How to write Seedance 2.5 prompts: numbered shots, camera notes and a Sound line, with a reusable template, three example prompts and the common mistakes.
Seedance 2.5, made by ByteDance, renders up to 30 seconds in one generation and rewards prompts written like a shot list. It reads numbered shots, camera notes and a line about sound, and it can hold several cuts inside one clip. This guide covers how the model reads a prompt, a template you can reuse, three example prompts with the technique behind each, and the mistakes that cost you takes.
The prompt is the main driver. With a prompt alone, Seedance 2.5 works as text-to-video. You can add an optional first frame, an optional last frame, or both, to pin where the shot starts and ends. For recurring characters, props or locations, Seedance 2.5 References takes up to 30 images, 10 videos and 10 audio clips. Frames and references cannot be combined in one generation, so decide which one drives the shot. Seedance 2.5 Live Action is a separate lane for real-looking AI characters built from up to 9 reference images.
Clips run from 4 to 30 seconds at 480p, 720p or 1080p, in ratios from 21:9 wide to 9:16 vertical, or adaptive. Sound is optional. When it is on, describe it in the prompt like any other part of the shot.
ByteDance documents multi-shot prompts for this model. One prompt can hold several shots, written either as "Shot N" headings or as a continuous timeline in whole seconds with no gaps (0-3s, 3-8s and so on). A single continuous take is one block with neither. Cinemagiq also rewrites Seedance prompts automatically into the structure ByteDance recommends, so you can focus on what happens in each shot.
For a multi-shot clip:
Shot 1: [shot size] of [subject + one concrete detail] [action], [camera move]. Shot 2: [shot size] of [subject] [next action], [camera move]. Shot 3: [shot size] of [detail or reaction]. [Style and light]. Sound: [ambience], [effects], [music or "no music"].
For a single continuous take, drop the headings:
[Camera move] [shot size] of [subject] [action], [lens or depth], [light], [style]. Sound: [ambience and effects], [music or "no music"].
Keep one idea per shot. Every heading should change something a viewer can see: the framing, the subject or the camera.
Made in Cinemagiq. The exact prompt is under each one.
PromptShot 1: extreme wide shot of a red hot-air balloon rising over a green valley at sunrise. Shot 2: medium shot inside the basket as the pilot opens the burner and a flame roars. Shot 3: close-up of the balloon fabric filling the frame. Sound: wind and the burner, no music.
PromptHandheld tracking shot following a teenage runner sprinting through a crowded night market, food stalls and lanterns whipping past, shallow depth of field. Sound: crowd chatter, sizzling food and footsteps, no music.
Prompt2D anime style: a girl rides her bicycle down a seaside hill in summer, white clouds, the camera tracks alongside her as her hair blows in the wind. Sound: cicadas, wind and the bicycle chain, no music.
Example 1 fits three numbered beats into a 5-second clip: "Shot 1: extreme wide shot", then "Shot 2: medium shot inside the basket", then "Shot 3: close-up". Each heading changes the shot size, which is what makes a cut read as a cut. The closing "Sound: wind and the burner, no music" names two effects and rules out a score.
Example 2 is a single continuous take, so it has no headings at all. It leads with the camera, "Handheld tracking shot following", then adds the lens cue "shallow depth of field". The Sound line layers three sources ("crowd chatter, sizzling food and footsteps") to build the space without music.
Example 3 opens with the style, "2D anime style:", so the look is set before any subject appears. The camera note, "the camera tracks alongside her", sits inside the action, and the Sound line ("cicadas, wind and the bicycle chain") picks sounds that place the scene in summer.
For the shot vocabulary behind these prompts, see the prompt engineering cheat sheet.
Use numbered headings such as Shot 1, Shot 2 and Shot 3, or a continuous timeline in whole seconds with no gaps, such as 0-3s and 3-8s. Pick one format per prompt, and give each shot its own framing and action so the cuts read clearly.
Seedance 2.5 renders from 4 to 30 seconds in one generation, at 480p, 720p or 1080p. A longer clip leaves room for more shots inside one prompt, so match the number of beats you write to the length you set.
Yes, sound is optional on Seedance 2.5. When it is on, end the prompt with a Sound line that names the ambience and effects you want, and add 'no music' if you plan to score the scene later. Switch it off when the clip will be mixed elsewhere.
Yes, through Seedance 2.5 References, which takes up to 30 images, 10 videos and 10 audio clips. References cannot be combined with a first or last frame, so choose one approach per shot. Seedance 2.5 Live Action keeps real-looking characters from up to 9 reference images.
Run Seedance 2.5 and the other leading models in one AI-native workspace.