Verified Sep 14, 2026

Is Wan 3.0 Open Source?

Wan 3.0 works in ComfyUI through hosted API Nodes. No verified official weights, checkpoints, or runnable local inference release were found in the sources checked on September 14, 2026.

Nopublic Wan 3 weights
YesComfyUI API Nodes
Nolocal inference
Yeshosted API
InputText · image · video · audio · file · link
One modelwan3.0-video
OutputUp to 30s · 1080P · native audio

Wan 3.0 is hosted; Wan 2.2 remains the open-weight path

ComfyUI supports Wan 3.0 text-to-video, image-to-video, and reference-to-video via API Nodes. The AlibabaCloud-Official/Wan3.0 repository is publicly accessible, but its visible files are README and LICENSE only; its official ownership has not been independently confirmed.

Open-source and local-run status

This is a dated status check, not a claim that weights will never be released.

SurfaceStatusWhat it means
Official weights / checkpointsNot verified as releasedNo verified downloadable Wan 3.0 weights in the checked sources.
GitHub repositoryPublic; ownership unverifiedAlibabaCloud-Official/Wan3.0 contains README and LICENSE, without runnable inference code or checkpoints.
ComfyUI API NodesAvailableOfficial text-to-video, image-to-video, and reference-to-video templates call a hosted service.
Local inferenceNot available as a verified official releaseRunning ComfyUI locally does not move Wan 3.0 generation onto your GPU.
Hosted APIAvailableModel Studio offers wan3.0-video and wan3.0-video-prime.
Open-weight alternativeWan 2.2Official code and weights support local ComfyUI workflows.

Do not download unverified 'Wan 3.0' checkpoints

A repository name, a pinned project, or an Apache-2.0 license on documentation does not verify ownership or provide model weights. Check official backlinks and actual release files before treating a repository as a local Wan 3.0 release.

Choose the path that matches the job

Use Wan 3.0 API Nodes for a visual ComfyUI workflow with cloud generation. Choose Wan 2.2 when you need downloadable official weights and inference on your own GPU.

Hosted and local alternatives