HumanScore: Benchmarking Human Motions in Generated Videos

2026-04-21

AI Summary

Current AI-generated videos lack systematic evaluation regarding the realism and biomechanical plausibility of human motion. To address this gap, this work proposes HumanScore, a novel multidimensional evaluation framework specifically designed for human actions, introducing six interpretable metrics that assess kinematic plausibility, temporal stability, and biomechanical consistency, among other dimensions. Leveraging a carefully curated set of motion prompts and integrating kinematic modeling, temporal analysis, and biomechanical constraints, the framework quantitatively evaluates 13 state-of-the-art video generation models. The results not only yield a robust ranking of models based on motion quality but also uncover a significant discrepancy between visual fidelity and physical plausibility, while identifying common failure modes such as temporal jittering and anatomically implausible poses.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Computer Vision: Video Understanding & Activity Analysis
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Recent advances in model architectures, compute, and data scale have driven rapid progress in video generation, producing increasingly realistic content. Yet, no prior method systematically measures how faithfully these systems render human bodies and motion dynamics. In this paper, we present HumanScore, a systematic framework to evaluate the quality of human motions in AI-generated videos. HumanScore defines six interpretable metrics spanning kinematic plausibility, temporal stability, and biomechanical consistency, enabling fine-grained diagnosis beyond visual realism alone. Through carefully designed prompts, we elicit a diverse set of movements at varying intensities and evaluate videos generated by thirteen state-of-the-art models. Our analysis reveals consistent gaps between perceptual plausibility and motion biomechanical fidelity, identifies recurrent failure modes (e.g., temporal jitter, anatomically implausible poses, and motion drift), and produces robust model rankings from quantitative and physically meaningful criteria.

Problem

• human motion evaluation
• video generation
• biomechanical consistency
• kinematic plausibility
• temporal stability

Innovation

• HumanScore
• human motion evaluation
• biomechanical consistency
• video generation benchmarking
• kinematic plausibility