2025 ICML ICML 2025

GRADEO: Towards Human-Like Evaluation for Text-to-Video Generation via Multi-Step Reasoning