Run ComfyUI with OneDiff

Download the Image

Run the following command to download the pre-installed image of ComfyUI + OneDiff:

docker pull oneflowinc/comfyui-onediff:latest

Run the Image

docker run -it --shm-size=8G -P --privileged --runtime=nvidia --rm \
    --gpus all --network host \
    -e ONEDIFF_INITIAL_PACKAGE_NAMES_FOR_CLASS_PROXIES=diffusers,/app/ComfyUI/comfy \
    -v /path/to/comfyui/models/:/app/ComfyUI/models/ \
    oneflowinc/comfyui-onediff python /app/ComfyUI/main.py

NOTE

• If you would like to run graph by VM, set env variable ONEFLOW_RUN_GRAPH_BY_VM=1
• The env variable ONEDIFF_INITIAL_PACKAGE_NAMES_FOR_CLASS_PROXIES=diffusers,/app/ComfyUI/comfy is required
• The directory structure of /path/to/comfyui/models should follow the structure of ComfyUI/models which means:

models/
├── checkpoints
├── clip
├── clip_vision
├── configs
├── controlnet
├── diffusers
├── embeddings
├── gligen
├── hypernetworks
├── loras
├── style_models
├── unet
├── upscale_models
├── vae
└── vae_approx

How to Use

In the menu's section "Add Node/utils", there is a "Model Speed Up" Node that takes a MODEL as input and outputs an accelerated model.

Related Workflows

Load the screenshot images below in ComfyUI to try the workflow

SDXL Workflow

LoRA