Get Started
Home
Topics
Search
Library
Research questionHow can we verify physical obligations in generated videos and locate evidence for each failure?Generated videos can appear visually fluent while violating physical requirements. Scalar quality scores do not reveal which obligation failed, when it failed, or what evidence supports the judgment.
AI
AI Agents
Computer Vision
Evaluation & Benchmarks
Image & Video Processing
Reasoning
Research Paper
Technology
Video Generation
Latest papersRecent research connected to this question, newest first.VeriPhy: Agentic Physical Reasoning for World Model Evaluation and RefinementThe source describes VeriPhy, which uses a text-only planner, observations, and frozen low-level experts to produce provenance-carrying evidence and supported, contradicted, or unknown verdicts. Evidence comes from a 1,500-clip corpus of human-annotated flaw records; the reported core evaluation covers 149 clips and 304 records, with comparisons against another evaluator on the same clips and claims.research paper · Sep 2, 2026
Related questions
How can we evaluate physical reasoning in video models without reliable motion calibration?How should generated interactive videos be evaluated for action adherence and visual-temporal coherence?How can we diagnose audio generation and audiovisual grounding failures in text-to-audio-video systems?How can text-guided 3D scene synthesis be evaluated against fine-grained spatial constraints?