MDL-6065EST.2026 · IDX.502
VideoIn production

Wan2.7-Video

Alibaba Cloud · China · 2026

Four models under one name - text, image and reference to video plus a video editor - aimed at changing a finished clip by instruction: an actor's line, the camera move, the weather, without regenerating the scene.

wujec.ai score

/10

Community score

no votes yet
Sign in to rate

Wan2.7-Video, announced by Alibaba on 7 April 2026, is less a single model than a set of four sharing a name: text-to-video, image-to-video, reference-to-video and a dedicated video editor. The pitch behind the packaging is that the interesting work in video is not the first generation but everything after it, so the family is built around changing an existing clip by instruction - a character's action, a line of dialogue, an appearance, a scene, a style or the camera work - rather than generating a new one and hoping it lands closer. Some of the specifics are unusual enough to be worth listing. Dialogue can be rewritten after the fact with lip movement re-synchronised and the original voice preserved. Consistency of up to five distinct characters is maintained across separate clips, each keeping its own voice and appearance. An audio clip can be used as a control input to change conditions such as weather, and a multi-panel image can dictate story order, framing and character setup. Video continuation accepts a specified end frame, which is Alibaba's answer to clips that stop as if braking. Generation runs from two to fifteen seconds at 720p or 1080p. Two details for anyone integrating it. The prompt field takes up to 5,000 characters, and Alibaba dropped the explicit shot-type parameter used by earlier versions - shot structure is now described in the prompt, and if it is not described the model decides for itself whether to cut. The company's own SDK does not cover this generation: the documentation states plainly that the SDK supports Wan2.6 and earlier, and Wan2.7 must be called over the new HTTP protocol. The published API snapshot, wan2.7-t2v-2026-06-12, is a June refresh rather than the April debut. Like every served video generation since Wan2.2, this one has no published weights and no technical report. The closure covers the flagship line only: the same team went on releasing open weights built on the Wan2.2 base, among them Wan-Dancer in July 2026 and the Animate-2 character-animation models in July and August 2026. It was superseded as the flagship on 24 August 2026 by Wan3.0, which doubled the ceiling to thirty seconds, but it remains available and documented in parallel.

#video generation#video editing#audio#closed weights
Official website

Videos

No videos yet.

Reviews

No reviews yet. Be the first!

Sign in to write a review