Text to video
Describe a scene in words and get an uncensored AI video up to 30 seconds long. How it works, what to write and which settings to use.
Text to video turns a written description into a video. It is the most popular way to start and the cheapest way to test an idea.
How to use it
- Go to Create → Text to video.
- Write your prompt, or switch to Scene builder and describe your idea in any language (Scene builder).
- Pick the model, quality, duration and aspect ratio (Models and settings).
- Optional: open the Timeline to plan what happens when, set Explicitness, pick Scene details or Match a look.
- Press Generate.
What to write
A strong prompt covers five things, in this order:
- Place and time: "a 1970s hotel room by the sea at night, warm lamp light"
- The people: how many, age (adults), build, hair, clothing or nudity. "a woman in her late twenties with long auburn hair and a man in his thirties"
- The action: visible, physical actions in order. "she pulls him down onto the bed, they kiss slowly, then…"
- The camera: shot size and movement. "slow push-in from the foot of the bed, 35mm lens"
- The look: light and color. "golden lamp light, soft shadows, warm film grain"
See the full prompt guide and ready-made recipes.
Using your characters
Type @ and pick one of your characters, for example @lena. If the character has photos, the video is made in References mode so their face and body stay the same, and you need to tick the rights checkbox.
Cost
Credits per second × duration. A 5 s Wan 3.0 clip at 480p is 125 credits; a 10 s Seedance 2.5 clip at 720p is 1,650 credits. See Pricing.
Questions
How long can a text-to-video clip be?
Up to 30 seconds with every model. Wan 3.0 starts at 2 seconds, Seedance 2.5 at 4 seconds.
Can I write my prompt in another language?
Yes in the Scene builder, which accepts any language and writes the English prompt for you. In Simple mode, English gives the most reliable results.
Why does my video look different every time?
Every generation is a new take. Keep the same prompt and settings and generate again, or use characters and a fixed look (Match a look, Scene details) to keep results consistent.