fix: PEP 508 compliance for StreamDiffusion dependency with extras

Add Github Action for Publishing to Comfy Registry

.github/workflows/publish.yml

@@ -0,0 +1,21 @@
+name: Publish to Comfy registry
+on:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+    paths:
+      - "pyproject.toml"
+
+jobs:
+  publish-node:
+    name: Publish Custom Node to registry
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check out code
+        uses: actions/checkout@v4
+      - name: Publish Custom Node
+        uses: Comfy-Org/publish-node-action@main
+        with:
+          ## Add your own personal access token to your Github Repository secrets and reference it here.
+          personal_access_token: ${{ secrets.REGISTRY_ACCESS_TOKEN }}

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "StreamDiffusion"]
+	path = StreamDiffusion
+	url = https://github.com/cumulo-autumn/StreamDiffusion

README.md

@@ -6,7 +6,7 @@ This repository is a custom node in ComfyUI.
 
 ### Workflow 1
 
-This is a program that allows you to use Huggingface Diffusers module with ComfyUI. Additionally, Stream Diffusion is also available.
+This is a program that allows you to use Hugging Face Diffusers module with ComfyUI. Additionally, Stream Diffusion is also available.
 
 ![Workflow](img/workflow.png)
 
@@ -33,10 +33,9 @@ The execution looks like this:
 Run the following command inside ComfyUI/custom_nodes.
 
 ```cmd
-git clone https://github.com/Limitex/ComfyUI-Diffusers.git
+git clone --recurse-submodules https://github.com/Limitex/ComfyUI-Diffusers.git
 cd ComfyUI-Diffusers
 pip install -r requirements.txt
-git clone https://github.com/cumulo-autumn/StreamDiffusion.git
 python -m streamdiffusion.tools.install-tensorrt
 ```
 ## Recommended Custom Nodes
@@ -68,3 +67,7 @@ https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite.git
 ### StreamDiffusion Warmup (StreamDiffusionWarmup)
 
 ### StreamDiffusion Fast Sampler (StreamDiffusionFastSampler)
+
+## Reference 
+
+https://github.com/cumulo-autumn/StreamDiffusion

nodes.py

@@ -1,8 +1,6 @@
 import copy
 import os
-import gc
 import torch
-import traceback
 from safetensors.torch import load_file
 from .utils import SCHEDULERS, token_auto_concat_embeds, vae_pt_to_vae_diffuser, convert_images_to_tensors, convert_tensors_to_images, resize_images
 from comfy.model_management import get_torch_device
@@ -10,7 +8,7 @@ import folder_paths
 from streamdiffusion import StreamDiffusion
 from streamdiffusion.image_utils import postprocess_image
 from diffusers import StableDiffusionPipeline, AutoencoderKL, AutoencoderTiny
-from pathlib import Path
+
 
 class DiffusersPipelineLoader:
     def __init__(self):
@@ -120,14 +118,13 @@ class DiffusersModelMakeup:
     CATEGORY = "Diffusers"
 
     def makeup_pipeline(self, pipeline, scheduler, autoencoder):
-        pipe = pipeline[0]
-        ckpt_cache_path = pipeline[1]
-        pipe.vae = autoencoder
-        pipe.scheduler = scheduler
-        pipe.safety_checker = None if pipe.safety_checker is None else lambda images, **kwargs: (images, [False])
-        pipe.enable_attention_slicing()
-        pipe = pipe.to(self.torch_device)
-        return ((pipe, ckpt_cache_path),)
+        pipeline = pipeline[0]
+        pipeline.vae = autoencoder
+        pipeline.scheduler = scheduler
+        pipeline.safety_checker = None if pipeline.safety_checker is None else lambda images, **kwargs: (images, [False])
+        pipeline.enable_attention_slicing()
+        pipeline = pipeline.to(self.torch_device)
+        return (pipeline,)
 
 class DiffusersClipTextEncode:
     @classmethod
@@ -146,7 +143,7 @@ class DiffusersClipTextEncode:
     CATEGORY = "Diffusers"
 
     def concat_embeds(self, maked_pipeline, positive, negative):
-        positive_embeds, negative_embeds = token_auto_concat_embeds(maked_pipeline[0], positive,negative)
+        positive_embeds, negative_embeds = token_auto_concat_embeds(maked_pipeline, positive,negative)
 
         return (positive_embeds, negative_embeds, positive, negative, )
 
@@ -174,7 +171,7 @@ class DiffusersSampler:
     CATEGORY = "Diffusers"
 
     def sample(self, maked_pipeline, positive_embeds, negative_embeds, height, width, steps, cfg, seed):
-        images = maked_pipeline[0](
+        images = maked_pipeline(
             prompt_embeds=positive_embeds,
             height=height,
             width=width,
@@ -239,11 +236,9 @@ class StreamDiffusionCreateStream:
                 "use_denoising_batch": ("BOOLEAN", {"default": True}),
                 "frame_buffer_size": ("INT", {"default": 1, "min": 1, "max": 10000}),
                 "cfg_type": (["none", "full", "self", "initialize"], {"default": "none"}),
-                "acceleration": (["none", "xformers", "tensorrt"], {"default": "tensorrt"}),
+                "xformers_memory_efficient_attention": ("BOOLEAN", {"default": False}),
                 "lcm_lora" : ("LCM_LORA", ),
-                "tiny_vae" : ("STRING", {"default": "madebyollin/taesd"}),
-                "use_lcm_lora" : ("BOOLEAN", {"default": True}),
-                "use_tiny_vae" : ("BOOLEAN", {"default": True}),
+                "tiny_vae" : ("STRING", {"default": "madebyollin/taesd"})
             }, 
         }
 
@@ -252,9 +247,8 @@ class StreamDiffusionCreateStream:
 
     CATEGORY = "Diffusers/StreamDiffusion"
 
-    def load_stream(self, maked_pipeline, t_index_list, width, height, do_add_noise, use_denoising_batch, frame_buffer_size, cfg_type, acceleration, lcm_lora, tiny_vae, use_lcm_lora, use_tiny_vae):
-        model_id_or_path = maked_pipeline[1]
-        maked_pipeline: StableDiffusionPipeline = copy.deepcopy(maked_pipeline[0])
+    def load_stream(self, maked_pipeline, t_index_list, width, height, do_add_noise, use_denoising_batch, frame_buffer_size, cfg_type, xformers_memory_efficient_attention, lcm_lora, tiny_vae):
+        maked_pipeline = copy.deepcopy(maked_pipeline)
         lcm_lora = copy.deepcopy(lcm_lora)
         stream = StreamDiffusion(
             pipe = maked_pipeline,
@@ -278,179 +272,8 @@ class StreamDiffusionCreateStream:
             dtype=maked_pipeline.dtype
         )
         
-        batch_size = (
-            len(t_index_list) * frame_buffer_size
-            if use_denoising_batch
-            else frame_buffer_size
-        )
-
-        try:
-            if acceleration == "xformers":
-                stream.pipe.enable_xformers_memory_efficient_attention()
-            if acceleration == "tensorrt":
-                from polygraphy import cuda
-                from streamdiffusion.acceleration.tensorrt import (
-                    TorchVAEEncoder,
-                    compile_unet,
-                    compile_vae_decoder,
-                    compile_vae_encoder,
-                )
-                from streamdiffusion.acceleration.tensorrt.engine import (
-                    AutoencoderKLEngine,
-                    UNet2DConditionModelEngine,
-                )
-                from streamdiffusion.acceleration.tensorrt.models import (
-                    VAE,
-                    UNet,
-                    VAEEncoder,
-                )
-
-                def create_prefix(
-                    model_id_or_path: str,
-                    max_batch_size: int,
-                    min_batch_size: int,
-                ):
-                    maybe_path = Path(model_id_or_path)
-                    if maybe_path.exists():
-                        return f"{maybe_path.stem}--lcm_lora-{use_lcm_lora}--tiny_vae-{use_tiny_vae}--max_batch-{max_batch_size}--min_batch-{min_batch_size}--mode-{self.mode}"
-                    else:
-                        return f"{model_id_or_path}--lcm_lora-{use_lcm_lora}--tiny_vae-{use_tiny_vae}--max_batch-{max_batch_size}--min_batch-{min_batch_size}--mode-{self.mode}"
-
-                engine_dir = Path(engine_dir)
-                unet_path = os.path.join(
-                    engine_dir,
-                    create_prefix(
-                        model_id_or_path=model_id_or_path,
-                        max_batch_size=stream.trt_unet_batch_size,
-                        min_batch_size=stream.trt_unet_batch_size,
-                    ),
-                    "unet.engine",
-                )
-                vae_encoder_path = os.path.join(
-                    engine_dir,
-                    create_prefix(
-                        model_id_or_path=model_id_or_path,
-                        max_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                        min_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    ),
-                    "vae_encoder.engine",
-                )
-                vae_decoder_path = os.path.join(
-                    engine_dir,
-                    create_prefix(
-                        model_id_or_path=model_id_or_path,
-                        max_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                        min_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    ),
-                    "vae_decoder.engine",
-                )
-
-                if not os.path.exists(unet_path):
-                    os.makedirs(os.path.dirname(unet_path), exist_ok=True)
-                    unet_model = UNet(
-                        fp16=True,
-                        device=stream.device,
-                        max_batch_size=stream.trt_unet_batch_size,
-                        min_batch_size=stream.trt_unet_batch_size,
-                        embedding_dim=stream.text_encoder.config.hidden_size,
-                        unet_dim=stream.unet.config.in_channels,
-                    )
-                    compile_unet(
-                        stream.unet,
-                        unet_model,
-                        unet_path + ".onnx",
-                        unet_path + ".opt.onnx",
-                        unet_path,
-                        opt_batch_size=stream.trt_unet_batch_size,
-                    )
-
-                if not os.path.exists(vae_decoder_path):
-                    os.makedirs(os.path.dirname(vae_decoder_path), exist_ok=True)
-                    stream.vae.forward = stream.vae.decode
-                    vae_decoder_model = VAE(
-                        device=stream.device,
-                        max_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                        min_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    )
-                    compile_vae_decoder(
-                        stream.vae,
-                        vae_decoder_model,
-                        vae_decoder_path + ".onnx",
-                        vae_decoder_path + ".opt.onnx",
-                        vae_decoder_path,
-                        opt_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    )
-                    delattr(stream.vae, "forward")
-
-                if not os.path.exists(vae_encoder_path):
-                    os.makedirs(os.path.dirname(vae_encoder_path), exist_ok=True)
-                    vae_encoder = TorchVAEEncoder(stream.vae).to(torch.device("cuda"))
-                    vae_encoder_model = VAEEncoder(
-                        device=stream.device,
-                        max_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                        min_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    )
-                    compile_vae_encoder(
-                        vae_encoder,
-                        vae_encoder_model,
-                        vae_encoder_path + ".onnx",
-                        vae_encoder_path + ".opt.onnx",
-                        vae_encoder_path,
-                        opt_batch_size=batch_size
-                        if self.mode == "txt2img"
-                        else stream.frame_bff_size,
-                    )
-
-                cuda_steram = cuda.Stream()
-
-                vae_config = stream.vae.config
-                vae_dtype = stream.vae.dtype
-
-                stream.unet = UNet2DConditionModelEngine(
-                    unet_path, cuda_steram, use_cuda_graph=False
-                )
-                stream.vae = AutoencoderKLEngine(
-                    vae_encoder_path,
-                    vae_decoder_path,
-                    cuda_steram,
-                    stream.pipe.vae_scale_factor,
-                    use_cuda_graph=False,
-                )
-                setattr(stream.vae, "config", vae_config)
-                setattr(stream.vae, "dtype", vae_dtype)
-
-                gc.collect()
-                torch.cuda.empty_cache()
-
-                print("TensorRT acceleration enabled.")
-            if acceleration == "sfast":
-                from streamdiffusion.acceleration.sfast import (
-                    accelerate_with_stable_fast,
-                )
-
-                stream = accelerate_with_stable_fast(stream)
-                print("StableFast acceleration enabled.")
-        except Exception:
-            traceback.print_exc()
-            print("Acceleration has failed. Falling back to normal mode.")
+        if xformers_memory_efficient_attention:
+            maked_pipeline.enable_xformers_memory_efficient_attention()
         return (stream, )
 
 class StreamDiffusionSampler:

pyproject.toml

@@ -0,0 +1,15 @@
+[project]
+name = "comfyui-diffusers"
+description = "This extension enables the use of the diffuser pipeline in ComfyUI. It also includes nodes related to Stream Diffusion."
+version = "1.0.0"
+license = "LICENSE"
+dependencies = ["diffusers[torch]", "accelerate", "transformers", "safetensors", "omegaconf", "pytorch_lightning", "xformers", "git+https://github.com/cumulo-autumn/StreamDiffusion.git@main#egg=streamdiffusion[tensorrt]"]
+
+[project.urls]
+Repository = "https://github.com/Limitex/ComfyUI-Diffusers"
+#  Used by Comfy Registry https://comfyregistry.org
+
+[tool.comfy]
+PublisherId = "limitex"
+DisplayName = "ComfyUI-Diffusers"
+Icon = ""

requirements.txt

@@ -5,4 +5,4 @@ safetensors
 omegaconf 
 pytorch_lightning
 xformers
-git+https://github.com/cumulo-autumn/StreamDiffusion.git@main#egg=streamdiffusion[tensorrt]
+streamdiffusion[tensorrt] @ git+https://github.com/cumulo-autumn/StreamDiffusion.git@main