seedance-2-prompt-engineering-to-video（Seedance 2 提示工程到视频）

Seedance 2.0 prompt engineering to video

Display name: Seedance 2.0 prompt engineering to video · Package folder: seedance-2-prompt-engineering-to-video

Combine high-control Seedance-style prompt design (modes, asset roles, timeline beats, IP-safe language) with direct WeryAI generation for SEEDANCE_2_0 text-to-video and image-to-video (single image by default; optional multi-image / first–last frame within model limits).

Resource library (prompt guidance — under resources/ next to WERYAI_VIDEO_API.md after sync):

Dependencies: WERYAI_API_KEY + Node.js 18+. After assembly, scripts/video_gen.js and resources/ must include resources/WERYAI_VIDEO_API.md (your publish/sync pipeline copies the API doc from the WeryAI video skill creator; prompt guides ship in-repo). Full JSON fields and commands: resources/WERYAI_VIDEO_API.md. No other Cursor skills required.

Prerequisites

• WERYAI_API_KEY must be set before running video_gen.js.
• Node.js 18+. Prefer public https URLs for image, images, first_frame, last_frame, and last_image. If the bundled runtime supports local file paths, review scripts/video_gen.js, verify read-and-upload behavior, and obtain explicit user consent before using local paths.
• model is mandatory in every submit-* and wait JSON body. This skill only documents SEEDANCE_2_0. If model is missing or blank, the CLI exits with MISSING_PARAM (no default model in the script).
• Each successful **submit-* / wait** may consume credits; retries create new paid tasks.
• SEEDANCE_2_0 does not support API negative_prompt on text or image paths—do not send that field for this model.

Security, secrets, and API hosts

• WERYAI_API_KEY: Treat as a secret; never commit its value inside the skill package. OpenClaw metadata declares primaryEnv / requires.env so installers see the runtime contract.
• API hosts (fixed in video_gen.js): Video tasks use https://api.weryai.com; the models list uses https://api-growth-agent.weryai.com. Do not rely on undocumented URL overrides.
• Local image disclosure: Prefer https. If the runtime can read local files and upload them to WeryAI for a public URL, require review, verification, and explicit consent before that path.
• Higher assurance: Use an isolated or short-lived environment; review scripts/video_gen.js before production paid use.

Pre-submit gate (mandatory)

⚠️ No paid submit without explicit user confirmation. Do not call submit-text, submit-image, submit-multi-image, or wait until the user has explicitly approved the full parameter table below, including the entire expanded prompt. Never infer consent from silence or vague “continue”. Explicit means confirm / go / approved / yes, generate (or clear equivalent in the user’s language).

Parameter confirmation table (show before any submit): model, duration, aspect_ratio, resolution, generate_audio, reference URLs (if any), and full expanded prompt (complete text, not a summary).

Workflow

1. Confirm the user wants Seedance 2.0 on WeryAI and whether the path is text-only or uses reference image(s).
2. Collect the brief, optional https URLs, target duration (5–15 s, integer), aspect_ratio, resolution, and whether generate_audio should be true or false.
3. Prompt engineering (mandatory): Use the resource library above. Start from resources/seedance2-prompt-engineering-playbook.md; for mode and JSON mapping use resources/modes-and-weryai-mapping.md; match a scenario from resources/recipes-weryai.md when helpful; sharpen wording with resources/camera-and-styles.md. Choose mode, map each URL to a role, draft timecoded beats, then compress into one API prompt. The API does not use @image1 tokens; roles belong in prose while URLs sit in JSON. Unless the user supplied a finished long prompt and asked not to rewrite, always expand short input into a production English prompt.
4. Validate prompt length against prompt_length_limit (2000); trim lower-priority clauses if over limit.
5. Pre-submit gate: Show the confirmation table; stop until the user explicitly confirms or edits.
6. Submit (async, default): After confirmation, run submit-text or submit-image (or submit-multi-image when using multiple URLs), using the same JSON shape as wait (see resources/WERYAI_VIDEO_API.md). Do not start a long blocking wait in the same turn unless the user already asked to block until the video is ready.
7. Immediate notify: On success, report taskId / batchId (or documented id fields), short status (e.g. queued), and ask whether to poll status or use blocking wait next. Do not run long invisible polling without that choice.
8. Continue (user-driven): Only after the user agrees, use status --task-id … at reasonable intervals or wait --json '…' with the same payload.
9. On completion, return playable URLs as Markdown inline links only—e.g. [Video](https://…)—not inside fenced code blocks.

Prompt expansion (mandatory)

video_gen.js does not auto-expand prompts.

• When: The user gives a short or vague brief. Exception: They provide a final long prompt within 2000 characters and explicitly opt out of rewriting—still show the full text in the pre-submit table.
• What to add: Mode; asset mapping (for each URL); timecoded beats for the chosen duration; shot scale, angle, camera move (see resources/camera-and-styles.md); lighting and materials; one clear payoff; framing for aspect_ratio; if generate_audio is true, describe ambience/SFX in generic, non-copyrighted terms; dialogue/sound as separate labeled lines when needed (see resources/modes-and-weryai-mapping.md §4).
• IP / moderation: resources/modes-and-weryai-mapping.md §8 and recipe blocks in resources/recipes-weryai.md (D, E, F).
• Length: Stay within 2000 characters for prompt.
• Confirmation: The pre-submit table must include the full expanded prompt.

CLI reference

After confirmation — async submit, then optional status:

~~~sh node scripts/video_gen.js submit-text --json '{"model":"SEEDANCE_2_0","prompt":"<expanded English prompt>","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' node scripts/video_gen.js submit-image --json '{"model":"SEEDANCE_2_0","prompt":"<expanded English prompt>","image":"https://example.com/ref.png","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' node scripts/video_gen.js status --task-id <TASK_ID> ~~~

Blocking until done — only if the user explicitly chooses wait:

~~~sh node scripts/video_gen.js wait --json '{"model":"SEEDANCE_2_0","prompt":"…","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' ~~~

Multi-image or first/last frame — use submit-multi-image or the field names documented in resources/WERYAI_VIDEO_API.md; keep images length ≤ 3 for SEEDANCE_2_0.

Full CLI and API: resources/WERYAI_VIDEO_API.md.

Definition of done

Done when the user receives at least one playable Video link or a clear error with a fix. Pre-submit: full prompt and parameters were explicitly confirmed. Post-submit: the user was notified with task id(s) and chose status vs wait. Submitted prompt must be the engineered production string unless the user opted out. Do not wrap user-facing video URLs in code fences.

Boundaries (out of scope)

• Do not send negative_prompt for SEEDANCE_2_0 (unsupported on this model’s rows).
• Do not exceed upload_image_limit (3) or claim multi-image support for other models inside this skill—this skill documents only SEEDANCE_2_0.
• Do not embed WERYAI_API_KEY values in files. Do not link to weryai-model-capabilities.md or arbitrary paths outside this skill package for shared capability docs.
• Do not use local image paths as the default; if supported by the bundled script, disclose upload behavior and require explicit consent.
• This skill does not run JiMeng-native UIs; it targets WeryAI JSON + video_gen.js only.

Example prompts

• Plan a 10s 9:16 Seedance 2.0 clip from a one-line cyberpunk alley idea, then generate on WeryAI after I confirm
• I have an https product still—map it as identity anchor, expand the prompt with beats, then image-to-video with SEEDANCE_2_0
• Help me write an IP-safe fantasy creature prompt and submit text-to-video with audio on

Model and API constraints (frozen for this skill)

Derived from node scripts/video_gen.js models (text_to_video + image_to_video) at authoring time. Re-run after platform upgrades. This skill uses only SEEDANCE_2_0.

Text-to-video

Image-to-video

Recommended models

Default parameters (when the user is silent)

Scenario: Text-to-video with prompt engineering

1. Read resources/seedance2-prompt-engineering-playbook.md and pick a matching recipe from resources/recipes-weryai.md if applicable; set Text-only mode; draft beats for the chosen duration.
2. Build one English prompt; keep within 2000 characters.
3. Pre-submit gate with full prompt and numeric fields.
4. After explicit confirmation:

~~~sh node scripts/video_gen.js submit-text --json '{"model":"SEEDANCE_2_0","prompt":"<expanded>","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' ~~~

1. Notify task id; offer status vs wait. Return final URLs as [Video](url).

Scenario: Image-to-video (single reference)

1. Obtain one https image URL (or local path only with script review + explicit consent).
2. Plan single-image mode and asset mapping (identity, composition, palette to preserve); see resources/modes-and-weryai-mapping.md §2.3 and recipes F / H in resources/recipes-weryai.md.
3. Expand prompt with motion beats and camera; merge dialogue/sound lines if needed.
4. Pre-submit gate including the image URL and full prompt.
5. After explicit confirmation:

~~~sh node scripts/video_gen.js submit-image --json '{"model":"SEEDANCE_2_0","prompt":"<expanded>","image":"https://example.com/input.png","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' ~~~

1. Same notify / poll / link rules as above.

Scenario (advanced): Multi-image or first/last frame

Use only when the user supplies multiple references or explicit start/end frames and agrees to SEEDANCE_2_0 limits (≤ 3 uploads). Map order with resources/modes-and-weryai-mapping.md §2.4 / §6 and recipe J in resources/recipes-weryai.md; then use submit-multi-image and field names from resources/WERYAI_VIDEO_API.md. Pre-submit gate must list every URL and the full prompt.

Pack: seedance-2-prompt-engineering-to-video · slug: seedance-2-prompt-engineering-video-gen.