OpenClaw Seedance Prompt Library

Search, rewrite, and generate Seedance 2.0 video prompts with cinematic variants and bilingual output. Seedance提示词检索/改写/生成

Security Scan

The skill is largely consistent with a Seedance prompt library, but it includes guidance that downplays required CC BY 4.0 attribution and could encourage redistributing upstream prompts without proper attribution — this legal/incoherence concern warrants caution.

Purpose & Capability

Name/description, included reference files, and the search helper script all align with a prompt library for Seedance prompts. The helper script fetches README files from the stated upstream repo, which is coherent with the skill's stated purpose.

Instruction Scope

SKILL.md stays focused on searching, rewriting, and generating prompts, and limits output sizes. However, it explicitly instructs to 'keep any required license handling at the repository level, not in the normal user-facing output flow.' The included references/source-attribution.md states the upstream is CC BY 4.0 and requires attribution. Encouraging omission of attribution from user-facing outputs is a legal/consistency problem and could cause downstream license violations (or produce outputs that omit required attribution).

Install Mechanism

No install spec and only a small helper script are included. There are no downloads, extracted archives, or external installers — minimal on-disk impact and low install risk.

Credentials

The skill requests no environment variables, no credentials, and no config paths. The helper script fetches content from GitHub raw URLs, which is appropriate for searching a public repository and proportionate to the stated function.

Persistence & Privilege

The skill is not always-enabled and does not request elevated persistence or modify other skills. It does not request special privileges or access to other agents' configurations.

What to consider before installing

This skill appears to do what it says (search and generate Seedance prompts) and contains only a small helper script that fetches the upstream README files. However, the upstream material is marked CC BY 4.0 (attribution required) and the SKILL.md suggests keeping license handling out of normal user-facing outputs — that is inconsistent and could lead to redistributing prompts without the required attribution. Before installing or using the skill: 1) Decide whether you will include attribution in user-visible outputs or otherwise comply with CC BY 4.0 (e.g., include a short attribution line or link in generated prompt packs or a repository README). 2) Audit any outputs you plan to redistribute to ensure they are not verbatim copyrighted prompts that require attribution or that exceed reasonable reuse under the license. 3) If you will run the included script in an environment with network access, be aware it fetches raw files from GitHub (expected for searching) — confirm that network access and remote fetches are acceptable in your environment. 4) If you want to be conservative, update SKILL.md/response templates to include a visible attribution note when upstream examples are used. Overall the implementation is low-risk technically, but the license-handling guidance is the main red flag and should be corrected before wide deployment.

Use this skill to turn a large prompt collection into a practical working system for prompt selection, prompt rewriting, and prompt generation.

Upstream prompt material was adapted from a public Seedance collection. Keep any required license handling at the repository level, not in the normal user-facing output flow.

Core use cases

• Find Seedance prompt examples by style, subject, motion, or camera language
• Rewrite a weak prompt into a stronger cinematic prompt
• Convert Chinese prompts to English and English prompts to Chinese
• Turn a user idea into a Seedance-ready prompt pack
• Extract reusable prompt patterns from the library instead of copying blindly

Working model

Do not dump a giant wall of prompts.

Instead:

1. classify the user's goal
2. search for 3-8 relevant examples
3. identify what actually makes them work
4. synthesize a better prompt for the user's use case
5. return a small, usable set of outputs

Output modes

Choose the output format that best fits the request:

1) Prompt shortlist

Use when the user wants inspiration.

Return:

• 3-8 relevant prompt examples
• one-line note on why each is relevant
• optional source title / ID when available

2) Prompt rewrite

Use when the user already has a rough prompt.

Return:

• original prompt
• improved prompt
• what changed: subject / motion / camera / lighting / duration / aspect ratio / pacing

3) Prompt pack

Use when the user wants multiple options.

Return 3 variants:

• safe — closest to user's request
• stylized — stronger cinematic/style choices
• experimental — more ambitious motion or scene design

4) Bilingual prompt delivery

Use when the user needs Chinese + English.

Return:

• Chinese version
• English version
• short note on any non-literal adaptation

5) Prompt generator

Use when the user gives only an idea, image concept, mood, or rough scene.

Return:

• core prompt — strongest balanced version
• safe — most reliable generation version
• stylized — stronger visual identity and camera language
• experimental — higher-risk, higher-novelty version
• optional negative / avoid notes if they help reduce model drift
• optional CN + EN pair when bilingual output helps

Prompt construction rules

When rewriting or creating prompts, prefer this structure:

1. Subject — who / what is on screen
2. Scene — environment and setting
3. Action — what changes over time
4. Camera — shot type, movement, framing
5. Style — cinematic, anime, realistic, surreal, etc.
6. Lighting / mood — color, atmosphere, time of day
7. Output constraints — duration, resolution, aspect ratio if useful

If the source example is strong, extract the structure, not just the wording.

Generator workflow

When the user gives only a rough idea, generate prompts in this order:

1. lock the subject and scene
2. choose one dominant motion arc
3. choose one camera logic
4. add one style layer, not five competing ones
5. add constraints only if they improve reliability
6. output a core prompt first
7. then derive safe / stylized / experimental variants

If the request is weak or underspecified, fill gaps with plausible cinematic defaults instead of asking too many questions.

Quality rules

• Prefer specificity over adjectives spam
• Prefer concrete motion over vague “dynamic” wording
• Prefer camera instructions that imply visual change
• Avoid contradictory style directions
• Keep prompt length proportional to complexity
• If a user asks for a simple result, do not over-engineer it

Safety notes

• If copyright or character/IP risk is high, flag it and offer an original alternative
• Prefer transformed prompt synthesis over verbatim copying
• If a user wants a close clone of a known character or franchise scene, offer an adjacent original version too

References

Read these files as needed:

• references/usage-patterns.md — how to search, adapt, and package prompts
• references/source-attribution.md — source repo, license, and reuse rules
• upstream dataset: https://github.com/YouMind-OpenLab/awesome-seedance-2-prompts

Optional helper scripts

• scripts/search-seedance-readme.mjs — keyword search against upstream README prompt entries via GitHub raw URLs