Surreal mutation & impossible physics (Seedance 2.0)

Anti-physics storytelling: gravity tilts, rooms fold, materials swap mid-action, scale jokes land as clear visual punchlines.

Dependencies: WERYAI_API_KEY + Node.js 18+. This skill uses only SEEDANCE_2_0. When you run the CLI, scripts/video_gen.js must exist; WERYAI_VIDEO_API.md must exist under resources/ (supply both via publish or pre-use assembly). Full commands and JSON fields: see 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 image URLs. If the assembled scripts/video_gen.js accepts local paths, review/verify the script and explicitly consent before local read-and-upload to WeryAI.
• Model (caller / agent): The bundled video_gen.js requires a non-empty model in JSON—if model is missing or blank, the CLI exits with MISSING_PARAM (no default model). The script does not enforce this skill's allowed model in code: you must set "model":"SEEDANCE_2_0" for this package and show it in the confirmation table before submit—see WERYAI_VIDEO_API.md (model row).
• Each wait run may consume credits; re-run creates new paid tasks.

Security, secrets, and API hosts

• WERYAI_API_KEY: Treat as a secret. Only configure it if you trust this skill's source; it is listed in OpenClaw metadata as requires.env / primaryEnv so installers know it is mandatory at runtime (never commit it inside the skill package).
• API hosts (fixed in video_gen.js): Video tasks use https://api.weryai.com; the models list uses https://api-growth-agent.weryai.com. Child skills must not document or rely on URL environment-variable overrides—only WERYAI_API_KEY is read from the environment.
• Local image handling disclosure: Prefer public https image URLs. If the assembled scripts/video_gen.js supports local file paths, it may read a local image and upload it to WeryAI to obtain a public URL; require review / verification and explicit consent before using that path.
• Higher assurance: Run generation in a short-lived or isolated environment (separate account or container), and review scripts/video_gen.js (HTTPS submit + poll loop) before production use. Verify whether the runtime can read local image files and upload them to WeryAI, and obtain explicit consent before using that path.

Prompt expansion (mandatory)

video_gen.js does not expand prompts. Before every wait --json, turn the user's short or vague brief into a full English production prompt that re-styles the described event into this skill's look.

When: User input is short, vague, or missing cinematic detail. Exception: They supply a finished long prompt within 2000 characters and ask you not to rewrite—still show the full text in the confirmation table.

Always add: shot size and angle; camera move; lighting and color grade aligned to this skill; subject action paced to duration; one clear payoff; platform framing (9:16 unless user chose another allowed ratio); if generate_audio is true, describe ambience / SFX in generic words (no copyrighted music references).

Length: Stay within prompt_length_limit 2000 for SEEDANCE_2_0; trim adjectives before losing the core transformation beat.

Confirmation: The pre-submit table must include the full expanded prompt.

Style transformation checklist

• Rule breaks: melting clocks vibe, impossible stairs, duplicate reflections.
• Space warp: corridor stretch, sky on floor—one clear hero idea per clip.
• Materials: liquid metal, rubber stone—motivated by user beat.
• Scale play: giant hand, tiny city—readable silhouette.
• Motion payoff: expectation subversion in final third of timeline.
• Sound (if on): reversed whooshes, rubbery boings—sparse and odd.

Fits: Curiosity bait, contrast comedy, visual spectacle, pattern-interrupt ads.

### Example prompts below are richness targets only—always derive from the user's actual brief.

Workflow

1. Confirm the request matches this skill (text-to-video and/or single image-to-video).
2. Collect the user's brief, optional image URL, and tier (best / good / fast)—all map to SEEDANCE_2_0 with different duration defaults (see Recommended models).
3. Expand prompt (mandatory): Unless the user opted out with a finished long prompt, expand using ## Prompt expansion (mandatory). Do not submit only the user's minimal words.
4. Validate model is SEEDANCE_2_0 only; validate duration, aspect_ratio, resolution, generate_audio against the frozen tables.
5. Show the confirmation table with the full expanded prompt; wait for confirm or edits.
6. Run node scripts/video_gen.js wait --json '...'.
7. Parse stdout JSON; return videos URLs or explain errors.
8. When presenting playable URLs to the user, use Markdown inline links only (e.g. [Video](https://...)). Do not wrap user-facing links in code fences.

CLI reference

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

Full reference: WERYAI_VIDEO_API.md.

Definition of done

Done when the user gets at least one playable Video-style link, or a clear failure with next steps. Submitted prompt must be the expanded prompt unless the user explicitly opted out. Never use another model_key in this skill.

Boundaries (out of scope)

• Do not use any model other than SEEDANCE_2_0 for this package.
• Do not link to weryai-model-capabilities.md or shared ../references/ paths; use resources/WERYAI_VIDEO_API.md for CLI/API details.
• Prefer public https for image when easy; OpenClaw / chat attachments are often a local path—if video_gen.js can read it, pass it as image and the script uploads first (use an absolute path if a relative path fails). Never embed the secret value of WERYAI_API_KEY in files.
• Do not invent API fields; do not send negative_prompt (not supported for this model).
• Do not wrap user-facing playable URLs in Markdown code fences.

Example prompts

• Drinking coffee → cup becomes miniature swimming pool, diver splash gag, vertical
• Walking upstairs → steps melt into waterfall, seamless dream logic
• HTTPS sneaker → sole grows into slide, surreal playground

Model and API constraints (frozen for this skill)

Derived from node scripts/video_gen.js models alignment at authoring time; re-run models after platform upgrades. This skill is locked to SEEDANCE_2_0 only.

Text-to-video

Image-to-video (single image)

Recommended models

Tier only changes duration / resolution / audio defaults—never the model.

Default parameters

Scenario: Text-to-video style transform

Flow

1. Capture the user's mundane event in one or two sentences.
2. Expand into English with the Style transformation checklist; keep identity generic unless user names a character.
3. Confirmation table → user confirm.
4. Execute:

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

5. Return [Video](url) links.

Scenario: Image-to-video style transform

Before use: Prefer a public https:// URL reachable from the internet. Local paths (e.g. OpenClaw attachment paths) work when the Node process can read them—video_gen.js uploads then submits; use absolute paths if needed. Do not use plain http:// remote URLs.

1. Plan how the still re-styles in motion (props, light, environment) without breaking likeness if the user requests preservation.
2. Expand prompt; add image to the confirmation table.
3. After confirm:

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

4. Return [Video](url) links.

Loop seam (optional)

If the user asks for a seamless loop, append a short English trio at the end of the expanded prompt: seamless loop, perfect loop, ends where it begins—no separate API flag.

Generated for skill surreal-mutation-transform-video-gen-seedance2.0.