Updated
Wan 3.0: Alibaba’s video model for 30-second clips with sound
Wan is Alibaba’s video generation model. In version 3.0 it is one model for every job: shoot a clip from text, animate a frame, build a scene from your own photos and videos, edit or extend a finished clip. Clips run up to thirty seconds, and dialogue, sound and music are generated together with the picture. Every video on this page was made by Wan 3.0 itself in Twelver.
Below is the model itself, not screenshots. Describe a clip or upload a frame: the answer arrives right on this page, and your first try needs no sign-up.
What you are trying right now
This is not a demo mode: the request goes to the same model as the main chat. The only difference is that the conversation ends here after the first answer — you can carry on in the chat, and the clip moves there with you. A task takes two to five minutes; that is normal, and you do not need to close the page.
Several shots from one request
Wan 3.0 reads a storyboard: you can lay the clip out second by second, and the model cuts between the shots itself, changing framing and giving each its own sound. This clip was made from a single text request, with no source frames and no editing.
How to write a storyboard
Each shot is its own timestamped stretch: what is in frame, how the camera moves and what we hear. Put a character’s line in quotes and the model will voice it. Keep a single shot to four to six seconds so the model has time to show the whole action.
Frame to video: a continuation with no seam
In frame-to-video mode your picture becomes the clip’s first frame, exactly as you supplied it. That is handy for longer stories: the last frame of one clip becomes the first frame of the next, and the join between them does not show. The clip below starts precisely where the previous one ended.
You can also supply a last frame, and the model invents the motion between the two pictures. In this mode the clip’s shape follows the frame.
References: the product stays the product
Wan 3.0 takes more references than any other engine here: up to ten images, five videos and five audio clips at once. Images hold characters, products and style; videos carry a camera move or choreography; audio sets a voice’s timbre or a music beat. This clip was built from a single photo of a bottle: its shape and the lettering on the label stayed as they are in the picture.
References have to be named and described
The prompt cites them as Image 1, Video 1, Audio 1 — each kind numbered separately, in upload order — and describes them in words: “the woman in the red dress from Image 1 picks up the bottle from Image 2”.
An audio reference sets the voice, not the words: write the line in the prompt and say whose voice speaks it — “Image 1 says in the voice of Audio 1: ‘…’”.
Editing and extending a finished clip
You can rework your own video in words: remove or add an object, change clothing, restyle or relight it, change a character’s line. Whatever you did not ask to change stays as it was. On the left is the source; on the right, the same clip after the edit.
Extending adds new seconds to a clip — after its end or before its start. The clip below is the same five seconds with five more that Wan wrote on: the camera pulls back to show the whole counter.
How editing and extending are billed
The source is paid for along with the result. Editing a 5-second clip costs the same as 10 seconds. An extension returns the whole source plus the new part, so adding 5 seconds to a 5-second clip is 15 billed seconds.
The ceiling is 30 seconds for input and output together, and the source can be at most 15 seconds. A short clip can be extended further: a five-second one by another twenty seconds, a ten-second one by ten.
Say in words where the new part goes: “after the last frame” or “before the first frame”.
Four modes
| Mode | Input | When you want it |
|---|---|---|
| Text to video | A description only | No frame, just an idea. Timestamped storyboards; the aspect ratio is set by hand. |
| Frame to video | A first frame, optionally a last one | Bring a photo to life, or continue the previous clip without a seam. The shape follows the frame. |
| References | Up to 10 images, 5 videos and 5 audio clips | Keep characters and products consistent, repeat a camera move, set a voice or a beat. |
| Edit and extend | A video up to 15 s and up to 10 images | Change a finished clip in words, or write new seconds onto it. |
How Wan differs from HappyHorse
Two engines from the same provider with similar modes. The difference comes down to length, references and price.
| Wan 3.0 | HappyHorse 1.1 | |
|---|---|---|
| Clip length | 2–30 s, plus extending | 3–15 s |
| References | Up to 10 images, 5 videos and 5 audio clips | Up to 9 images |
| Editing input | Up to 15 s | Up to a minute |
| Price per second | Cheaper at 720p, has 480p | Cheaper at 1080p |
When to pick another model
- A talking character in your recorded voice — Hailuo: it syncs lips to speech noticeably more cleanly. Wan’s audio reference sets the timbre, and the words come from the prompt.
- Editing a minute-long source — HappyHorse, which takes video up to sixty seconds. Wan takes up to fifteen.
- Stitching several clips with invented transitions — Seedance. Wan extends a single clip.
- Above 1080p — Veo.
How not to overspend
- Draft in 480p. It costs half of 720p, and the composition, motion and sound already show there. 1080p costs twice as much as 720p — step up once, on the final take.
- Count the input, not only the output. Editing, extending and video references add the source’s length to the bill. Trim it to the part you need first — trimming is free.
- Build a long clip shot by shot. Test the idea on five seconds before ordering thirty.
- Editing on the timeline is free. Trimming, joining, laying sound and grabbing a frame are ordinary file operations, not generation.
Frequently asked questions
What is Wan, and whose model is it?
What is new in Wan 3.0 compared with 2.7?
How long a clip does Wan 3.0 make?
Can Wan voice a character in my voice?
How many references does Wan 3.0 take?
How do I extend a finished video?
How much does a Wan 3.0 clip cost?
Do I need to pick Wan in the chat by hand?
Examples and prices are on the video generation page. What changed in 3.0 is in the launch post. Other models are in the book’s contents; video in general is in the video generation book.
Try it yourself
Everything in this guide runs inside Twelver
One chat for text, images, video, music and voice — no separate services or subscriptions.
Open Twelver chat