Wan-Animate-2 14B: Apache 2.0 weights vs Sume Motion Transfer

Wan-Animate-2 is open: the Wan-AI/Wan2.2-Animate-2-14B card lists the license as apache-2.0 and says the inference scripts and the Base and Distillation weights were released on August 07, 2026. Sume's hosted counterpart for driving a character with a video is Motion Transfer, which the video docs describe as one source video plus 1-8 reference images.

Weights facts are from the model card, read 2026-10-01. Sume limits are from Video generation, read 2026-10-01.

What does Wan-Animate-2 do?

The card describes an end-to-end character animation framework that consumes a driving video directly, with text-driven viewpoint control to separate the output camera from the driving video, and a Lite variant aimed at real-time streaming. Its inference command takes --refer-img-file (the character) and --refer-video-file (the template motion). The defaults are tuned for 8 A800 GPUs at 720P, and 480P was tested on 2 A800 GPUs.

Which hosted model matches it?

The video docs list a Motion Transfer model that accepts 4-30 seconds, with 480p/720p, and say it is listed only when its provider is configured. Check GET /v1/videos/models to see whether it appears for your workspace. It accepts video references but not audio, per the docs. The earlier generation is covered in Wan 2.2 Animate versus Motion Transfer.

What about face replacement?

That is a separate endpoint. The face swap docs say Avatar Face Swap 1.0 is a Beta model-run endpoint; it is not the same operation as full-body motion transfer.

Does the Apache license settle commercial use?

The card's license field says apache-2.0. Read the license text and any model-specific terms before deciding; this post does not give legal advice. Hosted use on Sume is governed by Sume's own terms, not by the weights license.