🤗 Hugging Face | Paper | Repository
Cosmos-H-Surgical-Transfer is a multi-ControlNet framework for control-conditioned surgical video generation, part of the Cosmos-H-Surgical suite and the NVIDIA MedTech Open Models.
- [March 16, 2026] As part of the NVIDIA MedTech Open Models, we released Cosmos-H-Surgical-Transfer
Cosmos-H-Surgical-Transfer is a multi-ControlNet framework designed to accept structured inputs from multiple video modalities, including RGB, depth, segmentation, and more. Users can configure generation through JSON-based controlnet_specs and run inference with just a few commands. It supports single-video inference, automatic control map generation, and multi-GPU setups. Cosmos-H-Surgical-Transfer is built upon the Cosmos-Transfer2.5-2B model and is adapted specifically for surgical video data.
Physical AI trains upon data generated in two important data augmentation workflows.
Input prompt:
Real surgery scene: right instrument coagulates the cystic artery.
| Computed Control | Output Video |
|---|---|
coag_seg.mp4See more computed controlscoag_depth.mp4coag_edge.mp4 |
coagulation_m_output.mp4 |
Input prompt:
Real surgery scene: Use the robotic forceps to puncture a needle into a soft tissue structure.
| Input Video | Computed Control | Output Video |
|---|---|---|
guided_input.mp4 |
guided_seg.mp4See more computed controlsguided_mask.mp4 |
guided_ouput.mp4 |
transfer_examples.mp4
Cosmos-H-Surgical-Transfer: General checkpoints, trained from the ground up for Physical AI and robotics for surgical tasks.
We welcome contributions. Check out our Contributing Guide to get started, and share your feedback through issues.
This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.
Cosmos-H-Surgical source code is released under the Apache 2 License.
Cosmos-H-Surgical models are released under the NVIDIA License. You are responsible for ensuring that your use of NVIDIA AI Foundation Models complies with all applicable laws.