FilmBench is a cinematic video-generation benchmark co-developed with directors and faculty from the Beijing Film Academy and the Hujing Digital Media & Entertainment Group film studio. It evaluates text-to-video and reference-to-video systems using prompts reverse-engineered from award-winning live-action films across 20 genres. Of 1,169 prompts, 1,056 are multi-shot and follow real shot lists. Its taxonomy contains three axes, 12 components, and 35 T2V plus three R2V-only sub-metrics. An expert-grade automatic evaluator matches human model rankings with Spearman correlations of 0.95 for T2V and 0.96 for R2V.
No heat snapshots are available in the last 24 hours.