Video · alibaba/wan-3
WAN 3 — a clip from your description or motion in your own frame
WAN 3.0 is the current generation of the Alibaba video model. It builds a clip from a written description and can animate a still: the picture becomes the first frame and the prompt sets what happens next. limko offers three resolutions and clips of five or ten seconds, with the price on the button before you start.
from 41.2 ₽ per video
What WAN 3 does
- Text to video: the scene, the light and the camera move are all described in words.
- Animation of a still: the uploaded picture becomes the first frame and sets the format of the clip.
- Three resolutions — 480p for drafts, 720p for social media, 1080p for the final cut.
- Negative prompt, seed for repeatable takes and automatic expansion of a short description.
Resolution and price
The price below is for a five second clip. Ten seconds costs exactly twice as much, because the provider bills video by the second. The exact amount is always on the button before you start.
| Resolution | Price per clip |
|---|---|
| 480p — drafts, previews, social media | from 41.2 ₽ |
| 720p — the everyday working format | twice the 480p price |
| 1080p — final cut and big screens | four times the 480p price |
Text to video and animation of a still
With no picture attached the model works as plain text to video: you describe the scene, the light and the camera move and get a clip. The format is set separately — 16:9, 9:16, 1:1, 4:3 or 3:4 — or leave it on adaptive and the model picks the proportions that suit the description.
Attach an image and it becomes the first frame while the prompt describes what follows. In this mode the aspect ratio control has no effect: the clip repeats the format of the source picture, so crop it beforehand.
A short description gets expanded by the model itself, which helps when the prompt is a single phrase. When the scene has to hold exactly, write it out in detail: what you state outweighs what the model adds.
When WAN 3 and when a neighbour
WAN 3 is the choice when you need a solid clip for sensible money: 480p costs less than most video models in the catalogue and 720p covers social media and product cards. Veo 3.1 and Sora 2 hold a complex scene and dialogue more confidently, but cost noticeably more for the same running time.
WAN 3 declares no audio parameters — the output is picture only. When the clip needs a synchronised soundtrack, look at WAN 2.7 I2V, where audio is a separate mode.
You pay in credits and only for a generation that succeeded. If the model fails, the credits return to the balance automatically.
Prompts that work
Describe the motion, not only the object: where the camera goes, what changes in the frame, at what pace. Push the rest out with a negative prompt.
Text to video
Camera moves along an old town street in the morning, wet cobblestones, warm light in the bakery windows, steam above a cup on the table. Slow push forward, no people in frame.
Animating a photo
Same frame: the woman turns her head towards the camera and smiles, the wind moves her hair slightly. Keep the background and the clothes, calm motion.
Product clip
A perfume bottle on black glass, slow orbit around the object, highlights sliding across the edges, water drops on the surface. Studio light, macro shot.
Worth knowing before you start
One run returns one clip. For a set of takes start several generations: the same seed repeats the result, a different one gives other variants.
The model counts in minutes rather than seconds: five seconds at 480p usually lands within a couple of minutes, 1080p takes noticeably longer. You can close the page — the finished clip appears in the generation history.
A negative prompt clears the frame of captions, flicker and extra hands. It is optional, but it helps when the model repeats the same mistake.
Frequently asked about WAN 3
How does WAN 3 differ from the earlier WAN models?
It is the next generation of the line: one model covers both text to video and animation of a still, and it goes up to 1080p. The earlier versions stay in the catalogue, so a series started on them can continue without a change of style.
Can I make a video from my own photo?
Yes. Upload the shot and describe the motion — the picture becomes the first frame of the clip. The video repeats the format of the photo, so crop it beforehand.
What clip lengths are available?
Five or ten seconds per run. Ten seconds costs exactly twice as much as five, because video is billed by the second.
Does it understand a prompt in Russian?
Yes, the description can be written in Russian. For fine wording of camera motion English is still more predictable, which holds for every model in the catalogue.
How much does a clip cost?
It depends on resolution and length: 480p is the cheapest, 720p costs twice that, 1080p four times. The exact amount is shown on the run button before anything is charged.