Merge branch 'master' into master

modules/api/api.py

@@ -1,6 +1,8 @@
+import base64
+import io
 import time
 import uvicorn
-from gradio.processing_utils import encode_pil_to_base64, decode_base64_to_file, decode_base64_to_image
+from gradio.processing_utils import decode_base64_to_file, decode_base64_to_image
 from fastapi import APIRouter, Depends, HTTPException
 import modules.shared as shared
 from modules import devices
@@ -29,6 +31,12 @@ def setUpscalers(req: dict):
     return reqDict
 
 
+def encode_pil_to_base64(image):
+    buffer = io.BytesIO()
+    image.save(buffer, format="png")
+    return base64.b64encode(buffer.getvalue())
+
+
 class Api:
     def __init__(self, app, queue_lock):
         self.router = APIRouter()
@@ -40,6 +48,7 @@ class Api:
         self.app.add_api_route("/sdapi/v1/extra-batch-images", self.extras_batch_images_api, methods=["POST"], response_model=ExtrasBatchImagesResponse)
         self.app.add_api_route("/sdapi/v1/png-info", self.pnginfoapi, methods=["POST"], response_model=PNGInfoResponse)
         self.app.add_api_route("/sdapi/v1/progress", self.progressapi, methods=["GET"], response_model=ProgressResponse)
+        self.app.add_api_route("/sdapi/v1/interrupt", self.interruptapi, methods=["POST"])
 
     def text2imgapi(self, txt2imgreq: StableDiffusionTxt2ImgProcessingAPI):
         sampler_index = sampler_to_index(txt2imgreq.sampler_index)
@@ -176,6 +185,11 @@ class Api:
 
         return ProgressResponse(progress=progress, eta_relative=eta_relative, state=shared.state.dict(), current_image=current_image)
 
+    def interruptapi(self):
+        shared.state.interrupt()
+
+        return {}
+
     def launch(self, server_name, port):
         self.app.include_router(self.router)
         uvicorn.run(self.app, host=server_name, port=port)

modules/extras.py

@@ -141,7 +141,7 @@ def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_
                             upscaling_resize_w, upscaling_resize_h, upscaling_crop)
             cache_key = LruCache.Key(image_hash=hash(np.array(image.getdata()).tobytes()),
                                      info_hash=hash(info),
-                                     args_hash=hash(upscale_args))
+                                     args_hash=hash((upscale_args, upscale_first)))
             cached_entry = cached_images.get(cache_key)
             if cached_entry is None:
                 res = upscale(image, *upscale_args)

modules/images.py

@@ -510,8 +510,9 @@ def save_image(image, path, basename, seed=None, prompt=None, extension='png', i
 
     if extension.lower() == '.png':
         pnginfo_data = PngImagePlugin.PngInfo()
-        for k, v in params.pnginfo.items():
-            pnginfo_data.add_text(k, str(v))
+        if opts.enable_pnginfo:
+            for k, v in params.pnginfo.items():
+                pnginfo_data.add_text(k, str(v))
 
         image.save(fullfn, quality=opts.jpeg_quality, pnginfo=pnginfo_data)
 

modules/img2img.py

@@ -137,6 +137,8 @@ def img2img(mode: int, prompt: str, negative_prompt: str, prompt_style: str, pro
         if processed is None:
             processed = process_images(p)
 
+    p.close()
+
     shared.total_tqdm.clear()
 
     generation_info_js = processed.js()

modules/interrogate.py

@@ -56,9 +56,9 @@ class InterrogateModels:
         import clip
 
         if self.running_on_cpu:
-            model, preprocess = clip.load(clip_model_name, device="cpu")
+            model, preprocess = clip.load(clip_model_name, device="cpu", download_root=shared.cmd_opts.clip_models_path)
         else:
-            model, preprocess = clip.load(clip_model_name)
+            model, preprocess = clip.load(clip_model_name, download_root=shared.cmd_opts.clip_models_path)
 
         model.eval()
         model = model.to(devices.device_interrogate)

modules/processing.py

@@ -202,6 +202,10 @@ class StableDiffusionProcessing():
     def sample(self, conditioning, unconditional_conditioning, seeds, subseeds, subseed_strength):
         raise NotImplementedError()
 
+    def close(self):
+        self.sd_model = None
+        self.sampler = None
+
 
 class Processed:
     def __init__(self, p: StableDiffusionProcessing, images_list, seed=-1, info="", subseed=None, all_prompts=None, all_seeds=None, all_subseeds=None, index_of_first_image=0, infotexts=None):
@@ -597,9 +601,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
     if p.scripts is not None:
         p.scripts.postprocess(p, res)
 
-    p.sd_model = None
-    p.sampler = None
-
     return res
 
 

modules/script_callbacks.py

@@ -26,6 +26,24 @@ class ImageSaveParams:
         """dictionary with parameters for image's PNG info data; infotext will have the key 'parameters'"""
 
 
+class CFGDenoiserParams:
+    def __init__(self, x, image_cond, sigma, sampling_step, total_sampling_steps):
+        self.x = x
+        """Latent image representation in the process of being denoised"""
+        
+        self.image_cond = image_cond
+        """Conditioning image"""
+        
+        self.sigma = sigma
+        """Current sigma noise step value"""
+        
+        self.sampling_step = sampling_step
+        """Current Sampling step number"""
+        
+        self.total_sampling_steps = total_sampling_steps
+        """Total number of sampling steps planned"""
+
+
 ScriptCallback = namedtuple("ScriptCallback", ["script", "callback"])
 callbacks_app_started = []
 callbacks_model_loaded = []
@@ -33,6 +51,7 @@ callbacks_ui_tabs = []
 callbacks_ui_settings = []
 callbacks_before_image_saved = []
 callbacks_image_saved = []
+callbacks_cfg_denoiser = []
 
 
 def clear_callbacks():
@@ -41,7 +60,7 @@ def clear_callbacks():
     callbacks_ui_settings.clear()
     callbacks_before_image_saved.clear()
     callbacks_image_saved.clear()
-
+    callbacks_cfg_denoiser.clear()
 
 def app_started_callback(demo: Blocks, app: FastAPI):
     for c in callbacks_app_started:
@@ -95,6 +114,14 @@ def image_saved_callback(params: ImageSaveParams):
             report_exception(c, 'image_saved_callback')
 
 
+def cfg_denoiser_callback(params: CFGDenoiserParams):
+    for c in callbacks_cfg_denoiser:
+        try:
+            c.callback(params)
+        except Exception:
+            report_exception(c, 'cfg_denoiser_callback')
+
+
 def add_callback(callbacks, fun):
     stack = [x for x in inspect.stack() if x.filename != __file__]
     filename = stack[0].filename if len(stack) > 0 else 'unknown file'
@@ -147,3 +174,12 @@ def on_image_saved(callback):
         - params: ImageSaveParams - parameters the image was saved with. Changing fields in this object does nothing.
     """
     add_callback(callbacks_image_saved, callback)
+
+
+def on_cfg_denoiser(callback):
+    """register a function to be called in the kdiffussion cfg_denoiser method after building the inner model inputs.
+    The callback is called with one argument:
+        - params: CFGDenoiserParams - parameters to be passed to the inner model and sampling state details.
+    """
+    add_callback(callbacks_cfg_denoiser, callback)
+

modules/sd_samplers.py

@@ -12,6 +12,7 @@ from modules import prompt_parser, devices, processing, images
 
 from modules.shared import opts, cmd_opts, state
 import modules.shared as shared
+from modules.script_callbacks import CFGDenoiserParams, cfg_denoiser_callback
 
 
 SamplerData = namedtuple('SamplerData', ['name', 'constructor', 'aliases', 'options'])
@@ -280,6 +281,12 @@ class CFGDenoiser(torch.nn.Module):
         image_cond_in = torch.cat([torch.stack([image_cond[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [image_cond])
         sigma_in = torch.cat([torch.stack([sigma[i] for _ in range(n)]) for i, n in enumerate(repeats)] + [sigma])
 
+        denoiser_params = CFGDenoiserParams(x_in, image_cond_in, sigma_in, state.sampling_step, state.sampling_steps)
+        cfg_denoiser_callback(denoiser_params)
+        x_in = denoiser_params.x
+        image_cond_in = denoiser_params.image_cond
+        sigma_in = denoiser_params.sigma
+
         if tensor.shape[1] == uncond.shape[1]:
             cond_in = torch.cat([tensor, uncond])
 

modules/shared.py

@@ -51,6 +51,7 @@ parser.add_argument("--realesrgan-models-path", type=str, help="Path to director
 parser.add_argument("--scunet-models-path", type=str, help="Path to directory with ScuNET model file(s).", default=os.path.join(models_path, 'ScuNET'))
 parser.add_argument("--swinir-models-path", type=str, help="Path to directory with SwinIR model file(s).", default=os.path.join(models_path, 'SwinIR'))
 parser.add_argument("--ldsr-models-path", type=str, help="Path to directory with LDSR model file(s).", default=os.path.join(models_path, 'LDSR'))
+parser.add_argument("--clip-models-path", type=str, help="Path to directory with CLIP model file(s).", default=None)
 parser.add_argument("--xformers", action='store_true', help="enable xformers for cross attention layers")
 parser.add_argument("--force-enable-xformers", action='store_true', help="enable xformers for cross attention layers regardless of whether the checking code thinks you can run it; do not make bug reports if this fails to work")
 parser.add_argument("--deepdanbooru", action='store_true', help="enable deepdanbooru interrogator")
@@ -288,11 +289,12 @@ options_templates.update(options_section(('system', "System"), {
 }))
 
 options_templates.update(options_section(('training', "Training"), {
-    "unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training hypernetwork. Saves VRAM."),
+    "unload_models_when_training": OptionInfo(False, "Move VAE and CLIP to RAM when training if possible. Saves VRAM."),
     "dataset_filename_word_regex": OptionInfo("", "Filename word regex"),
     "dataset_filename_join_string": OptionInfo(" ", "Filename join string"),
     "training_image_repeats_per_epoch": OptionInfo(1, "Number of repeats for a single input image per epoch; used only for displaying epoch number", gr.Number, {"precision": 0}),
     "training_write_csv_every": OptionInfo(500, "Save an csv containing the loss to log directory every N steps, 0 to disable"),
+    "training_xattention_optimizations": OptionInfo(False, "Use cross attention optimizations while training"),
 }))
 
 options_templates.update(options_section(('sd', "Stable Diffusion"), {

modules/textual_inversion/textual_inversion.py

@@ -235,6 +235,7 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
     filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
 
     log_directory = os.path.join(log_directory, datetime.datetime.now().strftime("%Y-%m-%d"), embedding_name)
+    unload = shared.opts.unload_models_when_training
 
     if save_embedding_every > 0:
         embedding_dir = os.path.join(log_directory, "embeddings")
@@ -272,6 +273,8 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
     shared.state.textinfo = f"Preparing dataset from {html.escape(data_root)}..."
     with torch.autocast("cuda"):
         ds = modules.textual_inversion.dataset.PersonalizedBase(data_root=data_root, width=training_width, height=training_height, repeats=shared.opts.training_image_repeats_per_epoch, placeholder_token=embedding_name, shuffle_tags=shuffle_tags, model=shared.sd_model, device=devices.device, template_file=template_file, batch_size=batch_size)
+    if unload:
+        shared.sd_model.first_stage_model.to(devices.cpu)
 
     embedding.vec.requires_grad = True
     optimizer = torch.optim.AdamW([embedding.vec], lr=scheduler.learn_rate)
@@ -328,6 +331,9 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
         if images_dir is not None and steps_done % create_image_every == 0:
             forced_filename = f'{embedding_name}-{steps_done}'
             last_saved_image = os.path.join(images_dir, forced_filename)
+
+            shared.sd_model.first_stage_model.to(devices.device)
+
             p = processing.StableDiffusionProcessingTxt2Img(
                 sd_model=shared.sd_model,
                 do_not_save_grid=True,
@@ -355,6 +361,9 @@ def train_embedding(embedding_name, learn_rate, batch_size, data_root, log_direc
             processed = processing.process_images(p)
             image = processed.images[0]
 
+            if unload:
+                shared.sd_model.first_stage_model.to(devices.cpu)
+
             shared.state.current_image = image
 
             if save_image_with_stored_embedding and os.path.exists(last_saved_file) and embedding_yet_to_be_embedded:
@@ -400,6 +409,7 @@ Last saved image: {html.escape(last_saved_image)}<br/>
 
     filename = os.path.join(shared.cmd_opts.embeddings_dir, f'{embedding_name}.pt')
     save_embedding(embedding, checkpoint, embedding_name, filename, remove_cached_checksum=True)
+    shared.sd_model.first_stage_model.to(devices.device)
 
     return embedding, filename
 

modules/textual_inversion/ui.py

@@ -25,8 +25,10 @@ def train_embedding(*args):
 
     assert not shared.cmd_opts.lowvram, 'Training models with lowvram not possible'
 
+    apply_optimizations = shared.opts.training_xattention_optimizations
     try:
-        sd_hijack.undo_optimizations()
+        if not apply_optimizations:
+            sd_hijack.undo_optimizations()
 
         embedding, filename = modules.textual_inversion.textual_inversion.train_embedding(*args)
 
@@ -38,5 +40,6 @@ Embedding saved to {html.escape(filename)}
     except Exception:
         raise
     finally:
-        sd_hijack.apply_optimizations()
+        if not apply_optimizations:
+            sd_hijack.apply_optimizations()
 

modules/txt2img.py

@@ -47,6 +47,8 @@ def txt2img(prompt: str, negative_prompt: str, prompt_style: str, prompt_style2:
     if processed is None:
         processed = process_images(p)
 
+    p.close()
+
     shared.total_tqdm.clear()
 
     generation_info_js = processed.js()

modules/ui.py

@@ -671,6 +671,8 @@ def create_ui(wrap_gradio_gpu_call):
     import modules.img2img
     import modules.txt2img
 
+    reload_javascript()
+
     parameters_copypaste.reset()
 
     with gr.Blocks(analytics_enabled=False) as txt2img_interface:
@@ -1573,8 +1575,7 @@ def create_ui(wrap_gradio_gpu_call):
         reload_script_bodies.click(
             fn=reload_scripts,
             inputs=[],
-            outputs=[],
-            _js='function(){}'
+            outputs=[]
         )
 
         def request_restart():
@@ -1586,7 +1587,7 @@ def create_ui(wrap_gradio_gpu_call):
             fn=request_restart,
             inputs=[],
             outputs=[],
-            _js='function(){restart_reload()}'
+            _js='restart_reload'
         )
 
         if column is not None:
@@ -1785,4 +1786,3 @@ def load_javascript(raw_response):
 
 
 reload_javascript = partial(load_javascript, gradio.routes.templates.TemplateResponse)
-reload_javascript()