Wan 3.0 AI Video Generator: Clips with Sound You Can Edit and Extend - Wan3AI

Sign in, describe the shot and press the button. What you attach picks the job: nothing means text to video, one picture opens the clip, two pictures can be the first and last frame, more pictures or a sound file become references, and a clip can be edited, extended or used as a reference. Finished videos land in your library.

Alibaba's video model, released to the public on 24 August 2026. Alibaba says it renders up to 30 seconds at 480p, 720p or 1080p with sound made in the same pass, and the same model also edits and extends clips. It is closed: there are no open weights. Wan3AI reaches it through WaveSpeed's API.

Both are in the model menu. Wan 3.0 costs 10, 20 or 40 credits a second and goes to 1080p, 30 seconds, video editing and extension; our runs took 64 to 174 seconds. MiniMax H3 Lite costs 5 or 8 credits a second, stops at 768p and 15 seconds, cannot edit, and usually answers in under 30 seconds. Rough ideas out on H3 Lite, finish on Wan 3.0.

Up to 10 pictures, 5 video clips and 5 sound files. Reference clips may run 15 seconds in total, and clips plus the new video stay within 30 seconds. Refer to them in the prompt as Image 1, Video 1, Audio 1 in the order you uploaded them.

Not for video. A new account gets 10 credits, enough for two pictures in the Image tab. The shortest Wan 3.0 clip, 2 seconds at 480p, costs 20 credits, so video starts with a credit pack or a plan.

Yes, on every clip. Describe what should be heard (wind, footsteps, a door, music) and it is generated with the picture. When you extend a clip, its original sound is kept and the new seconds get their own. Lip sync on speech is uneven, so check dialogue before you publish.

Yes. Attach a clip, keep Edit selected and say what to change: the season, the weather, an object, the style. The first 15 seconds are read. You pay for the seconds read plus the seconds made, so a 3-second clip edited into 3 new seconds at 480p costs 60 credits. Reference pictures can guide the edit.

Attach a clip, choose Extend and write what happens next. Wan 3.0 continues from the last frame and returns your clip with 2 to 30 new seconds added. Only the new seconds are billed; a source longer than 120 seconds is cut to its last 120.

Mostly, with known habits. Reviewers and our own runs saw it cut to a new angle nobody asked for (our greenhouse clip does this in its last second), drift on busy scenes, and sync lips loosely. Short prompts with one action per shot hold together best. Failed or refused runs are refunded automatically.

10 credits a second at 480p, 20 at 720p and 40 at 1080p, rounded up to the whole second. A 5-second 720p clip is 100 credits; 30 seconds at 1080p is 1,200. First and last frames, reference pictures and sound files add nothing. Reference clips and edited footage are billed by their length at the same rate; an extension bills only the new seconds. The button shows the total before you start.

Any whole number of seconds from 2 to 30 (an edit returns 2 to 15), at 480p, 720p or 1080p, in 16:9, 9:16, 4:3, 3:4 or 1:1. A clip that starts from your picture follows that picture's proportions, and an edit or extension keeps the shape of your footage.

That depends on Alibaba's and WaveSpeed's current terms, which we cannot interpret for you. Read them before client work, and keep real people's faces, other companies' logos and material you don't own out of prompts and uploads.

No. Wan3AI is an independent site that sends your requests to Wan 3.0 through WaveSpeed, a third-party API platform. Alibaba does not run, endorse or partner with it.