AI Video and video models
AI Video creates video clips from text, images, video and audio references. You can choose from several AI video models.
Two ways to create
- Generate Video — write a prompt. Optionally add reference images (characters, products, places), reference videos (motion or camera movement) or audio. Without references, the model makes the video from text alone.
- Frame to Video — upload a start image and describe what should happen. Some models also accept an end frame, so the video moves from one picture to the other.
How many references you can add, and which file types, depends on the model. The upload area shows the limits for the model you’ve selected.
Choosing a model
Models differ in maximum length, resolution, reference support and price. A quick guide:
| Model | Good to know |
|---|---|
| Seedance 2.0 Mini | The default. Low cost, 480p or 720p, 4–15 seconds. Great for trying ideas. |
| Seedance 2.0 / 2.5 | Up to 1080p with rich reference support. Seedance 2.5 makes clips up to 30 seconds. |
| Wan 3.0 | 2–30 seconds, up to 1080p, and accepts many image, video and audio references. |
| Kling 3.0 | Up to 4K, 3–15 seconds, with optional audio. |
| Veo 3.1 / Veo 3.1 Fast | 4, 6 or 8 second clips up to 4K with generated audio. Fast is the lower-cost version. |
| Vidu Q3, MiniMax H3 Max, Gemini Omni Flash | Alternatives with their own strengths; Vidu Q3 accepts photos of people that some other models reject. |
| Wan 2.2 Turbo | Fixed 5-second clips at a low price. |
The model list changes as new models are added. The app always shows the current models, their options and prices.
What affects the price
Most video models charge per second of output, so longer and higher-resolution videos cost more. On some models, turning on audio also changes the price. The exact cost is shown before you press Generate. Some models occasionally run limited-time discounts, shown as a sale price in the model list.
AI Director (Beta)
If you’d rather describe an idea than write a prompt, open AI Director from the menu. It asks a few questions, plans the video, and fills in the prompt, model and settings for you. Chatting doesn’t generate anything; you’re charged only when you choose to generate the video.
Photos of real people: some models, such as Seedance and Gemini Omni, may refuse uploaded photos of real people. Use a Character, an image created with AI Image, or a model such as Vidu Q3 or Wan 2.2 Turbo.