UlazAI - AI Image & Video Tools
Now in Video Studio
Wan 3.0 stops making you tell the story in five seconds.
Most video models hand you five or ten seconds and let you work it out from there. Wan 3.0 Video runs to thirty in a single generation, keeps the sound with it, and takes direction from more than a prompt: first and last frames, up to ten images, five clips, five audio fragments, a document, or a web page you want it to read.
- Duration
- 2-30s
- Resolution
- 480p / 720p / 1080p
- References
- image, video, audio, document, link
- From
- 96 credits
96 credits buys five seconds at 720p with audio. 480p halves that. Full rates below.
Five seconds is enough to show a thing. It is not enough to show a thing changing. Thirty seconds is a different unit: something can start, turn and land inside one shot, without a cut you had to hide. That is the whole difference here, and everything below is about how you steer it.
Thirty seconds, one shot
The clip beside this was one generation. Twenty seconds, 720p, one continuous take, audio written by the model. Nothing was stitched. You set the length between 2 and 30 seconds and pay by the second, so a long clip costs what a long clip costs and nothing extra.
Choose it when the idea needs time to arrive: a journey, a transformation, a scene that has to breathe before it means anything.
Five kinds of reference
First and last frame
Give it the picture it opens on, the picture it closes on, or both. The model fills the road between them. Images can be 240 to 8000 pixels on a side, up to 20 MB.
Up to ten images
Numbered in the order you supply them, so the prompt can point at Image1 and Image2 directly. Use it for a face, a product, a set, a palette that has to stay the same.
Up to five clips
Each one to fifteen seconds, fifteen seconds combined. Motion, staging and style carry over. With video in, input plus output has to stay inside thirty seconds.
Up to five audio fragments
One to fifteen seconds each. Give it the voice, the room tone or the music it should move to.
A document or a link
One file up to 100 MB and 50 pages, or one public web page. The model reads it and works from what is in there. Not both at once, and neither of them alongside frames.
Sound is not an afterthought
Audio is on by default and comes out of the same generation as the picture, so footsteps land on the step and a room sounds like the room you can see. Turn it off with one switch if you are scoring the clip yourself.
Pricing
Per second, charged on the seconds you asked for. If you supply reference video, the seconds you put in are billed too, because upstream counts them the same way.
| Resolution | Per second | 5 seconds | 15 seconds | 30 seconds |
|---|---|---|---|---|
| 480p | 9.6 | 48 | 144 | 288 |
| 720p | 19.2 | 96 | 288 | 576 |
| 1080p | 38.4 | 192 | 576 | 1152 |
A failed generation refunds itself. You are never charged for a clip you did not get.
Wan 3.0 next to Wan 2.7
| Model | Wan 2.7 | Wan 3.0 Video |
|---|---|---|
| Longest clip | 15s | 30s |
| Resolutions | 720p, 1080p | 480p, 720p, 1080p |
| Reference images | 5 | 10 |
| Reference video | 1 source clip | up to 5, 15s combined |
| Reference audio | no | up to 5 |
| Document or web page | no | yes |
| From (5s 720p) | 80 credits | 96 credits |
Wan 2.7 stays where it is. If fifteen seconds and one source clip cover the job, it is the cheaper way to do it.
Wan 3.0 questions
How long can one Wan 3.0 clip be?
Thirty seconds, in one generation, without stitching. If you supply reference video, the length of what you put in counts toward the same thirty.
Does Wan 3.0 make its own sound?
Yes. Audio is generated with the picture and is on by default. You can switch it off.
Can I combine reference images with a first and last frame?
No. The model takes one or the other. Pick frames when you know exactly where the shot starts and ends, and reference images when you care more about what stays consistent than where it begins.
What does Wan 3.0 cost?
19.2 credits per second at 720p, so 96 for five seconds. 480p is half of that, 1080p is double. Nothing is charged for a generation that fails.
What is the difference with Wan 3.0 Video Prime?
Same inputs, same limits, same controls. Prime finishes faster and costs more per second. In our own run of the same five second clip at 720p, Prime took 108 seconds against 200. One measurement, not a benchmark.
Start at 720p and five seconds.
96 credits, sound included. If it works, take the same prompt to thirty.