ComfyUI-WanVideoWrapper 🧩 Workflow Open source

What it is

ComfyUI-WanVideoWrapper by Kijai (Apache-2.0) wraps Alibaba's Wan video models and many related research models as ComfyUI nodes, often before they reach ComfyUI's native support. The author describes it as a perpetual work in progress and a personal sandbox, and advises using native Wan nodes when a feature is available there. It is widely used for text-to-video, image-to-video, video-to-video and control workflows with Wan and its derivatives.

Key features

• Text-, image- and video-to-video with Wan models
• Early support for Wan-based research models and LoRAs
• Block swapping and offloading to fit large models in limited VRAM
• torch.compile support and async prefetch for speed
• Example workflows in the repo
• Companion fp8-scaled model uploads on Hugging Face

How to install

Clone the repo into ComfyUI/custom_nodes, then run pip install -r requirements.txt (portable builds use the embedded Python). Download Wan models from the author's Hugging Face pages and place them in the matching models folders. It can also be installed through ComfyUI-Manager.

Pricing and rights

Free, Apache-2.0. The Wan model weights are released under Apache-2.0 by Alibaba, so commercial use is generally allowed, but check the licence of any third-party LoRA or research model you load.

Who it is for

Advanced ComfyUI users who want the newest Wan video features and can manage VRAM settings and breaking changes.

Verdict

The go-to place for bleeding-edge Wan workflows. It is experimental by design, issue posting is restricted for new accounts, and the last push (May 2026) is slower than Kijai's other repos.