Enregistré dans:
| Auteurs principaux: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2605.18396 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911695869313024 |
|---|---|
| author | Feng, Yuxiang Wang, Juncheng Xu, Chao Qian, Yijie Wang, Huihan Hou, Wenlong Liu, Yang Sun, Baigui Liu, Yong Wang, Shujun |
| author_facet | Feng, Yuxiang Wang, Juncheng Xu, Chao Qian, Yijie Wang, Huihan Hou, Wenlong Liu, Yang Sun, Baigui Liu, Yong Wang, Shujun |
| contents | Video generation models produce visually compelling results but systematically violate physical commonsense -- on VideoPhy-2, the best model achieves only 32.6% joint accuracy. We identify a specification bottleneck: text prompts are lossy compression of the physical world, omitting the parameters that fully determine dynamics, and no amount of model scaling can recover what was never specified. From this diagnosis we derive three properties that physics conditioning must satisfy -- sufficiency, dynamism, and verifiability -- and show that no existing approach satisfies all three. We present NEWTON, in which video generation is demoted from the system output to one action inside an agent's toolbox: a learned planner orchestrates physics-aware tools (keyframe generation, scientific computation, prompt refinement) to construct rich conditioning, and a verifier closes the loop for iterative re-planning. The planner is the sole trainable component, optimized on-policy via Flow-GRPO inside the live multi-turn loop. On VideoPhy-2, NEWTON improves joint accuracy from 21.4% to 29.7% on LTX-Video and from 30.7% to 37.4% on Veo-3.1, without modifying either generator. Our project page: https://Newton026.github.io/newton |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_18396 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | NEWTON: Agentic Planning for Physically Grounded Video Generation Feng, Yuxiang Wang, Juncheng Xu, Chao Qian, Yijie Wang, Huihan Hou, Wenlong Liu, Yang Sun, Baigui Liu, Yong Wang, Shujun Computer Vision and Pattern Recognition Video generation models produce visually compelling results but systematically violate physical commonsense -- on VideoPhy-2, the best model achieves only 32.6% joint accuracy. We identify a specification bottleneck: text prompts are lossy compression of the physical world, omitting the parameters that fully determine dynamics, and no amount of model scaling can recover what was never specified. From this diagnosis we derive three properties that physics conditioning must satisfy -- sufficiency, dynamism, and verifiability -- and show that no existing approach satisfies all three. We present NEWTON, in which video generation is demoted from the system output to one action inside an agent's toolbox: a learned planner orchestrates physics-aware tools (keyframe generation, scientific computation, prompt refinement) to construct rich conditioning, and a verifier closes the loop for iterative re-planning. The planner is the sole trainable component, optimized on-policy via Flow-GRPO inside the live multi-turn loop. On VideoPhy-2, NEWTON improves joint accuracy from 21.4% to 29.7% on LTX-Video and from 30.7% to 37.4% on Veo-3.1, without modifying either generator. Our project page: https://Newton026.github.io/newton |
| title | NEWTON: Agentic Planning for Physically Grounded Video Generation |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.18396 |