Prompts for Seedance: how to write requests that get the right shot

The quality of Seedance video depends directly on the text prompt. The model understands plain language, but results are noticeably better when you describe not only "what happens" but also "how it is shot": who is in frame, what the motion is, what the lighting and camera are. Below is a working prompt structure and ready templates.

The structure of a good prompt

Start with the action: who is in the shot, what happens and how the scene ends. Then set the location, lighting, mood and style. After that, describe the camera — wide or close-up, locked shot, push-in or pan. This order helps the model build the scene logically instead of assembling it from random details.

For a multi-beat clip, a short timeline helps: "0–4 seconds — the character enters, 4–8 — picks up the object, 8–12 — the camera pushes closer." Do not pack a long plot into four seconds: a short clip physically fits one or two key beats.

Templates for common tasks

Product clip: "Close-up: a [product] stands on a dark table, the camera slowly orbits it, soft studio light, light haze, premium product render." Landscape: "Wide shot: [place] at dawn, the camera slowly moves forward, warm golden light, light mist." Portrait in motion: "Medium shot: [character] walks through [place], the camera follows from the side, natural daylight, cinematic color grade."

Use references instead of long descriptions

If you need to keep a specific character, product or style, do not describe it in words — upload an image or clip and mention @Image1 or @Video1 in the prompt. The model takes the look and motion from the file, and the text stays short and precise: "@Image1 — the character, @Video1 — the motion style, scene: the character enters a cafe and sits by the window."

Common mistakes

A plot that is too long for a short clip — the model tries to fit everything in and loses detail. Contradictory instructions ("day" and "night" in one prompt) — the result becomes random. No camera description — the model picks the shot itself, and it does not always match your idea. Keep the prompt specific and check the result on the Seedance 2.5 page.

FAQ

What language should I write the prompt in?

The model understands Russian and English. Write in the language where you can describe the scene most precisely.

How many words should a prompt have?

20–60 words is usually enough. A short, specific prompt works better than a long list of details.

Can I set a shooting style?

Yes: specify "cinematic", "documentary", "animation" or a concrete visual style — the model will apply it during generation.