WAN 3.0连续镜头契约

Use this skill before an authorized WAN 3.0 generation when the brief requires one continuous camera take or readable in-frame text. It turns a redacted brief into a deterministic request contract; it does not upload media or call a model.

Run

CODEBLOCK0

The input must contain duration_seconds, resolution, ratio, single_take, camera_plan, and optional text_elements. Keep media IDs and private URLs out of the examples. Exit code 0 means the structure is acceptable for a human/API handoff; exit code 1 means hold the request.

Checks

• - Duration is 2-30 seconds and resolution is 480P, 720P, or 1080P.
• The brief explicitly requests one take and has a non-empty camera path.
• The prompt does not introduce numbered shots, hard cuts, dissolves, or montage language.
• Each on-screen text item has an exact string, time window, safe-area hint, and at most two lines; each line is limited to 32 characters.
• Text windows stay inside the requested duration and do not overlap another item in the same region.

The report includes a prompt block with the exact text and camera constraints. It does not claim that the model will render text correctly or avoid an uncommanded cut.

Human boundary

An authorized operator must confirm the current model, region, endpoint, API key, input rights, policy, cost, and whether the buyer really wants audio. After generation, use the companion auditor and manually watch the clip frame by frame. Do not place credentials or user media in this package.