Features · Models
The Previews, and the models.
Read this page before you pay.
The Previews.
Preview and Long take paint real clips in your browser with WebCodecs from your prompt and a seed. Frame 0 is your start picture and the last frame is your end picture. Preview makes clips of 2, 4, 6 or 8 seconds; Long take makes clips up to a minute that start on a picture but cannot end on one.
- Real files, no model
- No key needed
- Labelled Simulated in the composer
The models
What is ready where.
| Model | Runs | State |
|---|---|---|
| Preview | In your browser | Ready, simulated; clips of 2 to 8 s |
| Long take | In your browser | Ready, simulated; clips up to 60 s, no end picture |
| Veo 3.1 | On the server | Connected in code; ready where a key is set; up to 8 s with sound |
| Kling 2.1 | On the server | Connected in code; ready where a key is set; 10 s clips from a start picture |
Be clear about this
What we do not know for you.
- Pre-betaWhether a real model is on in the hosted app
It needs a key on the server. The models list in the app says what is ready where you are.
Questions
About models.
Are the videos made by an AI model?
Not by default today. The default models are Previews that paint clips in your browser from your prompt and a seed; they are not video models. Google Veo 3.1 and Kling 2.1 are connected in code and work only where their keys are set on the server.
Will it make what I describe?
The Previews will not: they paint a scene that leans with your words. A real model can, where one is switched on.
Where does my prompt go?
With the Previews, it is used in your browser. If a real model is on, your prompt and any start picture go to that model's provider.