How kijai/comfyui-wanvideowrapper works
This repository is a ComfyUI custom-node wrapper for WanVideo and related video-generation models. The principal workflow is to configure model inputs and conditioning in a ComfyUI workflow, load a transformer and supporting model assets, run video sampling, then decode latents for output and preview. Separate node families add model-specific animation, audio, camera, and control techniques. The graph combines README-documented capabilities with relationships visible in sampled files and the SOURCE INDEX; unsampled node wiring is not inferred.
Main entry points: nodes_sampler.py Video sampling · nodes.py ComfyUI nodes · nodes_model_loading.py Model loading · model.py Wan transformer
Components 21
Workflow nodes
- ComfyUI nodes
nodes.py - Model loading
nodes_model_loading.py - Video sampling
nodes_sampler.py - Latent preview
latent_preview.py
Wan runtime
- Wan transformer
wanvideo/modules/model.py - Attention
wanvideo/modules/attention.py - Video VAE
wanvideo/wan_video_vae.py - Flow schedulers
wanvideo/schedulers/fm_solvers_unipc.py - LCM scheduler
wanvideo/schedulers/scheduling_flow_match_lcm.py
Model extensions
- MultiTalk nodes
multitalk/nodes.py - MultiTalk runtime
multitalk/multitalk.py - LongCat layers
LongCat/layers.py - UniAnimate
unianimate/nodes.py - Related model nodes
ATI
Conditioning
- Control models
controlnet - Context windows
context_windows/context.py
Runtime support
- GGUF weights
gguf/gguf.py - Linear and LoRA
custom_linear.py - Radial attention
wanvideo/radial_attention/attn_mask.py - Model utilities
utils.py
Other components
- Workflow author
Connections 22
- Workflow author → ComfyUI nodes: builds workflow
- ComfyUI nodes → Model loading: loads models
- ComfyUI nodes → Video sampling: runs generation
- Model loading → Wan transformer: loads transformer
nodes_model_loading.py - Model loading → Video VAE: loads VAE
nodes_model_loading.py - Model loading → GGUF weights: supports GGUF
nodes_model_loading.py - Model loading → Linear and LoRA: patches weights
nodes_model_loading.py - Video sampling → Wan transformer: samples with
- Video sampling → Flow schedulers: uses scheduler
- Video sampling → LCM scheduler: supports scheduler
- Video sampling → Latent preview: streams previews
- Wan transformer → Attention: uses attention
wanvideo/modules/model.py - Wan transformer → Radial attention: uses masks
wanvideo/modules/model.py - MultiTalk nodes → MultiTalk runtime: runs audio motion
- MultiTalk runtime → Attention: calls attention
multitalk/multitalk.py - MultiTalk nodes → Video sampling: adds workflow nodes
- LongCat layers → Wan transformer: extends transformer
LongCat/layers.py - UniAnimate → ComfyUI nodes: adds animation nodes
- Control models → ComfyUI nodes: adds control nodes
- Context windows → Video sampling: windows frames
- Model utilities → Linear and LoRA: patches weights
utils.py - Related model nodes → ComfyUI nodes: adds model nodes