Guides
How to write AI video prompts
Write an AI video prompt in five parts: who or what, what they do, where, how the camera moves, and what we hear. Keep one main action per clip, use concrete nouns and verbs, and for image to video describe only the change, since the model already sees the photo. Our studio takes AI video prompts of up to 2,000 characters; one to three sentences is usually enough.
Updated
All guidesPhotos
0 / 9One photo starts the clip. With two or more, the model takes the people and things from them. Up to 9 photos, JPG, PNG or WebP.
0 / 2,000
Model
Why we ask
Putting a real person into a video without their OK can break privacy and image-rights laws, and that responsibility is yours. We keep the time you confirmed with the request.
Real examples
Seedance 2.5
The five parts of an AI video prompt
| Part | Ask yourself | Example words |
|---|---|---|
| Subject | Who or what is on screen? | a grey cat, two friends, a red vintage car |
| Action | What changes during the clip? | jumps onto the table, turns and waves, drives past |
| Setting | Where and when? | a rainy Tokyo street at night, a kitchen at sunrise |
| Camera | How do we watch it? | slow push-in, handheld, tracking shot, top-down |
| Sound | What do we hear? | rain on the window, crowd cheering, no music |
Text prompts and image prompts are different
In text to video, the AI video prompt is all the model has, so it must describe the look as well: the people, clothes, place, light and style.
In image to video, the photo already sets the look. Spend the words of the AI video prompt on motion and camera. Repeating what is in the photo can even pull the model away from it.
With several photos on Seedance 2.5, Seedance 2.0 and Wan 3.0, say who is who: "the man from photo 1 hands a gift to the woman from photo 2".
Camera words for AI video prompts
- Push-in / pull-back: the camera moves toward or away from the subject.
- Pan left or right: the camera turns on the spot.
- Tracking shot: the camera moves alongside the subject.
- Orbit: the camera circles the subject. See the drone shot prompts.
- Static shot: nothing moves but the subject. Useful when faces must stay stable.
AI video prompts before and after
Before A beautiful cinematic video of a girl, amazing, 4k, trending.
After A young woman in a yellow raincoat stops on a wet street, looks up and smiles as the rain stops. Slow push-in, evening streetlights, sound of dripping water.
Before (image to video) A man in a blue shirt standing on a beach with the sea behind him.
After (image to video) He turns toward the sea and walks into the shallow waves, the camera follows behind at shoulder height, wind and waves.
Common mistakes in AI video prompts
- Three actions in one short AI video prompt. Pick one; make a second clip for the next.
- Quality words instead of content: "masterpiece, 8k, ultra" tell the model nothing about what to show.
- Negations alone: "no people" can still add people. Describe what should be there instead.
- Text on screen: models often misspell it. Add titles in an editor afterwards.
- Public figures by name: the terms do not allow videos of real people without their consent.
Match the AI video prompt to the length
Clips run 2 to 30 s depending on the model. A 5 s clip holds one action and one camera move. Over 10 s you can describe a short sequence ("walks to the window, then turns around"), but keep the order clear with "then".
Try Capafy AI with one photo
Open the studioFAQ
Questions and answers
How long should an AI video prompt be?
One to three sentences, about 20 to 60 words, covers the five parts of an AI video prompt. The studio accepts up to 2,000 characters.
Should I write AI video prompts in English?
English usually works best, but the models also read other major languages. If a result ignores part of the prompt, try the English version.
Can I reuse the same AI video prompt on different models?
Yes, and it is a good way to compare them. Keep the photo and prompt the same and change only the model.