Video World Models · 2025-01-07
Cosmos World Foundation Model Platform for Physical AI
NVIDIA, :, Niket Agarwal, Arslan Ali, Maciej Bala, Yogesh Balaji, Erik Barker, Tiffany Cai, Prithvijit Chattopadhyay, Yongxin Chen, Yin Cui, Yifan Ding, Daniel Dworakowski, Jiaojiao Fan, Michele Fenzi, Francesco Ferroni, Sanja Fidler, Dieter Fox, Songwei Ge, Yunhao Ge, Jinwei Gu, Siddharth Gururani, Ethan He, Jiahui Huang, Jacob Huffman, Pooya Jannaty, Jingyi Jin, Seung Wook Kim, Gergely Klár, Grace Lam, Shiyi Lan, Laura Leal-Taixe, Anqi Li, Zhaoshuo Li, Chen-Hsuan Lin, Tsung-Yi Lin, Huan Ling, Ming-Yu Liu, Xian Liu, Alice Luo, Qianli Ma, Hanzi Mao, Kaichun Mo, Arsalan Mousavian, Seungjun Nah, Sriharsha Niverty, David Page, Despoina Paschalidou, Zeeshan Patel, Lindsey Pavao, Morteza Ramezanali, Fitsum Reda, Xiaowei Ren, Vasanth Rao Naik Sabavat, Ed Schmerling, Stella Shi, Bartosz Stefaniak, Shitao Tang, Lyne Tchapmi, Przemek Tredak, Wei-Cheng Tseng, Jibin Varghese, Hao Wang, Haoxiang Wang, Heng Wang, Ting-Chun Wang, Fangyin Wei, Xinyue Wei, Jay Zhangjie Wu, Jiashu Xu, Wei Yang, Lin Yen-Chen, Xiaohui Zeng, Yu Zeng, Jing Zhang, Qinsheng Zhang, Yuxuan Zhang, Qingqing Zhao, Artur Zolkowski
Editor pick
TL;DR
Cosmos is NVIDIA's platform for 'world foundation models' for Physical AI: a video curation pipeline, pre-trained diffusion and autoregressive world models, examples of post-training for downstream tasks, and video tokenizers, released open-source with open-weight models.
Why it matters
Positions a general-purpose pre-trained world model as something developers fine-tune into customized world models for robotics and driving; the code and weights are open with permissive licenses.
Method
- Video curation pipeline (motion, quality, text and video-type filtering) for training data.
- Pre-trained world foundation models (diffusion- and autoregressive-based) plus post-training examples.
- Cosmos Tokenizer for efficient video encoding and decoding.
Results
Editor-verified numbers only.
| Benchmark | Metric | Value | Baseline | Δ |
|---|---|---|---|---|
| Video curation pipeline | throughput increase from combined optimizations (x) | 6.5 | – | – |
| Video tokenizer speed | throughput vs prior tokenizers (x faster, up to) | 12 | prior tokenizers (–) | – |
Limitation
Failure cases: in the authors' autoregressive models, objects can unexpectedly appear from below in generated videos; they studied the failure rate on an evaluation set of 100 Physical AI inputs.
Lineage
- ← contrasts by V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning (Compares its latent world model's planning against Cosmos for robot manipulation.)
Abstract (from arXiv)
Physical AI needs to be trained digitally first. It needs a digital twin of itself, the policy model, and a digital twin of the world, the world model. In this paper, we present the Cosmos World Foundation Model Platform to help developers build customized world models for their Physical AI setups. We position a world foundation model as a general-purpose world model that can be fine-tuned into customized world models for downstream applications. Our platform covers a video curation pipeline, pre-trained world foundation models, examples of post-training of pre-trained world foundation models, and video tokenizers. To help Physical AI builders solve the most critical problems of our society, we make Cosmos open-source and our models open-weight with permissive licenses available via https://github.com/nvidia-cosmos/cosmos-predict1.