Diverse Video Generation with Determinantal Point Process-Guided Policy Optimization

2025-11-25

AI Summary

Current text-to-video (T2V) diffusion models suffer from insufficient diversity in multi-video generation from a single prompt. To address this, we propose DPP-GRPO—a novel framework that introduces Determinantal Point Processes (DPPs) into T2V generation for the first time, explicitly modeling inter-video diversity within a generated set. Coupled with Group Relative Policy Optimization (GRPO), a reinforcement learning algorithm, our method jointly optimizes prompt fidelity, visual quality, and diversity. DPP-GRPO adopts a plug-and-play, model-agnostic training paradigm, compatible with leading T2V models including WAN and CogVideoX. Extensive evaluations on VBench, VideoScore, and human preference studies demonstrate significant improvements in diversity across visual appearance, camera motion, and scene structure—without compromising generation quality. To foster reproducibility and further research, we publicly release both the implementation and a benchmark dataset comprising 30,000 diverse, high-quality prompts.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

While recent text-to-video (T2V) diffusion models have achieved impressive quality and prompt alignment, they often produce low-diversity outputs when sampling multiple videos from a single text prompt. We tackle this challenge by formulating it as a set-level policy optimization problem, with the goal of training a policy that can cover the diverse range of plausible outcomes for a given prompt. To address this, we introduce DPP-GRPO, a novel framework for diverse video generation that combines Determinantal Point Processes (DPPs) and Group Relative Policy Optimization (GRPO) theories to enforce explicit reward on diverse generations. Our objective turns diversity into an explicit signal by imposing diminishing returns on redundant samples (via DPP) while supplies groupwise feedback over candidate sets (via GRPO). Our framework is plug-and-play and model-agnostic, and encourages diverse generations across visual appearance, camera motions, and scene structure without sacrificing prompt fidelity or perceptual quality. We implement our method on WAN and CogVideoX, and show that our method consistently improves video diversity on state-of-the-art benchmarks such as VBench, VideoScore, and human preference studies. Moreover, we release our code and a new benchmark dataset of 30,000 diverse prompts to support future research.

Problem

• Generating diverse videos from single text prompts
• Optimizing set-level policies for diverse visual outcomes
• Reducing redundancy while maintaining prompt fidelity

Innovation

• DPP-GRPO framework enforces diversity via DPP and GRPO
• Plug-and-play method improves diversity without quality loss
• Groupwise feedback and diminishing returns prevent redundant samples