KeyFrame-Compass introduces what the paper describes as the first comprehensive benchmark for keyframe-conditioned video generation. It contains 386 curated samples spanning three application domains, two video structures, two prompt granularities, two conditioning formats, and four keyframe densities. The evaluation jointly measures keyframe execution and overall video quality. Six metrics cover keyframe presence, fidelity, temporal ordering, localization, persistence, and uniqueness, while evidence-grounded MLLM judgments and specialized perception models assess video quality. Experiments across nine representative systems expose a trade-off between following keyframes faithfully and producing natural videos. Performance worsens with denser constraints, and many open-source systems misread storyboard grids as ordered keyframe sequences.
No heat snapshots are available in the last 24 hours.