Tasks and inputs
Cosmos3-Nano-Sim-Bimanual is listed for Image-to-video, Video-to-video. It accepts image, text 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-Bimanual is NVIDIA's open 16B world model that simulates video of a two-armed AgiBot robot as it moves its camera and manipulates.
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
Most robot world models handle either the arm or the camera. Cosmos3-Nano-Sim-Bimanual handles both in one rollout. Starting from a single image, it generates what happens as a two-armed robot moves its gripper, changes its arm state and repositions its camera, all within the same action sequence.
That makes it a practical stand-in for teleoperation when you need demonstrations or want to evaluate a policy before it touches hardware.
Released on 20 October 2026 as one of the Cosmos3 examples, it starts from Cosmos3-Nano and is post-trained on a mix of manipulation videos paired with robot actions and navigation videos paired with camera trajectories. Causal distillation then turns it into a chunked, few-step generator that conditions on cached history and incoming controls.
Controls can come from anywhere. A vision-language-action policy, a world-action model or a person on a joystick can supply them at any point during a rollout.
| Field | Format |
|---|---|
| Initial frame | JPG, PNG, JPEG or WebP |
| Robot actions | JSON of shape (T, D), embodiment-specific |
| Camera trajectories | Pose sequence, e.g. 9-D camera motion |
| Text | Prompt string |
| Output | MP4 or streamed frames, 480p at 30 fps |
| Supported robot | AgiBot |
| Size | 16B trainable parameters |
| Licence | OpenMDW-1.1 |
Open-loop and closed-loop operation are both supported.
The single-embodiment restriction is the biggest caveat. If you do not run AgiBot hardware, expect to post-train before the actions mean anything.
Replay a recorded AgiBot episode's actions from its first frame and compare generated frames against the real video every second. Look for gripper state errors first; they are the failure that silently ruins policy evaluation. Then try a scripted camera pan mid-grasp and confirm the object stays where the arm left it.
For simulator-driven photorealism instead of action-driven simulation, Cosmos3-Nano-Sim-Transfer is the sibling to use. Other generators that start from a still frame are in the image-to-video directory, and NVIDIA's robotics models are collected on the provider page.
Compare the model on the work you need to complete.
Cosmos3-Nano-Sim-Bimanual is listed for Image-to-video, Video-to-video. It accepts image, text 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.
tbc
TBC mischief 1 is an image-to-video model built on Wan 2.2, sold on fal and OpenRouter at about half a cent per second for 3-5 s clips at up to 720p.
NVIDIA
Cosmos3-H-Dreams is NVIDIA's open real-time surgical world model: it streams Versius robot video from one endoscope frame and two-arm commands.
NVIDIA
Cosmos3-H-Embodied-Nano is NVIDIA's open 16B surgical world model that does forward dynamics, inverse dynamics and policy from one checkpoint.
noiz-ai
Helix World (preview) is an open real-time world model from Noiz AI and HKUST that streams explorable 768x512 video with stereo sound that follows the camera.
Alibaba
Wan-Animate-2 is Alibaba Wan's open 14B character-animation model that drives a character from video directly, with text-controlled camera angle.
Z.ai / Zhipu AI
SCAIL-2 is Zhipu's open character-animation model that drives a reference character from any video, with no pose skeletons or masks. MIT licence.