Human detectors are surprisingly powerful reward models

2026-01-15

AI Summary

Existing video generation models often suffer from missing limbs, distorted poses, or physically implausible motions when synthesizing complex non-rigid human actions such as dance or sports. This work proposes HuDA, a reward model that leverages the confidence scores of off-the-shelf human detectors as an efficient, fine-tuning-free reward signal—an insight not previously explored. Combined with a temporal prompt alignment score, HuDA guides post-training of generative models via Group Reward Policy Optimization (GRPO). Requiring no manual annotations, the approach achieves a 73% win rate over state-of-the-art models like Wan 2.1 on complex human motion generation and significantly enhances video quality in broader scenarios, including animal motion and human-object interactions.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Video generation models have recently achieved impressive visual fidelity and temporal coherence. Yet, they continue to struggle with complex, non-rigid motions, especially when synthesizing humans performing dynamic actions such as sports, dance, etc. Generated videos often exhibit missing or extra limbs, distorted poses, or physically implausible actions. In this work, we propose a remarkably simple reward model, HuDA, to quantify and improve the human motion in generated videos. HuDA integrates human detection confidence for appearance quality, and a temporal prompt alignment score to capture motion realism. We show this simple reward function that leverages off-the-shelf models without any additional training, outperforms specialized models finetuned with manually annotated data. Using HuDA for Group Reward Policy Optimization (GRPO) post-training of video models, we significantly enhance video generation, especially when generating complex human motions, outperforming state-of-the-art models like Wan 2.1, with win-rate of 73%. Finally, we demonstrate that HuDA improves generation quality beyond just humans, for instance, significantly improving generation of animal videos and human-object interactions.

Problem

• human motion
• video generation
• non-rigid motion
• temporal coherence
• visual fidelity

Innovation

• HuDA
• reward modeling
• video generation
• human motion realism
• GRPO