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):

Doc	Role
resources/seedance2-prompt-engineering-playbook.md	Hub: 60s workflow, @→JSON map, quick beats
resources/modes-and-weryai-mapping.md	Modes, WeryAI field mapping, dialogue/sound, multi-segment, pitfalls, pre-flight checklist
resources/recipes-weryai.md	A–K production recipes (adventure, extend, IP-safe, toy, product, drama, one-take, 30s plan)
resources/camera-and-styles.md	Shot, movement, angle, lighting, palette, render vocabulary

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

• Confirm the user wants Seedance 2.0 on WeryAI and whether the path is text-only or uses reference image(s).
• Collect the brief, optional https URLs, target duration (5–15 s, integer), aspect_ratio, resolution, and whether generate_audio should be true or false.
• 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.
• Validate prompt length against prompt_length_limit (2000); trim lower-priority clauses if over limit.
• Pre-submit gate: Show the confirmation table; stop until the user explicitly confirms or edits.
• 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.
• 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.
• Continue (user-driven): Only after the user agrees, use status --task-id … at reasonable intervals or wait --json '…' with the same payload.
• On completion, return playable URLs as Markdown inline links only—e.g. Video—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":"","duration":10,"aspect_ratio":"9:16","resolution":"720p","generate_audio":true}' node scripts/video_gen.js submit-image --json '{"model":"SEEDANCE_2_0","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 ~~~

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

model_key	durations (s)	aspect_ratios	resolutions	Audio (generate_audio)	negative_prompt (API)	prompt_length_limit
SEEDANCE_2_0	5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15	9:16, 1:1, 16:9	480p, 720p	Yes	Not supported — omit field	2000

Image-to-video

model_key	durations (s)	aspect_ratios	resolutions	Audio	negative_prompt (API)	Multi-image	First/last frame	upload_image_limit	prompt_length_limit
SEEDANCE_2_0	5–15 (integers as above)	9:16, 1:1, 16:9	480p, 720p	Yes	Not supported — omit field	Yes	Yes	3	2000

---

Recommended models

Tier	model_key	Notes
Default	SEEDANCE_2_0	Only model covered by this skill; text + image paths

---

Default parameters (when the user is silent)

Field	Value
model	SEEDANCE_2_0
aspect_ratio	9:16
duration	10 (must be one of the allowed integers)
resolution	720p
generate_audio	true unless the user wants silent motion

---

Scenario: Text-to-video with prompt engineering

• 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.
• Build one English prompt; keep within 2000 characters.
• Pre-submit gate with full prompt and numeric fields.
• After explicit confirmation:

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

• Notify task id; offer status vs wait. Return final URLs as Video.

---

Scenario: Image-to-video (single reference)

• Obtain one https image URL (or local path only with script review + explicit consent).
• 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.
• Expand prompt with motion beats and camera; merge dialogue/sound lines if needed.
• Pre-submit gate including the image URL and full prompt.
• After explicit confirmation:

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

• 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.