# Seedance Prompting Guide (2.0 and 2.5)

A practical guide to writing prompts for ByteDance's Seedance video models. It covers how to structure a prompt for **Seedance 2.0**, the timeline-based approach that works best in **Seedance 2.5**, how to write edit and extension prompts, and a ready-made instruction block you can give an AI assistant so it writes Seedance prompts for you.

## Which version and which workflow?

| You have… | Seedance 2.0 | Seedance 2.5 |
| --- | --- | --- |
| Only an idea | Text-to-video | Unlocked generation (text plus optional references) |
| A still image that already sets the look | Image-to-video | First/last-frame or keyframe generation (locked) |
| Assets whose look, motion, style or audio must stay consistent | Reference-to-video | Reference-driven generation with asset mapping |
| A clip you want to change or lengthen | Not supported | Instruction-based editing or extension (locked) |

**Unlocked vs. locked (2.5):** in an *unlocked* task, references, storyboards or keyframes guide the result, and the output's length and aspect ratio stay free. In a *locked* task (editing, first/last-frame generation, extension), the input pins the timeline or aspect ratio. Decide which one you're doing before you write anything.

## Principles that apply to both versions

- **Lead with the most important subject.**
- **Make action observable:** direction, pace, sequence and physical reaction, not just "she dances".
- **Describe light physically:** name the source, direction and texture instead of generic words like "cinematic" or "beautiful".
- **Direct the camera:** shot size, angle and movement, for example *slow push-in*, *lateral tracking shot*, *low-angle tracking* or *dutch angle*. If you use an uncommon camera term, describe what it should look like on screen.
- **Plan the sound:** ambience, sound effects, music and dialogue, stated on purpose.
- **Use negatives sparingly:** only for constraints that really matter.
- **Give every asset one job:** *"Image 1 controls the character's appearance. Video 1 supplies motion and camera rhythm only. Audio 1 supplies the musical rhythm only."* Say what must **not** transfer when that prevents a conflict.
- **Change one thing at a time** between generations, so you can tell what helped and what hurt.

## Seedance 2.0: the layered prompt

Write in this order, skipping what doesn't apply:

`Subject → action and motion → environment → visual style and lighting → camera → audio → essential exclusions`

Every prompt should include an opening shot, the camera movement or framing during the action, and a final framing. Name the visual treatment and lighting on purpose.

**Image-to-video:** don't re-describe the image you uploaded. Describe how it comes alive, in layers:

1. Foreground movement (rain, dust, reflections).
2. Main-subject motion and micro-expressions.
3. Background movement that adds depth.
4. Camera behavior.

If the image is an illustration, watercolor, animation or 3D render, say explicitly to keep that medium, or the style may drift.

```text
[Subject] [performs a clear action, with pace and sequence] in [environment].
[Foreground / subject / background motion].
[Visual medium, atmosphere, palette, and lighting source].
[Starting shot, camera movement, final framing].
[Ambience, sound effects, music, and/or dialogue].
[Only essential constraints].
```

> **My working standard:** I keep 2.0 prompts between about 1,950 and 1,999 characters (counting spaces and line breaks) and trim for density rather than going over. Dense, specific prompts give noticeably more controllable results.

## Seedance 2.5: the four-part timeline prompt

`Asset mapping → one-sentence creative brief → continuous timeline → global constraints`

1. **Asset mapping:** list every image, video and audio file in upload order, and give each one a subject and a single function.
2. **Creative brief:** one sentence covering subject, location, event, genre or style, and any special camera treatment.
3. **Timeline:** continuous whole-second ranges such as `0–3 seconds`, `3–7 seconds`, `7–15 seconds`, each with its action, camera, dialogue, sound and transition.
4. **Global constraints:** anything that holds for the whole clip, such as camera position, lighting, palette, ambience, aspect ratio, subtitles, music, on-screen text or watermark rules.

Match how much story you put in to the clip's length. A 5-second clip can hold one beat, not three.

```text
Asset mapping:
Image 1: [subject / appearance only].
Video 1: [motion and camera reference only].
Audio 1: [voice / dialogue / melody / rhythm only].

Creative brief: [subject] in [location] [event], [style / lighting / camera treatment].

0–[n] seconds: [shot, action, camera, sound].
[n]–[n] seconds: [next shot, action, camera, sound].

Global constraints: [continuity, aspect ratio, lighting, color, subtitles, music, text, watermark, exclusions].
```

### Editing and extending clips (2.5)

Pattern: `Target → exact change → protected properties`

```text
Edit Video 1, seconds [start–end]. Change only [target] from [current state] to [desired state]; [say whether the change is instant, gradual, or continuous]. Preserve [identity, performance, background, camera, timing, lighting, sound, and all other named properties]. Keep [unaffected time ranges] unchanged.
```

With several clips, start with `Edit Video 1` or `Extend Video 2`. For an extension, say whether to extend forward or backward, and by how many seconds.

### Preflight checklist

- Is the task locked or unlocked?
- Does every asset have a number, a subject and one clear responsibility?
- Are the time ranges continuous and realistic for the clip length?
- Are action, sound and transition cues timed where timing matters?
- Are the global constraints stated explicitly?

## Timestamped shot blueprint

For short, punchy clips, break the sequence into 2- to 4-second beats, and give each beat one distinct camera move and clear lighting. For a 16-second video, two connected 8-second generations are often more reliable, and cheaper, than one long one.

```text
[Shots: X | Duration: Ys | Aspect Ratio: W:H]
- 0:00–0:02 (Hook): [camera movement] + [exact subject action] + [lighting and ambient detail] + [reference tag]
- 0:02–0:05 (Build): [camera shift] + [action escalating] + [interaction with the environment]
- 0:05–0:08 (Climax / transformation): [dynamic camera movement] + [visual transformation]
- Audio intent: [ambient sound, foley, and score direction]
```

Use your tool's reference syntax (for example `@material` / `@reference` tags, or "Image 1 / Video 1") to anchor character identity, opening frames or product images.

## Let an AI assistant write the prompts

Paste this into a Claude project, custom GPT or system prompt, then describe your video idea in plain English.

```text
You are an expert AI video director for ByteDance's Seedance models. Turn the user's
plain-English video idea into a structured, frame-accurate Seedance prompt.

Rules:
1. Start every prompt with a global line: [Shots: X | Duration: Ys | Aspect Ratio: W:H].
2. Split the sequence into 2–4 second timestamped beats. For 16-second videos, use two
   connected 8-second segments.
3. Anchor characters, first frames, and products with the user's asset references
   (@reference tags or "Image 1 / Video 1"), and give each asset exactly one job.
4. Use explicit camera language (slow push-in, lateral tracking, low-angle, dutch angle)
   and physical lighting (source, direction, texture) — never vague quality words.
5. For Seedance 2.5, structure the prompt as: asset mapping → one-sentence creative
   brief → continuous whole-second timeline → global constraints.
6. End with an audio line: ambience, foley, music, and any dialogue.

Workflow:
- If the idea is too vague, ask one focused question (the hook, the subject, or which
  assets they have) before writing.
- Check that every beat has one distinct camera move and clear lighting.
- Return the finished prompt in a code block so it's easy to copy.
```

## Sources

- [Seedance 2.0 Complete Prompting Guide](https://docs.seedance.tv/en/seedance-2-complete-prompting-guide)
- [Seedance 2.5 Prompting Guide, Part 1](https://docs.seedance.tv/en/seedance-2-5-prompting-guide)
- [Seedance 2.5 Prompting Guide, Part 2: video editing](https://docs.seedance.tv/en/seedance-2-5-video-editing-guide)

*Sources consulted August 2026.*