Benchmark RadarKeyFrame-Compass
KeyFrame-Compass evaluates keyframe-conditioned video generation across 386 curated samples spanning three application domains, two video structures, two prompt granularities, two conditioning formats, and four keyframe densities. It jointly measures keyframe execution (presence, fidelity, temporal ordering, localization, persistence, uniqueness) and overall video quality via evidence-grounded MLLM judgments and specialized perception models.
- Released
- 2026-07-15
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
As keyframe-based workflows grow in video production, there is no standardized way to assess whether models faithfully reproduce prescribed keyframes while maintaining natural video quality. KeyFrame-Compass provides a controlled testbed that reveals trade-offs and degradation patterns, helping practitioners choose models based on constraint density and input format.
Motivation
Video generation increasingly relies on keyframe-based workflows, where creators specify a sequence of reference images to guide generation.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.