ComfyUI v0.18.2

There was an issue where the resample split was too early and dropped one
of the rolling convolutions a frame early. This is most noticable as a
lighting/color change between pixel frames 5->6 (latent 2->3), or as a
lighting change between the first and last frame in an FLF wan flow.

comfy/ldm/wan/vae.py

@@ -376,11 +376,16 @@ class Decoder3d(nn.Module):
             return
 
         layer = self.upsamples[layer_idx]
-        if isinstance(layer, Resample) and layer.mode == 'upsample3d' and x.shape[2] > 1:
-            for frame_idx in range(x.shape[2]):
+        if feat_cache is not None:
+            x = layer(x, feat_cache, feat_idx)
+        else:
+            x = layer(x)
+
+        if isinstance(layer, Resample) and layer.mode == 'upsample3d' and x.shape[2] > 2:
+            for frame_idx in range(0, x.shape[2], 2):
                 self.run_up(
-                    layer_idx,
-                    [x[:, :, frame_idx:frame_idx + 1, :, :]],
+                    layer_idx + 1,
+                    [x[:, :, frame_idx:frame_idx + 2, :, :]],
                     feat_cache,
                     feat_idx.copy(),
                     out_chunks,
@@ -388,11 +393,6 @@ class Decoder3d(nn.Module):
             del x
             return
 
-        if feat_cache is not None:
-            x = layer(x, feat_cache, feat_idx)
-        else:
-            x = layer(x)
-
         next_x_ref = [x]
         del x
         self.run_up(layer_idx + 1, next_x_ref, feat_cache, feat_idx, out_chunks)

comfy/sample.py

@@ -8,12 +8,12 @@ import comfy.nested_tensor
 
 def prepare_noise_inner(latent_image, generator, noise_inds=None):
     if noise_inds is None:
-        return torch.randn(latent_image.size(), dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")
+        return torch.randn(latent_image.size(), dtype=torch.float32, layout=latent_image.layout, generator=generator, device="cpu").to(dtype=latent_image.dtype)
 
     unique_inds, inverse = np.unique(noise_inds, return_inverse=True)
     noises = []
     for i in range(unique_inds[-1]+1):
-        noise = torch.randn([1] + list(latent_image.size())[1:], dtype=latent_image.dtype, layout=latent_image.layout, generator=generator, device="cpu")
+        noise = torch.randn([1] + list(latent_image.size())[1:], dtype=torch.float32, layout=latent_image.layout, generator=generator, device="cpu").to(dtype=latent_image.dtype)
         if i in unique_inds:
             noises.append(noise)
     noises = [noises[i] for i in inverse]

comfy/samplers.py

@@ -985,8 +985,8 @@ class CFGGuider:
         self.inner_model, self.conds, self.loaded_models = comfy.sampler_helpers.prepare_sampling(self.model_patcher, noise.shape, self.conds, self.model_options)
         device = self.model_patcher.load_device
 
-        noise = noise.to(device)
-        latent_image = latent_image.to(device)
+        noise = noise.to(device=device, dtype=torch.float32)
+        latent_image = latent_image.to(device=device, dtype=torch.float32)
         sigmas = sigmas.to(device)
         cast_to_load_options(self.model_options, device=device, dtype=self.model_patcher.model_dtype())
 
@@ -1028,6 +1028,7 @@ class CFGGuider:
                 denoise_mask, _ = comfy.utils.pack_latents(denoise_masks)
             else:
                 denoise_mask = denoise_masks[0]
+            denoise_mask = denoise_mask.float()
 
         self.conds = {}
         for k in self.original_conds:

comfy_api_nodes/apis/grok.py

@@ -29,13 +29,21 @@ class ImageEditRequest(BaseModel):
 class VideoGenerationRequest(BaseModel):
     model: str = Field(...)
     prompt: str = Field(...)
-    image: InputUrlObject | None = Field(...)
+    image: InputUrlObject | None = Field(None)
+    reference_images: list[InputUrlObject] | None = Field(None)
     duration: int = Field(...)
     aspect_ratio: str | None = Field(...)
     resolution: str = Field(...)
     seed: int = Field(...)
 
 
+class VideoExtensionRequest(BaseModel):
+    prompt: str = Field(...)
+    video: InputUrlObject = Field(...)
+    duration: int = Field(default=6)
+    model: str | None = Field(default=None)
+
+
 class VideoEditRequest(BaseModel):
     model: str = Field(...)
     prompt: str = Field(...)

comfy_api_nodes/nodes_grok.py

@@ -8,6 +8,7 @@ from comfy_api_nodes.apis.grok import (
     ImageGenerationResponse,
     InputUrlObject,
     VideoEditRequest,
+    VideoExtensionRequest,
     VideoGenerationRequest,
     VideoGenerationResponse,
     VideoStatusResponse,
@@ -21,6 +22,7 @@ from comfy_api_nodes.util import (
     poll_op,
     sync_op,
     tensor_to_base64_string,
+    upload_images_to_comfyapi,
     upload_video_to_comfyapi,
     validate_string,
     validate_video_duration,
@@ -33,6 +35,13 @@ def _extract_grok_price(response) -> float | None:
     return None
 
 
+def _extract_grok_video_price(response) -> float | None:
+    price = _extract_grok_price(response)
+    if price is not None:
+        return price * 1.43
+    return None
+
+
 class GrokImageNode(IO.ComfyNode):
 
     @classmethod
@@ -354,6 +363,8 @@ class GrokVideoNode(IO.ComfyNode):
         seed: int,
         image: Input.Image | None = None,
     ) -> IO.NodeOutput:
+        if model == "grok-imagine-video-beta":
+            model = "grok-imagine-video"
         image_url = None
         if image is not None:
             if get_number_of_images(image) != 1:
@@ -462,6 +473,244 @@ class GrokVideoEditNode(IO.ComfyNode):
         return IO.NodeOutput(await download_url_to_video_output(response.video.url))
 
 
+class GrokVideoReferenceNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="GrokVideoReferenceNode",
+            display_name="Grok Reference-to-Video",
+            category="api node/video/Grok",
+            description="Generate video guided by reference images as style and content references.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="Text description of the desired video.",
+                ),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "grok-imagine-video",
+                            [
+                                IO.Autogrow.Input(
+                                    "reference_images",
+                                    template=IO.Autogrow.TemplatePrefix(
+                                        IO.Image.Input("image"),
+                                        prefix="reference_",
+                                        min=1,
+                                        max=7,
+                                    ),
+                                    tooltip="Up to 7 reference images to guide the video generation.",
+                                ),
+                                IO.Combo.Input(
+                                    "resolution",
+                                    options=["480p", "720p"],
+                                    tooltip="The resolution of the output video.",
+                                ),
+                                IO.Combo.Input(
+                                    "aspect_ratio",
+                                    options=["16:9", "4:3", "3:2", "1:1", "2:3", "3:4", "9:16"],
+                                    tooltip="The aspect ratio of the output video.",
+                                ),
+                                IO.Int.Input(
+                                    "duration",
+                                    default=6,
+                                    min=2,
+                                    max=10,
+                                    step=1,
+                                    tooltip="The duration of the output video in seconds.",
+                                    display_mode=IO.NumberDisplay.slider,
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="The model to use for video generation.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    step=1,
+                    display_mode=IO.NumberDisplay.number,
+                    control_after_generate=True,
+                    tooltip="Seed to determine if node should re-run; "
+                    "actual results are nondeterministic 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=["model.duration", "model.resolution"],
+                    input_groups=["model.reference_images"],
+                ),
+                expr="""
+                (
+                  $res := $lookup(widgets, "model.resolution");
+                  $dur := $lookup(widgets, "model.duration");
+                  $refs := inputGroups["model.reference_images"];
+                  $rate := $res = "720p" ? 0.07 : 0.05;
+                  $price := ($rate * $dur + 0.002 * $refs) * 1.43;
+                  {"type":"usd","usd": $price}
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        model: dict,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, strip_whitespace=True, min_length=1)
+        ref_image_urls = await upload_images_to_comfyapi(
+            cls,
+            list(model["reference_images"].values()),
+            mime_type="image/png",
+            wait_label="Uploading base images",
+            max_images=7,
+        )
+        initial_response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/xai/v1/videos/generations", method="POST"),
+            data=VideoGenerationRequest(
+                model=model["model"],
+                reference_images=[InputUrlObject(url=i) for i in ref_image_urls],
+                prompt=prompt,
+                resolution=model["resolution"],
+                duration=model["duration"],
+                aspect_ratio=model["aspect_ratio"],
+                seed=seed,
+            ),
+            response_model=VideoGenerationResponse,
+        )
+        response = await poll_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/xai/v1/videos/{initial_response.request_id}"),
+            status_extractor=lambda r: r.status if r.status is not None else "complete",
+            response_model=VideoStatusResponse,
+            price_extractor=_extract_grok_video_price,
+        )
+        return IO.NodeOutput(await download_url_to_video_output(response.video.url))
+
+
+class GrokVideoExtendNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="GrokVideoExtendNode",
+            display_name="Grok Video Extend",
+            category="api node/video/Grok",
+            description="Extend an existing video with a seamless continuation based on a text prompt.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    tooltip="Text description of what should happen next in the video.",
+                ),
+                IO.Video.Input("video", tooltip="Source video to extend. MP4 format, 2-15 seconds."),
+                IO.DynamicCombo.Input(
+                    "model",
+                    options=[
+                        IO.DynamicCombo.Option(
+                            "grok-imagine-video",
+                            [
+                                IO.Int.Input(
+                                    "duration",
+                                    default=8,
+                                    min=2,
+                                    max=10,
+                                    step=1,
+                                    tooltip="Length of the extension in seconds.",
+                                    display_mode=IO.NumberDisplay.slider,
+                                ),
+                            ],
+                        ),
+                    ],
+                    tooltip="The model to use for video extension.",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=2147483647,
+                    step=1,
+                    display_mode=IO.NumberDisplay.number,
+                    control_after_generate=True,
+                    tooltip="Seed to determine if node should re-run; "
+                    "actual results are nondeterministic 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=["model.duration"]),
+                expr="""
+                (
+                  $dur := $lookup(widgets, "model.duration");
+                  {
+                    "type": "range_usd",
+                    "min_usd": (0.02 + 0.05 * $dur) * 1.43,
+                    "max_usd": (0.15 + 0.05 * $dur) * 1.43
+                  }
+                )
+                """,
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        video: Input.Video,
+        model: dict,
+        seed: int,
+    ) -> IO.NodeOutput:
+        validate_string(prompt, strip_whitespace=True, min_length=1)
+        validate_video_duration(video, min_duration=2, max_duration=15)
+        video_size = get_fs_object_size(video.get_stream_source())
+        if video_size > 50 * 1024 * 1024:
+            raise ValueError(f"Video size ({video_size / 1024 / 1024:.1f}MB) exceeds 50MB limit.")
+        initial_response = await sync_op(
+            cls,
+            ApiEndpoint(path="/proxy/xai/v1/videos/extensions", method="POST"),
+            data=VideoExtensionRequest(
+                prompt=prompt,
+                video=InputUrlObject(url=await upload_video_to_comfyapi(cls, video)),
+                duration=model["duration"],
+            ),
+            response_model=VideoGenerationResponse,
+        )
+        response = await poll_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/xai/v1/videos/{initial_response.request_id}"),
+            status_extractor=lambda r: r.status if r.status is not None else "complete",
+            response_model=VideoStatusResponse,
+            price_extractor=_extract_grok_video_price,
+        )
+        return IO.NodeOutput(await download_url_to_video_output(response.video.url))
+
+
 class GrokExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
@@ -469,7 +718,9 @@ class GrokExtension(ComfyExtension):
             GrokImageNode,
             GrokImageEditNode,
             GrokVideoNode,
+            GrokVideoReferenceNode,
             GrokVideoEditNode,
+            GrokVideoExtendNode,
         ]
 
 

comfy_extras/nodes_logic.py

@@ -8,36 +8,6 @@ from comfy_api.latest import _io
 MISSING = object()
 
 
-class BranchNode(io.ComfyNode):
-    @classmethod
-    def define_schema(cls):
-        mtemplate = io.MatchType.Template("switch")
-        minput = io.MatchType.Input("branch", template=mtemplate, lazy=True, optional=True)
-        template = _io.Autogrow.TemplatePrefix(input=minput, prefix="branch", min=1, max=10)
-        return io.Schema(
-            node_id="BranchNode",
-            display_name="Branch",
-            category="logic",
-            is_experimental=True,
-            inputs=[
-                io.Int.Input("branch"),
-                _io.Autogrow.Input("autogrow", template=template)
-            ],
-            outputs=[
-                io.MatchType.Output(template=mtemplate, display_name="output"),
-            ],
-        )
-
-    @classmethod
-    def check_lazy_status(cls, branch, autogrow):
-        print('lazy', branch)
-        return ['autogrow.' + list(autogrow.keys())[branch]]
-
-    @classmethod
-    def execute(cls, branch, autogrow) -> io.NodeOutput:
-        print(branch)
-        return list(autogrow.values())[branch],
-
 class SwitchNode(io.ComfyNode):
     @classmethod
     def define_schema(cls):
@@ -298,7 +268,6 @@ class LogicExtension(ComfyExtension):
             # AutogrowPrefixTestNode,
             # ComboOutputTestNode,
             # InvertBooleanNode,
-            BranchNode,
         ]
 
 async def comfy_entrypoint() -> LogicExtension:

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.18.0"
+__version__ = "0.18.2"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.18.0"
+version = "0.18.2"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.10"

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.41.21
-comfyui-workflow-templates==0.9.26
+comfyui-workflow-templates==0.9.36
 comfyui-embedded-docs==0.4.3
 torch
 torchsde