Seedance 2.0 Prompt Guide

One moment, not a sequence

A Seedance clip is one shot. Seedance 2.0 Mini, 2.0 Fast and 2.0 run from 4 to 15 seconds. Seedance 2.5, the flagship, runs from 4 to 30 seconds. That is not enough time for a scene change, a second location or a cut. If the prompt says "then," it is asking for two shots.

Write the prompt the way you would brief a camera operator. Name the subject, the single action, the light, the setting and one camera move. Put "no text" in the prompt when you do not want lettering. Video models often paint unreadable signs and labels if you leave that open.

• Subject and action: what is on screen and exactly how it moves
• Camera: locked, slow push-in, tracking, orbit, or a drone glide
• Light and setting: time of day, light direction, and the room or landscape
• Constraints: what must not change, plus "no text" when lettering would hurt the shot

Which Seedance version to prompt

The versions share the same prompt style, so a shot that works on a draft model can move to the flagship without a rewrite. Draft on Seedance 2.0 Fast or Mini at 480p or 720p while you are still testing motion. Move to Seedance 2.0 for 1080p and image, video and audio references. Use Seedance 2.5 when the shot needs to run longer than 15 seconds.

Aspect ratio is part of the brief. Use 16:9 for web and YouTube, 9:16 for Reels, TikTok and Shorts, 1:1 or 4:3 and 3:4 for feeds and product pages, and 21:9 for a wide cinematic frame. Resolution and duration both change the credit cost, and the studio shows that cost before you generate.

A text-to-video prompt

Text to video has to invent the whole frame, so the prompt carries the subject as well as the motion. Keep it to one action that is satisfying to watch for a few seconds.

This is the structure behind the latte pour in the gallery, generated with Seedance 2.5 at 4 seconds and 720p: "Slow-motion close-up of milk being poured into a cup of black espresso, forming a swirling latte art rosetta. Warm café light, a dark ceramic cup on a wooden counter, blurred coffee bar in the background, gentle steam. Static camera, smooth motion. No text."

The same shape works for a landscape. The sea-cliffs clip names one camera move, a slow forward drone glide, then stacks layers the camera can travel past: grass on the cliff top, dark rocks, waves, and low mist under a sunrise. One move, several depths.

An image-to-video prompt

When you already have a still, do not spend the prompt redescribing it. The image is frame one. Write what changes over time and what must stay fixed. Large motion and a moving camera are what make a product label, a face or a bottle shape drift.

The perfume clip in the gallery starts from a finished splash still. The bottle stays fixed and sharp, the framing stays the same, and only the water and the light move: ripples spread, the splash settles in slow motion, droplets fall, and a streak of light crosses the glass. The camera is locked. That is the right amount of motion for a product ad.

For a product photo, say the product must stay sharp and the framing unchanged, then name one physical change: liquid, fabric, steam, a reflection, or a slow light sweep. If the still has no empty space around the product, the motion has nowhere to go. Fix the still before you animate it.

• Keep: product shape, label, color, and framing
• Move: one material or one light change
• Camera: locked, unless the product can survive a very slow push-in

References, when the shot needs them

Seedance 2.0 and 2.5 can take references in one request: up to nine images, up to three video clips, and up to three audio clips. Use an image reference to hold a product or a character, a clip to carry motion you already like, and audio when the cut should follow a piece of music. References do not replace the prompt. You still need one sentence that says what the new shot does.

What to change when the shot is wrong

Change one thing at a time. If the product warps, shorten the motion and lock the camera before you rewrite the scene. If the clip looks like two shots, delete the second action. If signs and labels are gibberish, add "no text" and, for a product, describe the label as a locked detail rather than asking the model to invent lettering.

Generate the still you want first, with an image model, and only then send it to image to video. A good first frame fixes more problems than a longer prompt.

How do I write a Seedance 2.0 prompt?

Describe one continuous shot: the subject, a single action, the light and setting, one camera move, and what must stay still. Add "no text" when you do not want lettering. Avoid a sequence of scenes.

What is the difference between Seedance 2.0 and 2.5 for prompting?

The prompt style is the same. Seedance 2.0 makes clips from 4 to 15 seconds. Seedance 2.5 extends that to 30 seconds and is the flagship model. Draft on 2.0 Fast or Mini, then move the same prompt to 2.0 or 2.5.

How do I animate a product photo with Seedance?

Use image to video. Tell the model to keep the product sharp and the framing unchanged, then describe one motion, such as a splash settling or a light sweep. Keep the camera locked.

Which aspect ratio should I use?

16:9 for YouTube and web, 9:16 for Reels, TikTok and Shorts, 1:1 or 3:4 for feeds and product pages, and 21:9 for a wide frame. Seedance also supports 4:3.

Can I use Seedance without an API key?

Yes. In MiniMaxVion, open text to video or image to video and choose a Seedance model. Generation uses credits from your balance. New accounts get free credits.