What problem does it solve? AI-generated video clips often contain defects like face drift, morphing characters, motion glitches, or content mismatches, and teams lack a structured way to catch these errors before final assembly. This Skill provides a standardized QA gate that checks every clip against the shotlist and character bible, classifies defects with a consistent taxonomy, and blocks the pipeline until all shots pass. ## Core Features & Use Cases - Two-Layer Clip Inspection: Verifies content match against the shotlist and bible (subject, action, location, face similarity, camera, dialogue), then scans for AI defects using a standard taxonomy (morphing, face-drift, artifact, motion-glitch, audio-mismatch) with minor/major severity levels. - Two-Phase QA Workflow: Supports keyframe QA on static images before rendering to save credits, and video clip QA after animation or lip-sync, with batch review following the generation order. - Actionable Fix Recommendations: Maps each defect type to concrete fixes such as switching to lip-sync mode with Hedra or LivePortrait, regenerating keyframes with InstantID or Midjourney --cref, or rescuing shots via face swap with FaceFusion or Remaker AI. - Use Case: After generating 12 clips for a short AI film, run this Skill to produce a qa-report.md showing which shots passed, which failed with evidence, and exactly what the prompt engineer should change in the next fix round (up to 3 rounds). ## Quick Start Review the generated clips in my project against the shotlist and character bible, then write a qa-report.md with PASS or FAILED verdicts and fix suggestions for each shot.
