论文研究
x.com ·
@NVIDIAAI
NVIDIA 发布 Physis-Lang：为视频描述注入物理推理以改进世界模型
NVIDIA 研究者推出开源自演化框架 Physis-Lang，为视频描述补充解释场景成因与过程的物理推理，用于微调世界模型并在生成视频时加入推理提示。发布方称仅添加物理推理提示即可使 NVIDIA Cosmos 3 的 PhyGenBench 分数提升 5.62 分，无需重新训练，并附论文与项目页面。
AI推荐理由 NVIDIA 研究者发布开源自演化框架 Physis-Lang，为视频描述加入物理推理，用于微调世界模型并在生成时注入推理提示；称仅靠提示即让 Cosmos 3 的 PhyGenBench 分数提升 5.62 分且无需重训练，并附论文与项目页。有具体指标和可查链接，但为发布方自述，需以论文与基准复现验证。对视频世界模型的物理一致性改进有实际参考价值。
76
浏览 2,832
点赞 33
评论 4
转发 5
正文
A video world model can make a convincing clip and still get the physics wrong.
Our researchers just released Physis-Lang, an open self-evolving framework that adds physics reasoning to video captions. The captions explain why and how a scene unfolds. We use them to fine-tune world models and add that reasoning to prompts when generating video.
Adding physics reasoning to the prompt alone improved NVIDIA Cosmos 3’s PhyGenBench score by 5.62 points without retraining.
Paper and project: [physis-intelligence.github.io/physis-lang-we…](https://physis-intelligence.github.io/physis-lang-web/)
[Your browser does not support the video tag.](https://video.twimg.com/amplify_video/2105050806370369536/vid/avc1/1920x1080/7CsE3WEHQ3nISnMK.mp4?tag=29)
[🔗 View on Twitter](https://x.com/NVIDIAAI/status/2105051552029556942/video/1)