audit-faceless-video

Run the originality screen with the candidate script, one or more reference transcripts, and the production manifest, then run the motion screen on the exported video. Both scripts output JSON reports with PASS, REVIEW, BLOCK, or INCOMPLETE status, and manual review gates must still be completed before release.

Use the audit_motion.py script, which samples frames through FFmpeg and measures meaningful pixel change, distributed grid motion, scene cuts, and static hold duration. Renders below the motion thresholds or with static holds over 2 seconds are flagged as REVIEW.

The manifest needs asset_inventory_complete set to true, an assets array with provenance, license name, terms locator, explicit commercial_use_allowed boolean, and generation metadata for AI-generated assets. Ready assets also need a relative local_path and a valid 64-character SHA-256 hash.

The script locates FFmpeg automatically via the system PATH or the imageio-ffmpeg Python package. You can also pass an explicit FFmpeg executable path with the --ffmpeg argument if neither is available.

No. The audit is strictly read-only: it never modifies, re-renders, or replaces inputs, and it refuses output paths that resolve to any input path. Remediation requires a separate production pass.

Frame-difference analysis cannot prove that motion comes from independently controlled layers, so full-frame camera pans may pass numerically while still failing manual review. Manual playback verification of independent layer motion and interaction remains mandatory.