WAN 2.x inference with ComfyUI

A streamlined and automated environment for running ComfyUI with WAN 2.x video models, optimized for use on RunPod.io

🔧 Features

• Automatic model and LoRA downloads via environment variables.
• Supports advanced workflows for video generation and enhancement using pre-installed custom nodes.
• Compatible with high-performance NVIDIA GPUs.

🔧 Built-in authentication

• ComfyUI
• Code Server
• Hugging Face API
• CivitAI API

📦 Template Deployment

👉 One-click Deploy on RunPod WAN 2.2 t2v (lightx2v)

👉 One-click Deploy on RunPod WAN 2.2 i2v (lightx2v)

👉 One-click Deploy on RunPod WAN 2.2 animate (lightx2v)

👉 One-click Deploy on RunPod WAN 2.2 lightx2v Dyno

👉 One-click Deploy on RunPod WAN 2.2 v2v Lucy Edit

Workflows

• Open from ComfyUI's interface on the left
• View/Download from Workflow examples

💻 Hardware Requirements

T2V-A14B/I2V-A14B (high/low)

• precision fp16
• video settings 1024x768 122 frames

Recommended GPU	VRAM	RAM
L40S, RTX 6000 Ada	40 GB	90GB

Animate

• precision fp16
• video settings 1024x768 77 chunks/Sampler

Recommended GPU	VRAM	RAM
L40S, RTX 6000 Ada	40 GB	105GB

Storage

Component	Minimum
Volume Storage	90GB (/workspace)
Pod Storage	15GB

⚙️ Environment Variables

ComfyUI Configuration

Variable	Description
COMFYUI_EXTRA_ARGUMENTS	Additional arguments for ComfyUI CLI

Authentication Tokens

Token Source	Variable
Code Server	PASSWORD
Hugging Face	HF_TOKEN
CivitAI	CIVITAI_TOKEN

Hugging Face Model Configuration

Model Type	Model	Safetensors/GGUF
Diffusion Model	HF_MODEL_DIFFUSION_MODELS[1-10]	HF_MODEL_DIFFUSION_MODELS_FILENAME[1-10]
Checkpoints	HF_MODEL_CHECKPOINTS[1-10]	HF_MODEL_CHECKPOINTS_FILENAME[1-10]
Text Encoders	HF_MODEL_TEXT_ENCODERS[1-10]	HF_MODEL_TEXT_ENCODERS_FILENAME[1-10]
Clip Vision	HF_MODEL_CLIP_VISION[1-10]	HF_MODEL_CLIP_VISION_FILENAME[1-10]
Audio Encoders	HF_MODEL_AUDIO_ENCODERS[1-10]	HF_MODEL_AUDIO_ENCODERS_FILENAME[1-10]
Model patches	HF_MODEL_PATCHES[1-10]	HF_MODEL_PATCHES_FILENAME[1-10]
VAE	HF_MODEL_VAE[1-10]	HF_MODEL_VAE_FILENAME[1-10]
Upscalers	HF_MODEL_UPSCALER[1-10]	HF_MODEL_UPSCALER_PTH[1-10]
Loras	HF_MODEL_LORA[1-10]	HF_MODEL_LORA_FILENAME[1-10]
VLM/mmproj	HF_MODEL_VL[1-10]	HF_MODEL_VL_FILENAME[1-10]
SAM segmentation	HF_MODEL_SAMS[1-10]	HF_MODEL_SAMS_FILENAME[1-10]

CivitAI LORAs

Variable	Description
CIVITAI_MODEL_LORA_URL[1-10]	Direct download link for LoRAs

Workflows

Variable	Description
WORKFLOW[1-50]	download link (compressed or plain)

🌐 Network Services

Service	Port	Access Type
ComfyUI	8188	Web
Code Server	9000	Web
SSH/SCP	22	Terminal

📚 Tutorials & Resources

• WAN Video Homepage
• Docs ComfyUI
• Enhance a Video
• AccVideo
• CausVid
• NAG
• FusionX LoRA
• LightX2V
• Pusa-VidGen
• Animated preview
• Tutorial s2v
• Wan 2.2 animate
• Wan 2.2 animate workflow
• Samplers
• HuMo Website
• HuMo Github
• Example prompts

🧩 Pre-Installed Custom Nodes

Interface

• rgthree
• Login
• Manager
• KJNodes
• GGUF
• ComfyMath
• Resolution master
• Lucy edit
• PG Nodes
• ComfyUi-Scale-Image-to-Total-Pixels-Advanced

Video/Upscale

• Video Helper Suite
• Frame Interpolation
• VideoUpscale with Model
• SD Upscale
• VRgamedevgirl
• VideoStitcher

Controlnet

• controlnet_aux

Flow

• Power Flow

Segmentation

• Segment anything
• RMBG
• SecNodes

Wan

• Wrapper
• StartEndFrame
• iamccs
• Wan22FMLF

Animate

• WanAnimate Enhancer

Sampling

• RES4LYF
• Noise
• NAG
• WanMoeKSampler
• Was affine
• TripleKSampler
• WanMoEScheduler

📦 Model Sources

• Wan-Video Github
• Wan-Video 2.2 Github
• Kijai
• Wan-AI Huggingface
• WAN-VACE Github
• FusionX Huggingface
• MoviiGen Huggingface

⚙️ Setup

Component	Version
OS	Ubuntu 22.04 x86_64
Python	3.11.x
PyTorch	2.8.0
CUDA	12.9
Triton	3.4.0
onnxruntime-gpu	1.22.x
ComfyUI	0.3.68
CodeServer	Latest

⚙️ Installed Attentions

Wheels

Package	Version
flash_attn	2.8.3
sageattention	2.2.0

Build for

Processor	Compute Capability	SM
A40	8.6	sm_86
L40S	8.9	sm_89