# Seedance Prompt Studio

> Turn one video idea into a production-ready Seedance 2.0 prompt. Six lanes over one brief -
> plan, cast, storyboard, prompt, long-form segmentation and audit - plus a free bilingual prompt
> checker that runs entirely in the browser.

URL: https://seedance-prompt-studio.skillsafe.ai/

## What this is

A web app for writing prompts for **Seedance 2.0**, ByteDance/Jimeng's multimodal video generation
model. It does not generate video and it does not generate images. It writes the plans, tables and
prompts you take to Seedance yourself, and it checks them.

Not affiliated with or endorsed by ByteDance, Jimeng or Seedance.

## The free part

The prompt checker needs no account, spends no credits, and runs entirely in your browser. Paste a
Seedance prompt and it reports, in English or Chinese:

- **Budget** - more than 9 images, 3 videos, 3 audio files, or 12 files combined.
- **Duration** - outside the 4-15 second single-generation window.
- **@ reference resolution** - a reference to an asset that was never uploaded.
- **Vague references** - `reference @Video1` without saying what to take from it. The single most
  common defect in real Seedance prompts.
- **Unused assets** - an upload the prompt never mentions, which still costs you a file slot.
- **Camera conflicts** - a locked-off camera and an orbit in the same timed window.
- **Timed-window coverage** - gaps, overlaps, a breakdown that starts late, runs past the clip, or
  stops before the end.
- **Overloading** - more action beats than the seconds can physically hold.
- **Empty praise** - beautiful, stunning, epic, 唯美, 震撼: words a video model cannot act on.
- **Face risk** - a prompt instructing an upload of a real human face, which the platform refuses.
- **Missing audio direction** and **missing prohibition line**.

It understands both `@Image1`/`@Video1`/`@Audio1` and `@图片1`/`@视频1`/`@音频1`, including the
short `@图1` form that appears in the wild.

## The six lanes

Each takes the same work object - one video idea - and hands its result to the next.

| Lane | `task` | What it produces |
| --- | --- | --- |
| Video plan | `plan` | Scene-by-scene plan, every scene sized to one generation, plus a continuity bible |
| Character design | `cast` | Identity lock table and a character sheet prompt for an image model |
| Storyboard package | `board` | Panel grid sized by duration, panel table with timings, storyboard sheet prompt |
| Seedance 2.0 prompt | `prompt` | The prompt itself, with every @ reference given an explicit role |
| Long-form segmentation | `extend` | Past the 15s ceiling: segments, per-segment prompts, splice points |
| Prompt audit | `audit` | Findings against the named mistakes, and a corrected rewrite |

## Key Seedance 2.0 facts this app enforces

- One generation produces **4 to 15 seconds**. There is no way to exceed 15s in a single pass;
  longer pieces are spliced with the video-extension capability.
- Inputs: at most **9 images**, **3 videos** (2-15s total), **3 audio files** (15s total), and at
  most **12 files combined**.
- Sound is generated **with** the picture. Undirected audio is invented audio.
- Uploads containing **realistic human faces are refused** by the platform.
- Two entry points: first/last frame (one image plus a prompt) and omni reference (mixed image,
  video and audio). **@ references only resolve in omni reference.**
- When extending a video, the generation length you select is the length of the **new part**, not
  the running total. This is the most common way a long piece fails.

## Sources

Lanes are derived from open-source Agent Skills on GitHub:

- https://github.com/dexhunter/seedance2-skill - the @ reference system, camera vocabulary,
  capability patterns and the common-mistakes list.
- https://github.com/songguoxs/seedance-prompt-skill - the ten capabilities, timestamp
  segmentation, the long-form splicing strategy and the style vocabulary.
- https://github.com/agentara/skills/tree/main/skills/aigc - video-plan, video-storyboard and
  video-character-design.

## API

Full docs with worked examples in eight languages: https://seedance-prompt-studio.skillsafe.ai/api.html

Base URL `https://api.skillsafe.ai/v1/app-api`. Bearer token from
https://seedance-prompt-studio.skillsafe.ai/tokens.html. Every run body must carry a `task` field
naming the lane; the body **is** the input object and must not be wrapped in `{"input": ...}`.
`GET /me` and `POST /estimate` are free; `POST /run` and `POST /run-stream` are metered.

Output is Markdown with level-two section headings in a declared order, GitHub pipe tables with
declared columns, prompts in fenced `text` blocks, and `- [x]` checklists.

## Pricing

Runs are metered and priced per model - `gpt-5.6-luna`, `gpt-5.6-terra` (default) or
`gpt-5.6-sol`. `POST /estimate` returns the worst-case reservation for free before anything is
spent; the actual charge is usually far lower. The prompt checker and every lane's worked example
cost nothing.