Conan

ProofFrame — Evidence-Based QA for AI-Generated Video AI-generated videos often contain subtle errors: inconsistent characters, disappearing props, incorrect captions, or missing dialogue. Manually reviewing every render is slow and expensive. ProofFrame is a provider-agnostic verification engine that checks generated videos against pre-approved requirements. Using Meta's multimodal AI, it detects visual and narrative inconsistencies, grounds findings in actual frames and timestamps, and produces actionable repair instructions. Its fail-closed design prevents uncertain results from being marked as passed. After repairs, ProofFrame re-verifies the video and detects regressions. On a real Vietnamese film workprint, dual-ASR verification eliminated four false positives while preserving the genuine dialogue error, which was subsequently fixed and verified. ProofFrame turns AI video quality assurance into a measurable, repeatable, agent-ready workflow.