Wan 3.0

Name: Wan 3.0 Video
Author: Alibaba (Tongyi Lab)

byAlibaba (Tongyi Lab)

Thirty-second single takes that hold a character's face and arrive with dialogue, foley and room tone already in the mix

Roll the take

How Wan 3.0 Video works

You direct it like a shot: describe the moment, pin down references or start and end frames, choose a length up to 30 seconds, and get picture and sound back together.

Frame the scene

Write subject, place and movement, or anchor the shot with a start frame, an end frame and numbered reference media.

Set length and tier

Pick 2 to 30 seconds, draft at 480p or 720p on Prime, then lock the timing you like.

Render picture and sound

Export one continuous clip with lip-synced speech, foley and ambience generated alongside the image.

What Wan 3.0 Video is good at

Length, consistency and sound are the throughline: long takes without stitching, faces and products that stay put across references, and audio that arrives in the same pass.

Unbroken 30-second takes

Run a shot from 2 up to 30 seconds in a single pass, with subject identity and spatial layout holding steady. Add time brackets like [0-6s] so long takes keep moving instead of idling.

Reference-locked characters

Bind up to ten images, five video clips and five audio clips by number (Image 1, Video 1, Audio 1) so a face, a location and a voice stay consistent from shot to shot. Give each reference one clear role.

Standard finals, Prime drafts

Both tiers share one capability set. Prime runs about 1.35x the cost of Standard for faster turnaround, so test timing and framing there at lower resolution, then render the final on Standard.

Start-to-end frame control

Give it a start image and an end image and it interpolates the move between them. Keep lighting, lens and subject scale matched, and describe how the transition happens in the prompt.

Made with Wan 3.0 Video

The gallery runs from archival documentary and car commercials to a metronome visualizer, a color-clash lookbook and an underwater cave, all with sound generated in sync.

Archival craftsmanship documentary with synchronized soundscapes

High-velocity automotive spot with dynamic camera tracking

Surreal sonic visualizer with synchronized metronome cadence

High-contrast editorial lookbook with vivid color tension

Subterranean cinematic sequence with spatial underwater acoustics

What people build with Wan 3.0 Video

Directors, brand teams, social producers and music-video makers reach for it when one long, consistent, voiced shot beats a pile of short silent clips.

Single-take film directors

Block a scene as one uninterrupted 30-second shot with dialogue, ambience and foley already baked in, instead of stitching short clips and rebuilding the soundtrack.

Brand teams locking a product look

Feed in clean product and location references and keep the exact object consistent across spots, then run quick Prime drafts before committing to a Standard 1080p final.

Vertical social producers

Set 9:16, write a quoted line, and get a talking hook with synced voice and sound in one render for shorts and spots.

Music-video visual makers

Drive a performer with an audio reference for voice or soundtrack feel, then use time brackets to sequence beats across the track length.

Designers animating keyframes

Take a designed still, set it as the first frame, add an end frame, and describe only the motion to bring a product or poster concept to life.

Wan 3.0 Video specs

Plans and credits

Versions: Standard, Prime
Inputs: Prompt (text), Start Image (image), End Image (image), Reference Images (up to 10 images), Reference Videos (up to 5 videos), Reference Audio (up to 5 audios)
Output: video
Duration: 2–30 seconds
Resolution: 480p, 720p, 1080p
Aspect ratios: Adaptive, 16:9, 4:3, 1:1, 3:4, 9:16
Audio: Generates audio
Cost: 138–11,572 credits per run

Wan 3.0 questions

Prime is the faster tier for iteration, while Standard is the one to choose for finals where cost matters. They share the same capability set, so Prime does not add new features. It costs roughly 1.35x Standard, so draft timing and framing on Prime at 480p or 720p, then lock the final.

Match the variant to your anchor. Text-to-video suits concept exploration and establishing shots, but subjects can vary between runs. Image-to-video animates a designed first frame or bridges a start and end image, with the look locked to that frame. Reference-to-video keeps a character, product or location consistent using images, clips or voice audio, but needs careful numbering in the prompt.

Clips run from 2 to 30 seconds in one pass at 480p, 720p or 1080p, with 1080p as the default. If you use reference clips, their duration counts toward the 30-second budget alongside the output, so shorten the output when using long references.

Yes, it generates synchronized audio such as dialogue, foley, ambience and music, and Generate Audio is on by default. Turn it off if you'll score the piece yourself. Put spoken lines in quotation marks for lip-synced voice, and say "no background music, no voice lines" to keep the sound quiet.

It struggles with fast action involving many interacting bodies, readable on-screen text, and crowded spatial scenes. Sparse prompts on 20 to 30 second takes can drift or idle, and 1080p at 30 seconds is costly for throwaway clips.

It comes from Alibaba's Tongyi Lab. Wan 3.0 is a closed, commercial model, so it isn't suited to local open-weight workflows.