Write video prompts

A video prompt is not an image prompt that moves. The first frame already sets the look, so the prompt's job is to say what changes, how the camera follows it and what it sounds like. This lesson animates one approved image, a girl and her chameleon in a sunny kitchen, with Seedance 2.5: first as one shot written in six parts, then as a 15-second shot with four timed beats, then as three shots joined by cuts.

Before you start

• An approved first frame. Make and approve the image before you animate it. The video keeps its characters, setting, light and style, so fix any problem in the image, not in the video prompt.
• A model that takes a first frame. The film uses Seedance 2.5 in Create › Videos. Seedance 2.5 picks its mode from what you attach: a prompt alone, a first frame, a first and last frame, or references. A first frame and reference images can't be used together, so choose one anchor per shot.
• Draft runs. With Draft on in Settings, Seedance 2.5 renders a 480p preview. When a preview is the one you want, you finalize it in 1080p without generating it again.
• What a video prompt holds. Seedance's own format has slots for subject and action, scene, visual style, camera movement and audio. With a first frame, the scene and style are already in the image, which is why the prompt here spends its words on the action, the camera and the sound.

Step 1: Start from the first frame

Start from an image you've already approved, and use it as the First Frame. It already shows the girl, her chameleon, the kitchen and the light. So the prompt doesn't describe the look again: it describes what changes.

Describing the look a second time does not help. At best it repeats what the frame shows; at worst it asks for something slightly different, and the model drifts away from the frame to match the words.

Step 2: Write the shot in six parts

Write the shot in six parts, in this order:

1. Start state: where she is when the clip begins. She stands still, looking at the chameleon.
2. One main action: she picks up the lemon and holds it out.
3. End state: where the action lands. She ends holding the lemon up, smiling. Without it, the model may repeat the action, or add a new one.
4. Camera: the shot size, and one move. Here a medium-long shot at table height, with a slow push-in.
5. Pace: gentle, real-time.
6. Sound: what you hear in the room. Write "no music" if you'll add your own later.

The prompt the film uses:

The girl with the long white bubble braid stands still beside the kitchen table, looking at her green chameleon on the edge of the fruit bowl. She steps up to the table, picks up the lemon with both hands and holds it out to the chameleon, which slowly turns its head and both eyes toward the lemon. She ends standing at the table, holding the lemon up in front of the chameleon, smiling. Medium-long shot at table height, slow push-in toward the girl and the chameleon. Gentle, real-time pace. Sound: soft kitchen room tone, birds outside the window, light footsteps on the tiles. No music, no dialogue.

Name the camera move exactly ("slow push-in", "slow arc to the right") rather than with a vague word like "cinematic". Describe motion over time, not a frozen picture.

Step 3: Set the duration and generate

Open Settings and choose a Duration that fits the action. If it's too long, the model fills the time with extra moves. One action with a clear end state needs only a few seconds; the film uses 8s.

The other settings in the film:

• Resolution: 480p. While Draft is on, the output is always 480p, whatever this says.
• Aspect Ratio: Auto. With a first frame, the clip keeps the shape of the image.
• Generate Audio: on, so the model makes the sound the prompt describes.
• Output Format: MP4. MOV keeps more color for grading, in larger files.
• Draft: on, for a quick 480p preview.

The price shows on the Generate button. Click Generate.

She walks to the table, picks up the lemon, and stops on the end state, while the camera pushes in.

The banner above a draft shows how long you have left to finalize it. Click Finalize in 1080p to open Seedance 2.5 Draft Complete with the draft already loaded, then click Generate for the 1080p version. A draft can be finalized for 7 days.

Step 4: Time the beats of a longer shot

For a longer shot, split the action into beats, and give each beat its time. One action per beat: the lemon, the tongue, the color change, and the smile. Start each beat on its own line with its time in brackets, for example [0-4s].

Keep one camera move for the whole shot. Here, a slow arc around the table.

The prompt in the film, as far as its sound line (the end of the prompt is not shown in the film):

One continuous shot, no cuts. The girl with the long white bubble braid stands beside the kitchen table, looking at her green chameleon on the edge of the fruit bowl.
[0-4s] She steps up to the table, picks up the lemon with both hands and holds it out to the chameleon.
[4-8s] The chameleon shoots out its long sticky tongue, grabs a green grape from the bowl and pulls it into its mouth. She laughs, surprised.
[8-12s] She sets the lemon down right next to the chameleon. The chameleon's skin slowly changes color from green to the same bright lemon yellow, starting at its head.
[12-15s] She leans in with her chin on her hands, smiling at the yellow chameleon, and it blinks at her.
Camera: medium-long shot at table height. The camera slowly arcs to the right around the table and moves a little closer during the whole shot, ending on a medium shot of the girl and the chameleon.
Pace: gentle, real-time.
Sound: soft kitchen room tone, birds outside the window, a tiny wet snap when the tongue catches the grape.

Set the duration to the end of the last beat: here 15s. Click Generate.

Watch the result beat by beat: each action should land in its own window while the camera keeps arcing.

Step 5: Ask for cuts

You can also ask for cuts, so one clip holds several shots:

• Write each shot on its own line, with its time, its framing and its camera.
• Start each new shot with "Cut to".
• Say that the girl, the chameleon and the kitchen stay the same in every shot. Otherwise the model may treat each shot as a new scene.

The film's clip has three shots in 12 seconds: a medium-long static shot (0 to 4 s, she picks up the lemon), an extreme close-up (4 to 8 s, the chameleon turns yellow) and a close-up with a slow push-in (8 to 12 s, her smile). Its prompt opens like this:

Three shots with hard cuts. The same girl, the same chameleon, the same kitchen and the same sunlight in every shot.
Shot 1 [0-4s]: medium-long shot at table height, static camera. The girl with the long white bubble braid stands beside the table, then steps up and picks up the lemon from the fruit bowl.
Cut to shot 2 [4-8s]: extreme close-up of the green chameleon on the edge

The film shows only this part of the prompt. Shot 3 follows the same pattern, starting with "Cut to shot 3".

Set the duration to the end of the last shot, and click Generate.

What to avoid

• A prompt that only describes the scene gives a clip where almost nothing happens. Use active verbs, and say what moves.
• Several camera moves in one clip blend into one. For several moves, make several shots.
• Writing what should not happen. Write what should happen, not what shouldn't: "doesn't turn around" can make her turn. Write "She keeps facing the table." instead.

Sound in the prompt

With Generate Audio on, the model makes the sound from your words. Describe the room tone, the small sounds tied to the action, and say plainly what you don't want: "No music, no dialogue."

Seedance 2.5 also reads the developer's bracket channels for sound: ( ) for music and ambience, < > for sound effects, { } for spoken dialogue, and 【 】 for on-screen subtitles. For dialogue, name the language first. The Knowledge Center recommends keeping Generate Audio off and adding your own cleared track when the sound matters: the generated soundtrack is checked by moderation after the video is made, so a flagged track can block a finished clip.

The shape of each prompt

Common problems

What's next

To move the camera with more control, read Direct the camera and motion. To fix both the start and the end of a shot, use a first and a last frame in Generate video, and to keep a character the same across clips, see Reference to video.