Primary use stated or inferred cautiously from official documentation.
N
NVIDIA / nvidia/Cosmos3-Nano-Transfer-Auto
Cosmos3-Nano-Transfer-Auto
NVIDIA bidirectional model for generating driving scenes with synchronized multiview video and LiDAR. Published on October 20, 2026, it combines text, WSM world controls, and an optional initial image. It supports up to eleven standard or fisheye cameras and up to 300 frames per view. It is batch generation, not the causal simulation model Sim-Auto.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | nvidia/Cosmos3-Nano-Transfer-Auto |
|---|---|
| Model type | Robotics · Video · Multimodal |
| Access model | Open weights |
| License | OpenMDW 1.1 |
| Deployment | Self-hosted |
| Release | 20 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image · Video · LiDAR |
| Exit | Video · LiDAR |
| Context window | Text, WSM controls, and optional visual reference |
| maximum output | Up to 11 cameras; 300 frames per view; 480p or 720p |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Multiview video and LiDAR; aligned LiDAR controls are required to generate LiDAR. |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch |
Synthetic driving data with variations in weather, lighting, and environment. WSM controls include lanes, signs, traffic lights, and cuboids. The output supports 480p or 720p and playback between 10 and 30 fps. Maintain consistency across cameras, LiDAR, calibration, and labels; realistic imagery does not guarantee utility for detecting road obstacles.
The comparison with Cosmos-Transfer2.5 uses different prompts and does not demonstrate superiority under identical conditions. T2V and I2V results must also be kept separate. We do not transfer the performance of the causal version to this bidirectional variant. Review rare scenarios, geometry, synchronization, and diversity before incorporating synthetic data into driving systems.
03 / CAPABILITIES
What it can do
Orientation
Synthetic driving data with variations in weather, lighting, and environment. WSM controls include lanes, signs, traffic lights, and cuboids. The output supports 480p or 720p and playback between 10 and 30 fps. Maintain consistency across cameras, LiDAR, calibration, and labels; realistic imagery does not guarantee utility for detecting road obstacles.
Context
Text, WSM controls, and optional visual reference · Up to 11 cameras; 300 frames per view; 480p or 720p
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
TextImageVideoLiDAR
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Public 16B parameter checkpoint for self-hosted infrastructure. We have not verified a commercial endpoint or token rates: numerical costs are left blank. Downloadable weights do not make inference free. Budget for GPU, memory, video encoding, and operation; review the license of each variant and its dependencies. Playback frames per second do not equate to generation performance.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The Driving Simulation Evaluation dataset measures multiview RGB quality and LiDAR generation. We recorded FID and FVD-I3D for T2V and I2V with guidance 3, preserving the mode in each name. Both improve as they decrease. These are provider results at 720p, not accident rates or a road safety validation. Inferama has not reproduced the experiment.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Cosmos3 Driving T2V CFG3 FID ↓ | 45,5 | Score | View source ↗ |
| Cosmos3 Driving T2V CFG3 FVD-I3D ↓ | 488 | Score | View source ↗ |
| Cosmos3 Driving I2V CFG3 FID ↓ | 27,8 | Score | View source ↗ |
| Cosmos3 Driving I2V CFG3 FVD-I3D ↓ | 238 | Score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Cosmos3-Nano-Transfer-Auto
Version added to Inferama's verified catalog.
07 / SOURCES