Public benchmark
AI video continuity benchmark
Live — updates as new runs are scoredCalibrated character consistency scores for text-to-video models, computed on a fixed prompt suite by the open-source VideoConsistency pipeline.
Loading scores…
Methodology
Each model is scored on the same fixed prompt suite of long-form multi-shot briefs. Scores are calibrated (1.0 ≈ same subject re-rendered, 0.0 ≈ different subject) and computed by the open-source VideoContinuity pipeline — not by human raters or vibes. Full methodology on GitHub.
This benchmark is operated by the VideoContinuity project. Model vendors can request a re-run or dispute a score — see the fairness policy.