seedance-prompt-short
GitHub用于压缩Seedance 2.0提示词,在保留关键制作信号和连续性的前提下精简文本。需加载导演阅读记录,按优先级保留参考标签、主体身份及动作,去除填充词,确保输出简洁且符合审核标准。
Trigger Scenarios
Install
npx skills add Emily2040/seedance-2.0 --skill seedance-prompt-short -g -y
SKILL.md
Frontmatter
{
"name": "seedance-prompt-short",
"tags": [
"prompt-compression",
"chinese-prompt",
"seedance-20"
],
"license": "MIT",
"metadata": {
"author": "Iamemily2050 (@iamemily2050)",
"parent": "seedance-20",
"updated": "2026-08-01",
"version": "6.7.0",
"openclaw": {
"emoji": "🎬",
"homepage": "https:\/\/github.com\/Emily2040\/seedance-2.0"
},
"repository": "https:\/\/github.com\/Emily2040\/seedance-2.0"
},
"description": "This skill should be used when the user asks for a compact Seedance 2.0 prompt, short Chinese prompt, prompt compression, 30-100 word output, or removal of unnecessary prompt language.",
"user-invocable": true
}
seedance-prompt-short
Before producing prompt text, a prompt-ready block, a rewrite, an example, or a compiled clip, load the Director's Read, classify the brief, and complete its canonical narrative or non-narrative record. Translate that record into visible or audible carriers and keep its internal labels out of final generation prose.
Compress Seedance prompts without losing the production signal. A short prompt still needs mode, subject, action, camera, light, sound when useful, and constraints. Remove filler before removing physical details.
Load the Director's Read before compression, even when the requested output is only one line. Briefs with a dramatic story turn complete its ten-field narrative record first. Observation/performance without that turn and utility briefs use its two-line non-narrative intent and refusal without fabricated drama; short output must still preserve the requested expression or rhythm. Short means compressed output, not skipped judgment.
When sequence state is present, compression must preserve continuity locks, exact reference tags, actual opening state, current clip action, endpoint, completed beat exclusions, and reserved future beats. Do not compress away the words that keep a continuation from replaying completed action or leaking future action.
The moderation pre-screen runs on the compressed prompt too: compression that drops a clarifying word can expose a cue, and a short prompt is refused as readily as a long one. Never write an involuntary outcome as the endpoint; the model stages accidents as intentions. The short route still builds the shot table for any storyboard or multi-person clip and may deliver the prompt alone, but a cell it cannot fill is a question it must ask before delivering. Compression never removes a lock line, a stated position or a feeling word, never turns a literal action back into an idiom, and never replaces idle business with a freeze; see direct-for-the-model.
Storyboard clips keep their structure when compressed: numbered shots in event order, the cut written in words, no absolute seconds inside a block on Seedance 2.0. The shot-count ladder from multishot-grammar shrinks to one line per rung and one finished prompt at the chosen rung; when the user has fixed the count, state its rung in one line and write it.
Intent
Compression is an act of judgment about what the user loves most. What survives the cut is the soul of their shot; everything else goes first. If the user would mourn a deleted word, it was never filler.
Compression Priority
Preserve in this order:
- Reference tags and their role.
- Subject or product identity.
- Action verb and visible endpoint.
- For narrative work, the visible suppressed behavior and non-transferable detail that carry the Director's Read.
- One camera move.
- Physical light source or atmosphere.
- Audio cue or silence instruction.
- Safety, IP, or continuity constraint.
- Sequence state clauses: actual opening state, continuity locks, completed beats, and reserved beats.
Delete generic adjectives, duplicate style labels, obvious background details, secondary camera moves, and secondary actions before deleting preservation constraints.
For bilingual or mixed-language compression, load multilingual-community-examples. Keep only the language mix that clarifies reference roles, dialogue, camera terms, or safe production constraints.
Compact Templates
| Need | Template |
|---|---|
| T2V | [Subject] [action and endpoint] in [scene]. Camera: [one move]. Light/style: [physical source]. Sound: [cue]. Constraint: [risk/continuity]. |
| I2V | @Image1 preserved; only [motion/light/camera] changes. Camera: [one move]. Sound: [cue]. Constraint: [what must not change]. |
| V2V | @Video1 controls [motion/camera/timing] only; new subject [anchor]. [Action]. Do not transfer [identity/scene/logo]. |
| Chinese | @Image1为参考,严格保持[主体]不变;仅加入[动作/光线/镜头]。声音:[提示]。 |
| Storyboard | Shot 1: [cut or move], [action and expression], [space], [sound]. Shot 2: cut to [framing], [action to endpoint], [sound]. Constraint: [risk/continuity]. |
| Chinese storyboard | 镜头1:[运镜或切镜],[动作与表情],[声音]。镜头2:镜头切至[机位],[动作与终点],[声音]。保持无字幕。 |
Output Contract
Return one compact prompt, ideally 30-100 English words or an equivalent Chinese prompt when the user asks for Chinese or maximum compression. Translate the Director's Read into visible or audible carriers and keep its internal labels out of the prompt. Include a one-line note only if something important was removed.
Version History
-
4668457
Current 2026-09-27 12:53
引入Shot Table优先机制,确保多镜头提示词包含明确的镜头位置、灯光等结构信息,并集成自动化检查脚本以验证镜头块完整性。
-
9ea203f
2026-09-08 20:19
新增对无虚构戏剧性的观察/表演类提示的支持,明确非叙事场景的处理逻辑。
-
b8cdc5b
2026-08-04 19:06
新增强制加载Director's Read逻辑以规范提示词路由;修复嵌套运行时路由的可移植性问题;引入便携式导演阅读机制。
-
7f22eac
2026-08-02 22:01
更新版本至 v6.7.0,记录 Seedance 2.5 信息并修正前端排版与字体规范问题。
- 6c51262 2026-07-30 20:23


