ComfyUI Video Docker

Docker image for ComfyUI video generation, supporting models such as LTX 2.3, Wan 2.2, and more. Built on RunPod PyTorch.

Quick Start

# Pull and run
docker pull ghcr.io/loan-mgt/comfyui-wan:latest
docker run -d --name comfyui --gpus all -p 8188:8888 ghcr.io/loan-mgt/comfyui-wan:latest

Access at: http://localhost:8188

Features

• RunPod Base: Built on runpod/pytorch with CUDA 12.8.1 and PyTorch 2.8.0
• uv Package Manager: System-level Python dependencies managed with uv
• Fast HF Downloads: HF_XET_HIGH_PERFORMANCE=1 enabled for accelerated Hugging Face transfers
• Custom LoRA Support: Download custom LoRA models from Hugging Face at startup via --model
• Private Model Support: Pass a Hugging Face token at runtime via --hf-token
• Bundled Custom Nodes:
  • ComfyUI-Manager
  • hf-lora-loader
  • ComfyUI-RunpodDirect

Requirements

• Docker with GPU support
• NVIDIA GPU (16 GB+ VRAM recommended)

Advanced Usage

Persist Models

Mount a local directory so models survive container restarts:

docker run -d --name comfyui --gpus all -p 8188:8888 \
  -v /path/to/models:/app/ComfyUI/models \
  ghcr.io/loan-mgt/comfyui-wan:latest

Environment Variables

Variable	Default	Description
COMFYUI_DIR	/app/ComfyUI	ComfyUI installation directory
HF_HOME	/app/ComfyUI/models/.cache/huggingface/	Hugging Face cache location
HF_XET_HIGH_PERFORMANCE	1	Enables high-performance Xet-backed HF transfers

Python Environment

• Dependencies are installed system-wide using uv pip install --system
• ComfyUI is started with python3 main.py --listen 0.0.0.0 --port 8888

Custom LoRA Models

Download LoRA models from Hugging Face at startup using --model. Files are saved to models/loras/{filename}.

# Single LoRA
docker run -d --name comfyui --gpus all -p 8188:8888 \
  ghcr.io/loan-mgt/comfyui-wan:latest \
  --model "user/repo lora_weights.safetensors"

# Multiple LoRAs
docker run -d --name comfyui --gpus all -p 8188:8888 \
  ghcr.io/loan-mgt/comfyui-wan:latest \
  --model "user/repo style_lora.safetensors" \
  --model "another/repo character_lora.safetensors"

--model format: "repo_id filename" (space-separated)

Private Models (Hugging Face Token)

docker run -d --name comfyui --gpus all -p 8188:8888 \
  ghcr.io/loan-mgt/comfyui-wan:latest \
  --hf-token "hf_yourTokenHere" \
  --model "private-user/private-repo model.safetensors"