Research map / Video / Video generation
PhysFlow: Physics-Intrinsic Velocity Regularization for Motion-Intensive Video Generation
Xianglong Guo, Chang Yu, Haobo Xu, Junhao Ma, Zhen Lei
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes PhysFlow, a physics-intrinsic velocity regularization for flow-matching text-to-video models to reduce motion artifacts in motion-intensive scenarios.
Problem
Flow-matching text-to-video models degrade under motion-intensive prompts, producing object fragmentation, geometric deformation, and temporal flickering.
Approach
Regularizes the velocity field governing latent motion evolution, in contrast to existing training-free methods that act at the attention or guidance-scale level.
Experimental setup
Baselines: training-free attention-level methods guidance-scale methods
Builds on: flow matching
Closest papers by meaning
- Temporal-aware Flow Matching for Video Generation with Temporally Coherent Motion (ICML 2026 · similarity 0.89)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (NeurIPS 2026 · similarity 0.89)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.89)
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · similarity 0.88)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.87)
- PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
