Tasks and inputs
Cosmos3-Nano-Sim-Transfer is listed for Text-to-video, Video-to-video. It accepts video, text, image and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 19, 2026
Cosmos3-Nano-Sim-Transfer is NVIDIA's open 16B model that renders simulator depth, segmentation or edge maps into photorealistic robot video in real time.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights; no hosted API price. GPU time is the cost, so confirm throughput on your own hardware.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Simulators are good at physics and bad at looking real. Cosmos3-Nano-Sim-Transfer keeps the simulator for the first job and takes over the second. Feed it per-frame depth maps, segmentation masks, edge maps or blurred renders from a tool like Isaac Lab, add a text prompt, and it returns photorealistic RGB video that preserves the geometry and motion the simulator specified.
Because the simulator still owns physics, scene state and ground-truth labels, a policy can train on realistic images while its reward signal stays exact.
| Item | Detail |
|---|---|
| Released | 20 October 2026, Cosmos3 examples family |
| Base | Cosmos3-Nano, post-trained for control and causally distilled |
| Architecture | Mixture-of-Transformers, 16B trainable parameters |
| Inputs | Control video (depth, segmentation, edge or blur) + text |
| Output | Up to 600 frames at 480p, 30 fps playback |
| Batching | Designed for up to eight parallel environments on one GB200 |
| Runtimes | PyTorch, vLLM-Omni, SGLang |
| Licence | OpenMDW-1.1, commercial use allowed |
The repository was opened at the end of August, but the release date given is 20 October.
Unlike offline transfer models, this one generates in small latent-frame chunks conditioned on cached history and the newest controls. That is what makes closed-loop use possible: the next frame can depend on what the policy just did. Throughput depends on batch size and settings, and no single FPS figure is published, so profile it.
The Cosmos3 examples split by domain. Video-to-video transfer for driving lives in Transfer-Auto and its causal sibling Sim-Auto; robot-action simulation in Sim-Bimanual. Sim-Transfer is the general robotics renderer that sits on top of a classical simulator.
Step three is the one that matters. NVIDIA itself notes that simulator annotations describe the simulated scene and their alignment with generated imagery should be validated for each use.
A quick-start skill ships in the cosmos-framework repository. Other NVIDIA physical-AI releases are on the NVIDIA provider page.
Compare the model on the work you need to complete.
Cosmos3-Nano-Sim-Transfer is listed for Text-to-video, Video-to-video. It accepts video, text, image and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.
The weights are downloadable under OpenMDW-1.1. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
NVIDIA
Cosmos3-Nano-Sim-Auto is NVIDIA's open causal driving simulator: real-time multi-camera video and LiDAR that react to an AV policy's planned path.
NVIDIA
Cosmos3-Nano-Transfer-Auto is NVIDIA's open 16B model that generates synchronised multi-camera driving video and LiDAR from map controls and a prompt.
skywork
Matrix-Game 4.0 is Skywork's open set of 5B real-time video renderers that turn a running game's frames into clean styled video, per scene.
Tencent
WeVisCut is Tencent's 9B tool-calling agent that edits narration scripts into multi-minute videos through 12 editing tools and an editable Studio timeline.
skywork
SkyReels-V3-V2V-14B is Skywork's 14B video-extension model that continues a clip for 5-30 seconds or cuts to a new shot, at 720p from a text instruction.
AInVFX
AInVFX/SeedVR2_comfyUI is a source-linked Hugging Face repository assigned to Video-to-video. Its recorded task interface accepts video and produces video.