Wan 3.0 Prime text to video: write a whole scene, not a moment
Wan 3.0 Prime makes video with sound from a text prompt and runs long enough for a small story. How to write the beats, set it up and test it cheaply.
Watch the example on the Wan 3.0 Prime text to video page before you write anything. A young chess hustler beats challenger after challenger in a city square, an elderly woman sits down, makes a quiet first move, and his grin slowly fades. That is not a moment, it is a scene with a setup, a turn and a payoff. Most text-to-video models give you a few seconds of one action. This one, from Alibaba's Wan 3.0 family, has a duration slider that runs long enough to tell something, and it writes the sound along with the picture.
Who should pick it
- Short stories and sketches. Anything with a before and an after: a reveal, a reaction, a small comedy beat.
- Ads and social clips that need sound. Audio is on by default, so a street scene comes with its crowd and a kitchen with its sizzle.
- Storyboard tests. A long single take shows whether a scene idea works before you plan it shot by shot; see our storyboard guide.
If you only need a few seconds of drifting fog or a slow product turn, you are paying for range you will not use. Its sibling Wan 3.0 text to video starts cheaper, and our video model guide compares the rest of the field.
Write in beats
A long clip with a one-line prompt drifts: the model has time to fill and nothing to fill it with. Give it an order of events instead, the way the chess example does. Who is there, what happens first, what changes, how the camera moves, and the look. Three or four beats is plenty.
A street food vendor at night flips noodles in a flaming wok while a queue of customers waits. A gust of wind blows his paper hat into the wok; he freezes, then fishes it out with chopsticks and serves it to the next customer with a straight face. The camera starts on the flames, pulls back to show the queue, and ends on the customer's shocked expression. Neon signs, steam, sizzling sound, cinematic comedy.Early morning in a quiet bakery. A baker opens the oven and slides out a tray of golden croissants, steam rising. She sets them in the window, turns the sign to OPEN, and the first customer walks in as the bell above the door rings. Slow dolly from the oven to the window, warm soft light, calm atmosphere.Vertical clip. A paper boat floats down a rain gutter, bumps over a small waterfall, spins in a puddle and comes to rest against a red rubber boot. Low camera at water level, following the boat, rain sounds, bright overcast light.The settings, briefly
- Resolution: three steps, the middle one is the default.
- Aspect ratio: 16:9 by default, plus 9:16, 1:1, 4:3 and 3:4.
- Duration: whole seconds. The default is short; drag it up only when your prompt has the beats to fill it.
- Generate audio: on by default. Switch it off if you will add your own music; in our checks that did not lower the price, so leave it on unless the sound is in the way.
- Prompt expansion (advanced): lets an optimizer rewrite your prompt with more detail. Handy for a vague idea, unhelpful when you wrote careful beats, so it is off by default.
- Seed (advanced): a fixed number repeats a result, so you can change one word and compare.
Lifehacks that save money
- Test the scene short and small. Run the prompt at the lowest resolution and a short duration first. If the characters and the look are right, raise both for the final take. The Generate button shows the price before each run.
- Fix the seed while you tune. Once a draft is close, keep its seed and adjust the wording; the changes you see are then caused by your words, not by luck.
- Animate a still instead, when the look matters more than the story. Make the frame first, then use Wan 3.0 Prime image to video so the first frame is exactly the one you approved.
The honest limitation
Length is a gift and a risk. The longer the clip, the more chances for a hand to blur, a face to shift between beats or a background detail to change. A long take that is almost right still costs a long take, so the cheap test above is not optional. And it is one continuous shot from one prompt: there are no cuts to separate scenes and no way to keep a character identical across two runs except a fixed seed and the same description. For a series, plan to edit several clips together.
Questions
Does Wan 3.0 Prime text to video make sound?
Yes, audio is on by default. Describe the sounds you want in the prompt, or turn Generate audio off for a silent clip.
How long can a clip be?
The duration slider on the model page shows the full range in whole seconds. The default is short; longer clips cost more, and the price shows on the Generate button.
Should I turn on prompt expansion?
Only for a rough idea you want filled in. If you wrote your scene beat by beat, leave it off so the model follows your words.
