display, surface)
-        if context is not None:
-            eglDestroyContext(display, context)
-        if display is not None:
-            eglTerminate(display)
-        raise
-
-
-def _init_osmesa():
-    """Initialize OSMesa for software rendering. Returns (context, buffer). Raises RuntimeError on failure."""
-    import ctypes
-
-    logger.debug("_init_osmesa: starting")
-    os.environ["PYOPENGL_PLATFORM"] = "osmesa"
-
-    logger.debug("_init_osmesa: importing OpenGL.osmesa")
-    from OpenGL import GL as _gl
-    from OpenGL.osmesa import (
-        OSMesaCreateContextExt, OSMesaMakeCurrent, OSMesaDestroyContext,
-        OSMESA_RGBA,
-    )
-    logger.debug("_init_osmesa: imports completed")
-
-    ctx = OSMesaCreateContextExt(OSMESA_RGBA, 24, 0, 0, None)
-    if not ctx:
-        raise RuntimeError("OSMesaCreateContextExt() failed")
-
-    width, height = 64, 64
-    buffer = (ctypes.c_ubyte * (width * height * 4))()
-
-    logger.debug("_init_osmesa: calling OSMesaMakeCurrent()")
-    if not OSMesaMakeCurrent(ctx, buffer, _gl.GL_UNSIGNED_BYTE, width, height):
-        OSMesaDestroyContext(ctx)
-        raise RuntimeError("OSMesaMakeCurrent() failed")
-
-    logger.debug("_init_osmesa: completed successfully")
-    return ctx, buffer
-
-
-class GLContext:
-    """Manages OpenGL context and resources for shader execution.
-
-    Tries backends in order: GLFW (desktop) → EGL (headless GPU) → OSMesa (software).
-    """
-
-    _instance = None
-    _initialized = False
-
-    def __new__(cls):
-        if cls._instance is None:
-            cls._instance = super().__new__(cls)
-        return cls._instance
-
-    def __init__(self):
-        if GLContext._initialized:
-            logger.debug("GLContext.__init__: already initialized, skipping")
-            return
-
-        logger.debug("GLContext.__init__: starting initialization")
-
-        global glfw, EGL
-
-        import time
-        start = time.perf_counter()
-
-        self._backend = None
-        self._window = None
-        self._egl_display = None
-        self._egl_context = None
-        self._egl_surface = None
-        self._osmesa_ctx = None
-        self._osmesa_buffer = None
-        self._vao = None
-
-        # Try backends in order: GLFW → EGL → OSMesa
-        errors = []
-
-        logger.debug("GLContext.__init__: trying GLFW backend")
-        try:
-            self._window, glfw = _init_glfw()
-            self._backend = "glfw"
-            logger.debug("GLContext.__init__: GLFW backend succeeded")
-        except Exception as e:
-            logger.debug(f"GLContext.__init__: GLFW backend failed: {e}")
-            errors.append(("GLFW", e))
-
-        if self._backend is None:
-            logger.debug("GLContext.__init__: trying EGL backend")
-            try:
-                self._egl_display, self._egl_context, self._egl_surface, EGL = _init_egl()
-                self._backend = "egl"
-                logger.debug("GLContext.__init__: EGL backend succeeded")
-            except Exception as e:
-                logger.debug(f"GLContext.__init__: EGL backend failed: {e}")
-                errors.append(("EGL", e))
-
-        if self._backend is None:
-            logger.debug("GLContext.__init__: trying OSMesa backend")
-            try:
-                self._osmesa_ctx, self._osmesa_buffer = _init_osmesa()
-                self._backend = "osmesa"
-                logger.debug("GLContext.__init__: OSMesa backend succeeded")
-            except Exception as e:
-                logger.debug(f"GLContext.__init__: OSMesa backend failed: {e}")
-                errors.append(("OSMesa", e))
-
-        if self._backend is None:
-            if sys.platform == "win32":
-                platform_help = (
-                    "Windows: Ensure GPU drivers are installed and display is available.\n"
-                    "         CPU-only/headless mode is not supported on Windows."
-                )
-            elif sys.platform == "darwin":
-                platform_help = (
-                    "macOS: GLFW is not supported.\n"
-                    "  Install OSMesa via Homebrew: brew install mesa\n"
-                    "  Then: pip install PyOpenGL PyOpenGL-accelerate"
-                )
-            else:
-                platform_help = (
-                    "Linux: Install one of these backends:\n"
-                    "  Desktop:           sudo apt install libgl1-mesa-glx libglfw3\n"
-                    "  Headless with GPU: sudo apt install libegl1-mesa libgl1-mesa-dri\n"
-                    "  Headless (CPU):    sudo apt install libosmesa6"
-                )
-
-            error_details = "\n".join(f"  {name}: {err}" for name, err in errors)
-            raise RuntimeError(
-                f"Failed to create OpenGL context.\n\n"
-                f"Backend errors:\n{error_details}\n\n"
-                f"{platform_help}"
-            )
-
-        # Now import OpenGL.GL (after context is current)
-        logger.debug("GLContext.__init__: importing OpenGL.GL")
-        _import_opengl()
-
-        # Create VAO (required for core profile, but OSMesa may use compat profile)
-        logger.debug("GLContext.__init__: creating VAO")
-        try:
-            vao = gl.glGenVertexArrays(1)
-            gl.glBindVertexArray(vao)
-            self._vao = vao  # Only store after successful bind
-            logger.debug("GLContext.__init__: VAO created successfully")
-        except Exception as e:
-            logger.debug(f"GLContext.__init__: VAO creation failed (may be expected for OSMesa): {e}")
-            # OSMesa with older Mesa may not support VAOs
-            # Clean up if we created but couldn't bind
-            if vao:
-                try:
-                    gl.glDeleteVertexArrays(1, [vao])
-                except Exception:
-                    pass
-
-        elapsed = (time.perf_counter() - start) * 1000
-
-        # Log device info
-        renderer = gl.glGetString(gl.GL_RENDERER)
-        vendor = gl.glGetString(gl.GL_VENDOR)
-        version = gl.glGetString(gl.GL_VERSION)
-        renderer = renderer.decode() if renderer else "Unknown"
-        vendor = vendor.decode() if vendor else "Unknown"
-        version = version.decode() if version else "Unknown"
-
-        GLContext._initialized = True
-        logger.info(f"GLSL context initialized in {elapsed:.1f}ms ({self._backend}) - {renderer} ({vendor}), GL {version}")
-
-    def make_current(self):
-        if self._backend == "glfw":
-            glfw.make_context_current(self._window)
-        elif self._backend == "egl":
-            from OpenGL.EGL import eglMakeCurrent
-            eglMakeCurrent(self._egl_display, self._egl_surface, self._egl_surface, self._egl_context)
-        elif self._backend == "osmesa":
-            from OpenGL.osmesa import OSMesaMakeCurrent
-            OSMesaMakeCurrent(self._osmesa_ctx, self._osmesa_buffer, gl.GL_UNSIGNED_BYTE, 64, 64)
-
-        if self._vao is not None:
-            gl.glBindVertexArray(self._vao)
-
-
-def _compile_shader(source: str, shader_type: int) -> int:
-    """Compile a shader and return its ID."""
-    shader = gl.glCreateShader(shader_type)
-    gl.glShaderSource(shader, source)
-    gl.glCompileShader(shader)
-
-    if gl.glGetShaderiv(shader, gl.GL_COMPILE_STATUS) != gl.GL_TRUE:
-        error = gl.glGetShaderInfoLog(shader).decode()
-        gl.glDeleteShader(shader)
-        raise RuntimeError(f"Shader compilation failed:\n{error}")
-
-    return shader
-
-
-def _create_program(vertex_source: str, fragment_source: str) -> int:
-    """Create and link a shader program."""
-    vertex_shader = _compile_shader(vertex_source, gl.GL_VERTEX_SHADER)
-    try:
-        fragment_shader = _compile_shader(fragment_source, gl.GL_FRAGMENT_SHADER)
-    except RuntimeError:
-        gl.glDeleteShader(vertex_shader)
-        raise
-
-    program = gl.glCreateProgram()
-    gl.glAttachShader(program, vertex_shader)
-    gl.glAttachShader(program, fragment_shader)
-    gl.glLinkProgram(program)
-
-    gl.glDeleteShader(vertex_shader)
-    gl.glDeleteShader(fragment_shader)
-
-    if gl.glGetProgramiv(program, gl.GL_LINK_STATUS) != gl.GL_TRUE:
-        error = gl.glGetProgramInfoLog(program).decode()
-        gl.glDeleteProgram(program)
-        raise RuntimeError(f"Program linking failed:\n{error}")
-
-    return program
-
-
-def _render_shader_batch(
-    fragment_code: str,
-    width: int,
-    height: int,
-    image_batches: list[list[np.ndarray]],
-    floats: list[float],
-    ints: list[int],
-) -> list[list[np.ndarray]]:
-    """
-    Render a fragment shader for multiple batches efficiently.
-
-    Compiles shader once, reuses framebuffer/textures across batches.
-    Supports multi-pass rendering via #pragma passes N directive.
-
-    Args:
-        fragment_code: User's fragment shader code
-        width: Output width
-        height: Output height
-        image_batches: List of batches, each batch is a list of input images (H, W, C) float32 [0,1]
-        floats: List of float uniforms
-        ints: List of int uniforms
-
-    Returns:
-        List of batch outputs, each is a list of output images (H, W, 4) float32 [0,1]
-    """
-    import time
-    start_time = time.perf_counter()
-
-    if not image_batches:
-        return []
-
-    ctx = GLContext()
-    ctx.make_current()
-
-    # Convert from GLSL ES to desktop GLSL 330
-    fragment_source = _convert_es_to_desktop(fragment_code)
-
-    # Detect how many outputs the shader actually uses
-    num_outputs = _detect_output_count(fragment_code)
-
-    # Detect multi-pass rendering
-    num_passes = _detect_pass_count(fragment_code)
-
-    # Track resources for cleanup
-    program = None
-    fbo = None
-    output_textures = []
-    input_textures = []
-    ping_pong_textures = []
-    ping_pong_fbos = []
-
-    num_inputs = len(image_batches[0])
-
-    try:
-        # Compile shaders (once for all batches)
-        try:
-            program = _create_program(VERTEX_SHADER, fragment_source)
-        except RuntimeError:
-            logger.error(f"Fragment shader:\n{fragment_source}")
-            raise
-
-        gl.glUseProgram(program)
-
-        # Create framebuffer with only the needed color attachments
-        fbo = gl.glGenFramebuffers(1)
-        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, fbo)
-
-        draw_buffers = []
-        for i in range(num_outputs):
-            tex = gl.glGenTextures(1)
-            output_textures.append(tex)
-            gl.glBindTexture(gl.GL_TEXTURE_2D, tex)
-            gl.glTexImage2D(gl.GL_TEXTURE_2D, 0, gl.GL_RGBA32F, width, height, 0, gl.GL_RGBA, gl.GL_FLOAT, None)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MIN_FILTER, gl.GL_LINEAR)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MAG_FILTER, gl.GL_LINEAR)
-            gl.glFramebufferTexture2D(gl.GL_FRAMEBUFFER, gl.GL_COLOR_ATTACHMENT0 + i, gl.GL_TEXTURE_2D, tex, 0)
-            draw_buffers.append(gl.GL_COLOR_ATTACHMENT0 + i)
-
-        gl.glDrawBuffers(num_outputs, draw_buffers)
-
-        if gl.glCheckFramebufferStatus(gl.GL_FRAMEBUFFER) != gl.GL_FRAMEBUFFER_COMPLETE:
-            raise RuntimeError("Framebuffer is not complete")
-
-        # Create ping-pong resources for multi-pass rendering
-        if num_passes > 1:
-            for _ in range(2):
-                pp_tex = gl.glGenTextures(1)
-                ping_pong_textures.append(pp_tex)
-                gl.glBindTexture(gl.GL_TEXTURE_2D, pp_tex)
-                gl.glTexImage2D(gl.GL_TEXTURE_2D, 0, gl.GL_RGBA32F, width, height, 0, gl.GL_RGBA, gl.GL_FLOAT, None)
-                gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MIN_FILTER, gl.GL_LINEAR)
-                gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MAG_FILTER, gl.GL_LINEAR)
-                gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_S, gl.GL_CLAMP_TO_EDGE)
-                gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_T, gl.GL_CLAMP_TO_EDGE)
-
-                pp_fbo = gl.glGenFramebuffers(1)
-                ping_pong_fbos.append(pp_fbo)
-                gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, pp_fbo)
-                gl.glFramebufferTexture2D(gl.GL_FRAMEBUFFER, gl.GL_COLOR_ATTACHMENT0, gl.GL_TEXTURE_2D, pp_tex, 0)
-                gl.glDrawBuffers(1, [gl.GL_COLOR_ATTACHMENT0])
-
-                if gl.glCheckFramebufferStatus(gl.GL_FRAMEBUFFER) != gl.GL_FRAMEBUFFER_COMPLETE:
-                    raise RuntimeError("Ping-pong framebuffer is not complete")
-
-        # Create input textures (reused for all batches)
-        for i in range(num_inputs):
-            tex = gl.glGenTextures(1)
-            input_textures.append(tex)
-            gl.glActiveTexture(gl.GL_TEXTURE0 + i)
-            gl.glBindTexture(gl.GL_TEXTURE_2D, tex)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MIN_FILTER, gl.GL_LINEAR)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_MAG_FILTER, gl.GL_LINEAR)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_S, gl.GL_CLAMP_TO_EDGE)
-            gl.glTexParameteri(gl.GL_TEXTURE_2D, gl.GL_TEXTURE_WRAP_T, gl.GL_CLAMP_TO_EDGE)
-
-            loc = gl.glGetUniformLocation(program, f"u_image{i}")
-            if loc >= 0:
-                gl.glUniform1i(loc, i)
-
-        # Set static uniforms (once for all batches)
-        loc = gl.glGetUniformLocation(program, "u_resolution")
-        if loc >= 0:
-            gl.glUniform2f(loc, float(width), float(height))
-
-        for i, v in enumerate(floats):
-            loc = gl.glGetUniformLocation(program, f"u_float{i}")
-            if loc >= 0:
-                gl.glUniform1f(loc, v)
-
-        for i, v in enumerate(ints):
-            loc = gl.glGetUniformLocation(program, f"u_int{i}")
-            if loc >= 0:
-                gl.glUniform1i(loc, v)
-
-        # Get u_pass uniform location for multi-pass
-        pass_loc = gl.glGetUniformLocation(program, "u_pass")
-
-        gl.glViewport(0, 0, width, height)
-        gl.glDisable(gl.GL_BLEND)  # Ensure no alpha blending - write output directly
-
-        # Process each batch
-        all_batch_outputs = []
-        for images in image_batches:
-            # Update input textures with this batch's images
-            for i, img in enumerate(images):
-                gl.glActiveTexture(gl.GL_TEXTURE0 + i)
-                gl.glBindTexture(gl.GL_TEXTURE_2D, input_textures[i])
-
-                # Flip vertically for GL coordinates, ensure RGBA
-                h, w, c = img.shape
-                if c == 3:
-                    img_upload = np.empty((h, w, 4), dtype=np.float32)
-                    img_upload[:, :, :3] = img[::-1, :, :]
-                    img_upload[:, :, 3] = 1.0
-                else:
-                    img_upload = np.ascontiguousarray(img[::-1, :, :])
-
-                gl.glTexImage2D(gl.GL_TEXTURE_2D, 0, gl.GL_RGBA32F, w, h, 0, gl.GL_RGBA, gl.GL_FLOAT, img_upload)
-
-            if num_passes == 1:
-                # Single pass - render directly to output FBO
-                gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, fbo)
-                if pass_loc >= 0:
-                    gl.glUniform1i(pass_loc, 0)
-                gl.glClearColor(0, 0, 0, 0)
-                gl.glClear(gl.GL_COLOR_BUFFER_BIT)
-                gl.glDrawArrays(gl.GL_TRIANGLES, 0, 3)
-            else:
-                # Multi-pass rendering with ping-pong
-                for p in range(num_passes):
-                    is_last_pass = (p == num_passes - 1)
-
-                    # Set pass uniform
-                    if pass_loc >= 0:
-                        gl.glUniform1i(pass_loc, p)
-
-                    if is_last_pass:
-                        # Last pass renders to the main output FBO
-                        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, fbo)
-                    else:
-                        # Intermediate passes render to ping-pong FBO
-                        target_fbo = ping_pong_fbos[p % 2]
-                        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, target_fbo)
-
-                    # Set input texture for this pass
-                    gl.glActiveTexture(gl.GL_TEXTURE0)
-                    if p == 0:
-                        # First pass reads from original input
-                        gl.glBindTexture(gl.GL_TEXTURE_2D, input_textures[0])
-                    else:
-                        # Subsequent passes read from previous pass output
-                        source_tex = ping_pong_textures[(p - 1) % 2]
-                        gl.glBindTexture(gl.GL_TEXTURE_2D, source_tex)
-
-                    gl.glClearColor(0, 0, 0, 0)
-                    gl.glClear(gl.GL_COLOR_BUFFER_BIT)
-                    gl.glDrawArrays(gl.GL_TRIANGLES, 0, 3)
-
-            # Read back outputs for this batch
-            # (glGetTexImage is synchronous, implicitly waits for rendering)
-            batch_outputs = []
-            for tex in output_textures:
-                gl.glBindTexture(gl.GL_TEXTURE_2D, tex)
-                data = gl.glGetTexImage(gl.GL_TEXTURE_2D, 0, gl.GL_RGBA, gl.GL_FLOAT)
-                img = np.frombuffer(data, dtype=np.float32).reshape(height, width, 4)
-                batch_outputs.append(img[::-1, :, :].copy())
-
-            # Pad with black images for unused outputs
-            black_img = np.zeros((height, width, 4), dtype=np.float32)
-            for _ in range(num_outputs, MAX_OUTPUTS):
-                batch_outputs.append(black_img)
-
-            all_batch_outputs.append(batch_outputs)
-
-        elapsed = (time.perf_counter() - start_time) * 1000
-        num_batches = len(image_batches)
-        pass_info = f", {num_passes} passes" if num_passes > 1 else ""
-        logger.info(f"GLSL shader executed in {elapsed:.1f}ms ({num_batches} batch{'es' if num_batches != 1 else ''}, {width}x{height}{pass_info})")
-
-        return all_batch_outputs
-
-    finally:
-        # Unbind before deleting
-        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, 0)
-        gl.glUseProgram(0)
-
-        for tex in input_textures:
-            gl.glDeleteTextures(tex)
-        for tex in output_textures:
-            gl.glDeleteTextures(tex)
-        for tex in ping_pong_textures:
-            gl.glDeleteTextures(tex)
-        if fbo is not None:
-            gl.glDeleteFramebuffers(1, [fbo])
-        for pp_fbo in ping_pong_fbos:
-            gl.glDeleteFramebuffers(1, [pp_fbo])
-        if program is not None:
-            gl.glDeleteProgram(program)
-
-class GLSLShader(io.ComfyNode):
-
-    @classmethod
-    def define_schema(cls) -> io.Schema:
-        image_template = io.Autogrow.TemplatePrefix(
-            io.Image.Input("image"),
-            prefix="image",
-            min=1,
-            max=MAX_IMAGES,
-        )
-
-        float_template = io.Autogrow.TemplatePrefix(
-            io.Float.Input("float", default=0.0),
-            prefix="u_float",
-            min=0,
-            max=MAX_UNIFORMS,
-        )
-
-        int_template = io.Autogrow.TemplatePrefix(
-            io.Int.Input("int", default=0),
-            prefix="u_int",
-            min=0,
-            max=MAX_UNIFORMS,
-        )
-
-        return io.Schema(
-            node_id="GLSLShader",
-            display_name="GLSL Shader",
-            category="image/shader",
-            description=(
-                "Apply GLSL ES fragment shaders to images. "
-                "u_resolution (vec2) is always available."
-            ),
-            inputs=[
-                io.String.Input(
-                    "fragment_shader",
-                    default=DEFAULT_FRAGMENT_SHADER,
-                    multiline=True,
-                    tooltip="GLSL fragment shader source code (GLSL ES 3.00 / WebGL 2.0 compatible)",
-                ),
-                io.DynamicCombo.Input(
-                    "size_mode",
-                    options=[
-                        io.DynamicCombo.Option("from_input", []),
-                        io.DynamicCombo.Option(
-                            "custom",
-                            [
-                                io.Int.Input(
-                                    "width",
-                                    default=512,
-                                    min=1,
-                                    max=nodes.MAX_RESOLUTION,
-                                ),
-                                io.Int.Input(
-                                    "height",
-                                    default=512,
-                                    min=1,
-                                    max=nodes.MAX_RESOLUTION,
-                                ),
-                            ],
-                        ),
-                    ],
-                    tooltip="Output size: 'from_input' uses first input image dimensions, 'custom' allows manual size",
-                ),
-                io.Autogrow.Input("images", template=image_template, tooltip=f"Images are available as u_image0-{MAX_IMAGES-1} (sampler2D) in the shader code"),
-                io.Autogrow.Input("floats", template=float_template, tooltip=f"Floats are available as u_float0-{MAX_UNIFORMS-1} in the shader code"),
-                io.Autogrow.Input("ints", template=int_template, tooltip=f"Ints are available as u_int0-{MAX_UNIFORMS-1} in the shader code"),
-            ],
-            outputs=[
-                io.Image.Output(display_name="IMAGE0", tooltip="Available via layout(location = 0) out vec4 fragColor0 in the shader code"),
-                io.Image.Output(display_name="IMAGE1", tooltip="Available via layout(location = 1) out vec4 fragColor1 in the shader code"),
-                io.Image.Output(display_name="IMAGE2", tooltip="Available via layout(location = 2) out vec4 fragColor2 in the shader code"),
-                io.Image.Output(display_name="IMAGE3", tooltip="Available via layout(location = 3) out vec4 fragColor3 in the shader code"),
-            ],
-        )
-
-    @classmethod
-    def execute(
-        cls,
-        fragment_shader: str,
-        size_mode: SizeModeInput,
-        images: io.Autogrow.Type,
-        floats: io.Autogrow.Type = None,
-        ints: io.Autogrow.Type = None,
-        **kwargs,
-    ) -> io.NodeOutput:
-        image_list = [v for v in images.values() if v is not None]
-        float_list = (
-            [v if v is not None else 0.0 for v in floats.values()] if floats else []
-        )
-        int_list = [v if v is not None else 0 for v in ints.values()] if ints else []
-
-        if not image_list:
-            raise ValueError("At least one input image is required")
-
-        # Determine output dimensions
-        if size_mode["size_mode"] == "custom":
-            out_width = size_mode["width"]
-            out_height = size_mode["height"]
-        else:
-            out_height, out_width = image_list[0].shape[1:3]
-
-        batch_size = image_list[0].shape[0]
-
-        # Prepare batches
-        image_batches = []
-        for batch_idx in range(batch_size):
-            batch_images = [img_tensor[batch_idx].cpu().numpy().astype(np.float32) for img_tensor in image_list]
-            image_batches.append(batch_images)
-
-        all_batch_outputs = _render_shader_batch(
-            fragment_shader,
-            out_width,
-            out_height,
-            image_batches,
-            float_list,
-            int_list,
-        )
-
-        # Collect outputs into tensors
-        all_outputs = [[] for _ in range(MAX_OUTPUTS)]
-        for batch_outputs in all_batch_outputs:
-            for i, out_img in enumerate(batch_outputs):
-                all_outputs[i].append(torch.from_numpy(out_img))
-
-        output_tensors = [torch.stack(all_outputs[i], dim=0) for i in range(MAX_OUTPUTS)]
-        return io.NodeOutput(
-            *output_tensors,
-            ui=cls._build_ui_output(image_list, output_tensors[0]),
-        )
-
-    @classmethod
-    def _build_ui_output(
-        cls, image_list: list[torch.Tensor], output_batch: torch.Tensor
-    ) -> dict[str, list]:
-        """Build UI output with input and output images for client-side shader execution."""
-        combined_inputs = torch.cat(image_list, dim=0)
-        input_images_ui = ui.ImageSaveHelper.save_images(
-            combined_inputs,
-            filename_prefix="GLSLShader_input",
-            folder_type=io.FolderType.temp,
-            cls=None,
-            compress_level=1,
-        )
-
-        output_images_ui = ui.ImageSaveHelper.save_images(
-            output_batch,
-            filename_prefix="GLSLShader_output",
-            folder_type=io.FolderType.temp,
-            cls=None,
-            compress_level=1,
-        )
-
-        return {"input_images": input_images_ui, "images": output_images_ui}
-
-
-class GLSLExtension(ComfyExtension):
-    @override
-    async def get_node_list(self) -> list[type[io.ComfyNode]]:
-        return [GLSLShader]
-
-
-async def comfy_entrypoint() -> GLSLExtension:
-    return GLSLExtension()

comfy_extras/nodes_images.py

@@ -6,7 +6,6 @@ import folder_paths
 import json
 import os
 import re
-import math
 import torch
 import comfy.utils
 
@@ -589,7 +588,6 @@ class ImageRotate(IO.ComfyNode):
     def define_schema(cls):
         return IO.Schema(
             node_id="ImageRotate",
-            display_name="Image Rotate",
             search_aliases=["turn", "flip orientation"],
             category="image/transform",
             essentials_category="Image Tools",
@@ -683,172 +681,6 @@ class ImageScaleToMaxDimension(IO.ComfyNode):
     upscale = execute    # TODO: remove
 
 
-class SplitImageToTileList(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="SplitImageToTileList",
-            category="image/batch",
-            search_aliases=["split image", "tile image", "slice image"],
-            display_name="Split Image into List of Tiles",
-            description="Splits an image into a batched list of tiles with a specified overlap.",
-            inputs=[
-                IO.Image.Input("image"),
-                IO.Int.Input("tile_width", default=1024, min=64, max=MAX_RESOLUTION),
-                IO.Int.Input("tile_height", default=1024, min=64, max=MAX_RESOLUTION),
-                IO.Int.Input("overlap", default=128, min=0, max=4096),
-            ],
-            outputs=[
-                IO.Image.Output(is_output_list=True),
-            ],
-        )
-
-    @staticmethod
-    def get_grid_coords(width, height, tile_width, tile_height, overlap):
-        coords = []
-        stride_x = max(1, tile_width - overlap)
-        stride_y = max(1, tile_height - overlap)
-
-        y = 0
-        while y < height:
-            x = 0
-            y_end = min(y + tile_height, height)
-            y_start = max(0, y_end - tile_height)
-
-            while x < width:
-                x_end = min(x + tile_width, width)
-                x_start = max(0, x_end - tile_width)
-
-                coords.append((x_start, y_start, x_end, y_end))
-
-                if x_end >= width:
-                    break
-                x += stride_x
-
-            if y_end >= height:
-                break
-            y += stride_y
-
-        return coords
-
-    @classmethod
-    def execute(cls, image, tile_width, tile_height, overlap):
-        b, h, w, c = image.shape
-        coords = cls.get_grid_coords(w, h, tile_width, tile_height, overlap)
-
-        output_list = []
-        for (x_start, y_start, x_end, y_end) in coords:
-            tile = image[:, y_start:y_end, x_start:x_end, :]
-            output_list.append(tile)
-
-        return IO.NodeOutput(output_list)
-
-
-class ImageMergeTileList(IO.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        return IO.Schema(
-            node_id="ImageMergeTileList",
-            display_name="Merge List of Tiles to Image",
-            category="image/batch",
-            search_aliases=["split image", "tile image", "slice image"],
-            is_input_list=True,
-            inputs=[
-                IO.Image.Input("image_list"),
-                IO.Int.Input("final_width", default=1024, min=64, max=32768),
-                IO.Int.Input("final_height", default=1024, min=64, max=32768),
-                IO.Int.Input("overlap", default=128, min=0, max=4096),
-            ],
-            outputs=[
-                IO.Image.Output(is_output_list=False),
-            ],
-        )
-
-    @staticmethod
-    def get_grid_coords(width, height, tile_width, tile_height, overlap):
-        coords = []
-        stride_x = max(1, tile_width - overlap)
-        stride_y = max(1, tile_height - overlap)
-
-        y = 0
-        while y < height:
-            x = 0
-            y_end = min(y + tile_height, height)
-            y_start = max(0, y_end - tile_height)
-
-            while x < width:
-                x_end = min(x + tile_width, width)
-                x_start = max(0, x_end - tile_width)
-
-                coords.append((x_start, y_start, x_end, y_end))
-
-                if x_end >= width:
-                    break
-                x += stride_x
-
-            if y_end >= height:
-                break
-            y += stride_y
-
-        return coords
-
-    @classmethod
-    def execute(cls, image_list, final_width, final_height, overlap):
-        w = final_width[0]
-        h = final_height[0]
-        ovlp = overlap[0]
-        feather_str = 1.0
-
-        first_tile = image_list[0]
-        b, t_h, t_w, c = first_tile.shape
-        device = first_tile.device
-        dtype = first_tile.dtype
-
-        coords = cls.get_grid_coords(w, h, t_w, t_h, ovlp)
-
-        canvas = torch.zeros((b, h, w, c), device=device, dtype=dtype)
-        weights = torch.zeros((b, h, w, 1), device=device, dtype=dtype)
-
-        if ovlp > 0:
-            y_w = torch.sin(math.pi * torch.linspace(0, 1, t_h, device=device, dtype=dtype))
-            x_w = torch.sin(math.pi * torch.linspace(0, 1, t_w, device=device, dtype=dtype))
-            y_w = torch.clamp(y_w, min=1e-5)
-            x_w = torch.clamp(x_w, min=1e-5)
-
-            sine_mask = (y_w.unsqueeze(1) * x_w.unsqueeze(0)).unsqueeze(0).unsqueeze(-1)
-            flat_mask = torch.ones_like(sine_mask)
-
-            weight_mask = torch.lerp(flat_mask, sine_mask, feather_str)
-        else:
-            weight_mask = torch.ones((1, t_h, t_w, 1), device=device, dtype=dtype)
-
-        for i, (x_start, y_start, x_end, y_end) in enumerate(coords):
-            if i >= len(image_list):
-                break
-
-            tile = image_list[i]
-
-            region_h = y_end - y_start
-            region_w = x_end - x_start
-
-            real_h = min(region_h, tile.shape[1])
-            real_w = min(region_w, tile.shape[2])
-
-            y_end_actual = y_start + real_h
-            x_end_actual = x_start + real_w
-
-            tile_crop = tile[:, :real_h, :real_w, :]
-            mask_crop = weight_mask[:, :real_h, :real_w, :]
-
-            canvas[:, y_start:y_end_actual, x_start:x_end_actual, :] += tile_crop * mask_crop
-            weights[:, y_start:y_end_actual, x_start:x_end_actual, :] += mask_crop
-
-        weights[weights == 0] = 1.0
-        merged_image = canvas / weights
-
-        return IO.NodeOutput(merged_image)
-
-
 class ImagesExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -868,8 +700,6 @@ class ImagesExtension(ComfyExtension):
             ImageRotate,
             ImageFlip,
             ImageScaleToMaxDimension,
-            SplitImageToTileList,
-            ImageMergeTileList,
         ]
 
 

comfy_extras/nodes_nag.py

@@ -10,7 +10,7 @@ class NAGuidance(io.ComfyNode):
             node_id="NAGuidance",
             display_name="Normalized Attention Guidance",
             description="Applies Normalized Attention Guidance to models, enabling negative prompts on distilled/schnell models.",
-            category="advanced/guidance",
+            category="",
             is_experimental=True,
             inputs=[
                 io.Model.Input("model", tooltip="The model to apply NAG to."),

comfy_extras/nodes_post_processing.py

@@ -19,7 +19,6 @@ class Blend(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageBlend",
-            display_name="Image Blend",
             category="image/postprocessing",
             inputs=[
                 io.Image.Input("image1"),
@@ -77,7 +76,6 @@ class Blur(io.ComfyNode):
     def define_schema(cls):
         return io.Schema(
             node_id="ImageBlur",
-            display_name="Image Blur",
             category="image/postprocessing",
             essentials_category="Image Tools",
             inputs=[

comfy_extras/nodes_primitive.py

@@ -29,7 +29,6 @@ class StringMultiline(io.ComfyNode):
             node_id="PrimitiveStringMultiline",
             display_name="String (Multiline)",
             category="utils/primitive",
-            essentials_category="Basics",
             inputs=[
                 io.String.Input("value", multiline=True),
             ],

comfy_extras/nodes_textgen.py

@@ -42,7 +42,7 @@ class TextGenerate(io.ComfyNode):
     @classmethod
     def execute(cls, clip, prompt, max_length, sampling_mode, image=None) -> io.NodeOutput:
 
-        tokens = clip.tokenize(prompt, image=image, skip_template=False, min_length=1)
+        tokens = clip.tokenize(prompt, image=image, skip_template=False)
 
         # Get sampling parameters from dynamic combo
         do_sample = sampling_mode.get("sampling_mode") == "on"

comfy_extras/nodes_torch_compile.py

@@ -25,7 +25,7 @@ class TorchCompileModel(io.ComfyNode):
 
     @classmethod
     def execute(cls, model, backend) -> io.NodeOutput:
-        m = model.clone(disable_dynamic=True)
+        m = model.clone()
         set_torch_compile_wrapper(model=m, backend=backend, options={"guard_filter_fn": skip_torch_compile_dict})
         return io.NodeOutput(m)
 

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.15.0"
+__version__ = "0.14.1"

cuda_malloc.py

@@ -1,8 +1,10 @@
 import os
 import importlib.util
-from comfy.cli_args import args, PerformanceFeature
+from comfy.cli_args import args, PerformanceFeature, enables_dynamic_vram
 import subprocess
 
+import comfy_aimdo.control
+
 #Can't use pytorch to get the GPU names because the cuda malloc has to be set before the first import.
 def get_gpu_names():
     if os.name == 'nt':
@@ -85,6 +87,10 @@ if not args.cuda_malloc:
     except:
         pass
 
+if enables_dynamic_vram() and comfy_aimdo.control.init():
+    args.cuda_malloc = False
+    os.environ['PYTORCH_CUDA_ALLOC_CONF'] = ""
+
 if args.disable_cuda_malloc:
     args.cuda_malloc = False
 

execution.py

@@ -9,6 +9,7 @@ import traceback
 from enum import Enum
 from typing import List, Literal, NamedTuple, Optional, Union
 import asyncio
+from contextlib import nullcontext
 
 import torch
 
@@ -520,14 +521,19 @@ async def execute(server, dynprompt, caches, current_item, extra_data, executed,
                 # TODO - How to handle this with async functions without contextvars (which requires Python 3.12)?
                 GraphBuilder.set_default_prefix(unique_id, call_index, 0)
 
-            try:
-                output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
-            finally:
-                if comfy.memory_management.aimdo_enabled:
-                    if args.verbose == "DEBUG":
-                        comfy_aimdo.control.analyze()
-                    comfy.model_management.reset_cast_buffers()
-                    comfy_aimdo.model_vbar.vbars_reset_watermark_limits()
+            #Do comfy_aimdo mempool chunking here on the per-node level. Multi-model workflows
+            #will cause all sorts of incompatible memory shapes to fragment the pytorch alloc
+            #that we just want to cull out each model run.
+            allocator = comfy.memory_management.aimdo_allocator
+            with nullcontext() if allocator is None else torch.cuda.use_mem_pool(torch.cuda.MemPool(allocator.allocator())):
+                try:
+                    output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
+                finally:
+                    if allocator is not None:
+                        if args.verbose == "DEBUG":
+                            comfy_aimdo.model_vbar.vbars_analyze()
+                        comfy.model_management.reset_cast_buffers()
+                        comfy_aimdo.model_vbar.vbars_reset_watermark_limits()
 
             if has_pending_tasks:
                 pending_async_nodes[unique_id] = output_data

main.py

@@ -173,10 +173,6 @@ import gc
 if 'torch' in sys.modules:
     logging.warning("WARNING: Potential Error in code: Torch already imported, torch should never be imported before this point.")
 
-import comfy_aimdo.control
-
-if enables_dynamic_vram():
-    comfy_aimdo.control.init()
 
 import comfy.utils
 
@@ -192,9 +188,13 @@ import hook_breaker_ac10a0
 import comfy.memory_management
 import comfy.model_patcher
 
+import comfy_aimdo.control
+import comfy_aimdo.torch
+
 if enables_dynamic_vram():
     if comfy.model_management.torch_version_numeric < (2, 8):
         logging.warning("Unsupported Pytorch detected. DynamicVRAM support requires Pytorch version 2.8 or later. Falling back to legacy ModelPatcher. VRAM estimates may be unreliable especially on Windows")
+        comfy.memory_management.aimdo_allocator = None
     elif comfy_aimdo.control.init_device(comfy.model_management.get_torch_device().index):
         if args.verbose == 'DEBUG':
             comfy_aimdo.control.set_log_debug()
@@ -208,10 +208,11 @@ if enables_dynamic_vram():
             comfy_aimdo.control.set_log_info()
 
         comfy.model_patcher.CoreModelPatcher = comfy.model_patcher.ModelPatcherDynamic
-        comfy.memory_management.aimdo_enabled = True
+        comfy.memory_management.aimdo_allocator = comfy_aimdo.torch.get_torch_allocator()
         logging.info("DynamicVRAM support detected and enabled")
     else:
         logging.warning("No working comfy-aimdo install detected. DynamicVRAM support disabled. Falling back to legacy ModelPatcher. VRAM estimates may be unreliable especially on Windows")
+        comfy.memory_management.aimdo_allocator = None
 
 
 def cuda_malloc_warning():

nodes.py

@@ -70,6 +70,7 @@ class CLIPTextEncode(ComfyNodeABC):
     FUNCTION = "encode"
 
     CATEGORY = "conditioning"
+    ESSENTIALS_CATEGORY = "Basics"
     DESCRIPTION = "Encodes a text prompt using a CLIP model into an embedding that can be used to guide the diffusion model towards generating specific images."
     SEARCH_ALIASES = ["text", "prompt", "text prompt", "positive prompt", "negative prompt", "encode text", "text encoder", "encode prompt"]
 
@@ -2441,7 +2442,6 @@ async def init_builtin_extra_nodes():
         "nodes_wanmove.py",
         "nodes_image_compare.py",
         "nodes_zimage.py",
-        "nodes_glsl.py",
         "nodes_lora_debug.py",
         "nodes_textgen.py",
         "nodes_color.py",

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.15.0"
+version = "0.14.1"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.10"

requirements.txt

@@ -1,6 +1,6 @@
-comfyui-frontend-package==1.39.19
-comfyui-workflow-templates==0.9.3
-comfyui-embedded-docs==0.4.3
+comfyui-frontend-package==1.39.14
+comfyui-workflow-templates==0.8.43
+comfyui-embedded-docs==0.4.1
 torch
 torchsde
 torchvision
@@ -22,7 +22,7 @@ alembic
 SQLAlchemy
 av>=14.2.0
 comfy-kitchen>=0.2.7
-comfy-aimdo>=0.2.2
+comfy-aimdo>=0.1.8
 requests
 
 #non essential dependencies:
@@ -30,6 +30,3 @@ kornia>=0.7.1
 spandrel
 pydantic~=2.0
 pydantic-settings~=2.0
-PyOpenGL
-PyOpenGL-accelerate
-glfw