SSyncDev

Cosmos3-Nano-Transfer-Auto

SyncDev model directory · Profile updated Nov 19, 2026

Compare with another model

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.

Key specifications

Specification score
19.8
License
OpenMDW-1.1

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.

Inputs and outputs

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

  • Text input
  • Video input
  • Image input

Produces

  • Video output

Capabilities

VisionOpen weights

What is Cosmos3-Nano-Transfer-Auto?

Cosmos3-Nano-Transfer-Auto answers a specific request from autonomous-vehicle teams: give me the same driving scene, but at dusk, in rain, from eleven cameras and a LiDAR, all perfectly in sync. It generates exactly that from structured controls and a text description.

NVIDIA published it on 20 October 2026 among the Cosmos3 example models.

Inputs, briefly

  • A text prompt for weather, lighting and time of day. It also carries the camera configuration.
  • World Scenario Map (WSM) controls: one MP4 per camera showing rendered lane markings and object cuboids.
  • LiDAR controls as HD-map range-map sequences aligned with the cameras, needed when LiDAR is requested.
  • Optional initial camera images and an optional initial measured LiDAR sweep.

Out come one video stream per camera, standard or fisheye, plus LiDAR range and intensity maps that convert to point clouds with per-point intensity. Cameras and LiDAR are synchronised spatially and in time.

Limits worth knowing

Property Value
Camera views 1 to 11, Hyperion 8.1 rig (front-wide 120°, front-tele 30°, cross and rear views, four 200° fisheyes)
LiDAR 128-beam rotating
Length Up to 300 frames per view
Resolution 480p or 720p, 10–30 fps playback
Parameters 16B trainable, Mixture-of-Transformers
Licence OpenMDW-1.1

Other rigs need additional post-training. The WSM controls come from a separate renderer described in the model's quick-start skill.

Offline versus closed-loop

This is the bidirectional model. It denoises the whole multi-view sequence at once, using future context, which suits batch synthetic-data generation but not interactive driving. For closed-loop evaluation NVIDIA distilled it into Cosmos3-Nano-Sim-Auto, which generates causally. Pick Transfer-Auto when quality and cross-view consistency outweigh latency.

Measuring whether the data helps

Do not judge the videos by eye alone. Generate a few hundred clips covering conditions your fleet rarely sees, train your perception model with and without them, and test on real held-out drives in those conditions. Check LiDAR separately: project generated points onto generated frames and look for misaligned objects.

The NVIDIA autonomous-driving stack also includes Alpamayo 2 Edge. For other generators that start from video, browse the video-to-video directory.

How to evaluate Cosmos3-Nano-Transfer-Auto

Compare the model on the work you need to complete.

Tasks and inputs

Cosmos3-Nano-Transfer-Auto is listed for Text-to-video, Video-to-video. It accepts text, video, image and produces video. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Choose duration, resolution and frame rate together. Compare the cost of an accepted clip, including repeat generations.

Choosing a deployment

The weights are downloadable under OpenMDW-1.1. Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

Up to 11 standard and fisheye cameras on the Hyperion 8.1 rig and a 128-beam rotating LiDAR, all synchronised.

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

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.

— context

NVIDIA

Cosmos3-Nano-Sim-Transfer

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.

— context

skywork

Matrix-Game 4.0

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.

— context

Tencent

WeVisCut

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.

— context

skywork

SkyReels-V3-V2V-14B

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.

— context

AInVFX

AInVFX/SeedVR2_comfyUI

AInVFX/SeedVR2_comfyUI is a source-linked Hugging Face repository assigned to Video-to-video. Its recorded task interface accepts video and produces video.

— context