# cosmos | NVIDIA Corporation | Knowledge Base | Instagit

NVIDIA Cosmos is an open platform of world models, datasets, and tools that enables developers to build Physical AI for robots, autonomous vehicles, smart infrastructure, and more.

GitHub Stars: 9.3k

Repository: https://github.com/NVIDIA/cosmos

---

## Articles

### [Comparing Cosmos 3 Reasoner versus Generator Modes for Different Use Cases](/NVIDIA/cosmos/comparing-cosmos-3-reasoner-versus-generator-modes-for-different-use-cases)

Explore NVIDIA Cosmos 3's Reasoner and Generator modes. Learn which mode suits vision-language understanding and media creation tasks for optimal results.

- Tags: comparisons
- Published: 2026-07-03

### [How to Use Cosmos 3 for Embodied Reasoning and Next-Action Prediction](/NVIDIA/cosmos/using-cosmos-3-for-embodied-reasoning-and-next-action-prediction)

Leverage Cosmos 3 an omnimodal world model for embodied reasoning and next-action prediction. Use its causal transformer Reasoner surface to plan tasks from video and text inputs.

- Tags: how-to-guide
- Published: 2026-07-03

### [Running Supervised Fine-Tuning on Cosmos 3 Vision Generators: A Complete Guide](/NVIDIA/cosmos/running-supervised-fine-tuning-on-cosmos-3-vision-generators)

Learn to run supervised fine-tuning on Cosmos 3 vision generators with NVIDIA's cookbook scripts. Automate dataset prep, multi-GPU training, and checkpoint export for your custom video data.

- Tags: how-to-guide
- Published: 2026-07-03

### [Implementing Video-to-Video Transformation with Prompt Guidance in Cosmos 3](/NVIDIA/cosmos/implementing-video-to-video-transformation-with-prompt-guidance-in-cosmos-3)

Learn how to implement video-to-video transformation with prompt guidance in Cosmos 3. This guide covers using textual prompts and spatial controls for powerful video generation.

- Tags: how-to-guide
- Published: 2026-07-03

### [Using Cosmos 3 for Physical Plausibility Analysis and Classification: A Complete Guide](/NVIDIA/cosmos/using-cosmos-3-for-physical-plausibility-analysis-and-classification)

Learn how to use Cosmos 3 for physical plausibility analysis and classification. This guide covers the omni-model's two-step workflow for video synthesis and LLM-vision classification.

- Tags: how-to-guide
- Published: 2026-07-03

### [Troubleshooting CUDA Driver Compatibility with PyTorch Backends for Cosmos 3](/NVIDIA/cosmos/troubleshooting-cuda-driver-compatibility-with-pytorch-backends-for-cosmos-3)

Fix CUDA driver compatibility errors in Cosmos 3 with PyTorch. Ensure your CUDA wheel tag matches your NVIDIA driver for successful GPU acceleration. Avoid torch.cuda.is_available() failing.

- Tags: tutorial
- Published: 2026-07-03

### [Implementing 2D Grounding and Bounding Box Localization with Cosmos 3](/NVIDIA/cosmos/implementing-2d-grounding-and-bounding-box-localization-with-cosmos-3)

Implement 2D grounding and bounding box localization with Cosmos 3. Our Reasoner surface converts image-text prompts into JSON bounding boxes using a Mixture-of-Transformers.

- Tags: how-to-guide
- Published: 2026-07-03

### [Understanding mRoPE Position Embedding for Multimodal Reasoning in Cosmos 3](/NVIDIA/cosmos/understanding-mrope-position-embedding-for-multimodal-reasoning-in-cosmos-3)

Explore mRoPE position embedding for multimodal reasoning in NVIDIA Cosmos 3. Learn how this 3-axis embedding unifies vision, audio, and robot actions in transformers.

- Tags: deep-dive
- Published: 2026-07-03

### [How to Configure Video Resolution, Frame Rate, and Aspect Ratio in Cosmos 3](/NVIDIA/cosmos/configuring-video-resolution-frame-rate-and-aspect-ratio-settings-in-cosmos-3)

Learn to configure video resolution, frame rate, and aspect ratio in NVIDIA Cosmos 3. Control these settings using environment variables or CLI arguments for optimal video generation.

- Tags: how-to-guide
- Published: 2026-07-03

### [Running Inverse Dynamics Predictions for Autonomous Vehicle Ego-Motion with Cosmos 3](/NVIDIA/cosmos/running-inverse-dynamics-predictions-for-autonomous-vehicle-ego-motion-with-cosmos-3)

Learn to run inverse dynamics predictions for autonomous vehicle ego-motion with Cosmos 3. Estimate 9-DOF vehicle pose deltas from video using PyTorch or REST APIs.

- Tags: how-to-guide
- Published: 2026-07-03

### [How to Set Up the Guardrail Safety Checker for Cosmos 3 Generator](/NVIDIA/cosmos/setting-up-guardrail-safety-checker-for-cosmos-3-generator)

Learn to set up the Guardrail safety checker for the NVIDIA Cosmos 3 Generator. Easily enable or disable this essential feature for secure AI model generation.

- Tags: how-to-guide
- Published: 2026-07-03

### [Configuring Action Policy Predictions for DROID Robot Embodiment in Cosmos 3](/NVIDIA/cosmos/configuring-action-policy-predictions-for-droid-robot-embodiment-in-cosmos-3)

Learn to configure DROID robot action policy predictions in Cosmos 3. Set up the environment, register the experiment, and run the training script for seamless embodiment.

- Tags: how-to-guide
- Published: 2026-07-03

### [Configuring Tensor Parallelism for Cosmos3-Super Inference on Multiple GPUs](/NVIDIA/cosmos/configuring-tensor-parallelism-for-cosmos3-super-inference-on-multiple-gpus)

Configure tensor parallelism for Cosmos3-Super inference on multiple GPUs. Use vLLM-Omni and set tensor parallel size to shard weights and reduce memory usage by 75%.

- Tags: how-to-guide
- Published: 2026-07-03

### [Understanding the Mixture-of-Transformers Architecture in Cosmos 3](/NVIDIA/cosmos/understanding-mixture-of-transformers-architecture-in-cosmos-3)

Explore Cosmos 3's Mixture-of-Transformers architecture. Learn how a unified transformer backbone with dual heads enables simultaneous text reasoning and multimodal generation for images, video, audio, and actions.

- Tags: deep-dive
- Published: 2026-07-03

### [How to Use NVIDIA Cosmos 3 Reasoner for Video Captioning and Temporal Localization](/NVIDIA/cosmos/using-cosmos-3-reasoner-for-video-captioning-and-temporal-localization)

Learn to use NVIDIA Cosmos 3 Reasoner for video captioning and temporal localization. Process video inputs and generate detailed captions or structured temporal events with this powerful transformer architecture.

- Tags: how-to-guide
- Published: 2026-07-03

### [Setting Up Cosmos 3 Reasoner with vLLM for Production Inference](/NVIDIA/cosmos/setting-up-cosmos-3-reasoner-with-vllm-for-production-inference)

Deploy Cosmos 3 Reasoner with vLLM for production inference. Install the vllm-cosmos3 plugin, override the vLLM architecture, and configure I/O paths for seamless deployment.

- Tags: how-to-guide
- Published: 2026-07-03

### [How to Implement Custom Evaluation Metrics for Cosmos 3 Generated Videos Using Cosmos Evaluator](/NVIDIA/cosmos/implement-custom-evaluation-metrics-cosmos-3-cosmos-evaluator)

Learn how to implement custom evaluation metrics for Cosmos 3 videos with Cosmos Evaluator. Extend functionality by creating Python classes and registering them in metrics yaml for tailored video scoring.

- Tags: how-to-guide
- Published: 2026-06-14

### [Best Practices for Running Cosmos 3 in Docker Containers with NVIDIA NGC Images](/NVIDIA/cosmos/running-cosmos-3-docker-containers-ngc-images)

Run Cosmos 3 in Docker with NVIDIA NGC images. Optimize GPU performance using vLLM-Omni or NIM, authenticate to nvcr.io, and launch with key Docker flags for seamless deployment.

- Tags: best-practices
- Published: 2026-06-14

### [How to Set Up Multi-GPU Inference with Tensor-Parallel-Size, CFG-Parallel-Size, and Ulysses-Degree in Cosmos 3](/NVIDIA/cosmos/setup-multi-gpu-inference-cosmos-3-options)

Learn to set up multi-GPU inference with Cosmos 3 using tensor-parallel-size, cfg-parallel-size, and ulysses-degree. Optimize performance by multiplying these flags for total GPU count.

- Tags: how-to-guide
- Published: 2026-06-14

### [Security and Safety Considerations When Deploying Cosmos 3 in Production](/NVIDIA/cosmos/security-safety-considerations-deploying-cosmos-3-production)

Deploy Cosmos 3 securely in production. Learn essential guardrails, input sandboxing, output validation, and container best practices for NVIDIA NIM and vLLM-Omni. Ensure authentication and CUDA compatibility.

- Tags: best-practices
- Published: 2026-06-14

### [How to Configure UniPCMultistepScheduler with flow_shift Parameters for Optimal Cosmos Generation Quality](/NVIDIA/cosmos/configure-unipcmultistepscheduler-flow-shift-parameters)

Optimize Cosmos video generation quality by configuring UniPCMultistepScheduler with flow_shift parameters. Learn how to control temporal smoothing and enhance your output.

- Tags: how-to-guide
- Published: 2026-06-14

### [Forward Dynamics, Inverse Dynamics, and Policy Action Modes in Cosmos 3: Architecture and API Differences](/NVIDIA/cosmos/differences-forward-inverse-dynamics-policy-modes-cosmos-3)

Explore Cosmos 3 architecture and API differences. Understand forward dynamics, inverse dynamics, and policy action modes to generate video, extract actions, and predict future moves.

- Tags: architecture
- Published: 2026-06-14

### [How to Implement Video Transfer with Edge Detection, Blur, Depth Estimation, and Segmentation Controls in NVIDIA Cosmos](/NVIDIA/cosmos/implement-video-transfer-controls-cosmos-3)

Implement video transfer with edge detection blur depth estimation and segmentation using NVIDIA Cosmos. Control spatial maps for advanced video generation with the Cosmos Framework.

- Tags: how-to-guide
- Published: 2026-06-14

### [Cosmos 3 Supported Video Resolutions and Frame Rates: Complete Generation Settings Guide](/NVIDIA/cosmos/supported-video-resolutions-frame-rates-cosmos-3)

Explore Cosmos 3 video resolutions and frame rates. Discover settings for 256p, 480p, 720p, and frame rates up to 30 FPS. Optimize your generation settings today.

- Tags: api-reference
- Published: 2026-06-14

### [How Prompt Upsampling Works in NVIDIA Cosmos 3 Generator for Expanding Scene Descriptions](/NVIDIA/cosmos/how-prompt-upsampling-works-cosmos-3-generator)

Discover how prompt upsampling in NVIDIA Cosmos 3 expands scene descriptions by loading pre-computed LLM-expanded details for richer textual inputs during generation.

- Tags: deep-dive
- Published: 2026-06-14

### [Cosmos3-Nano vs Cosmos3-Super vs Cosmos3-Super-Text2Image: Complete Model Comparison](/NVIDIA/cosmos/differences-cosmos3-nano-super-text2image-models)

Compare Cosmos3-Nano, Cosmos3-Super, and Cosmos3-Super-Text2Image models from NVIDIA/cosmos. Understand parameter counts, GPU requirements, and use cases to choose the best fit for your project.

- Tags: deep-dive
- Published: 2026-06-14

### [How to Integrate Cosmos 3 with Custom Robotics Embodiments: A Complete Guide](/NVIDIA/cosmos/integrate-cosmos-3-custom-robotics-embodiments)

Integrate Cosmos 3 with custom robotics embodiments by defining domain name, raw action dim, and passing action trajectories via extra_params. No core model modifications needed. Get the complete guide from NVIDIA/cosmos.

- Tags: how-to-guide
- Published: 2026-06-14

### [Best Practices for Fine-Tuning Cosmos 3 Models Using the Cosmos Framework](/NVIDIA/cosmos/best-practices-fine-tuning-cosmos-3-cosmos-framework)

Master fine-tuning Cosmos 3 models with the NVIDIA Cosmos Framework. Learn best practices for data curation, mixed precision training, and safety validation to optimize your models.

- Tags: best-practices
- Published: 2026-06-14

### [How to Optimize Memory Usage for Cosmos 3 Super on Multiple GPUs with Layer Offloading](/NVIDIA/cosmos/optimize-memory-usage-cosmos-3-multi-gpu-layer-offloading)

Optimize Cosmos 3 Super GPU memory with layer offloading and tensor parallelism. Stream transformer blocks from CPU RAM to reduce per-GPU needs from 90GB.

- Tags: performance
- Published: 2026-06-14

### [How Multi-Dimensional Rotary Position Embedding (mRoPE) Encodes Spatial-Temporal Structure Across Modalities in NVIDIA Cosmos](/NVIDIA/cosmos/how-mrope-encodes-spatial-temporal-structure)

Discover how NVIDIA Cosmos leverages mRoPE to encode spatial-temporal structure across modalities like image, video, and audio. Understand its unified 3-D approach for consistent coordinates.

- Tags: deep-dive
- Published: 2026-06-14

### [How to Resolve CUDA Driver Mismatch Errors with Cosmos 3: Fixing `torch.cuda.is_available()` is False](/NVIDIA/cosmos/resolve-cuda-driver-mismatch-errors-cosmos-3)

Fix CUDA driver mismatch errors in Cosmos 3, especially when torch.cuda.is_available() is False, by reinstalling PyTorch with the correct `--torch-backend` flag. Ensure CUDA compatibility.

- Tags: how-to-guide
- Published: 2026-06-14

### [Diffusers vs vLLM-Omni vs Cosmos Framework: NVIDIA Cosmos Inference Backends Explained](/NVIDIA/cosmos/difference-diffusers-vllm-omni-cosmos-framework-backends)

Explore NVIDIA Cosmos inference backends: Diffusers for prototyping, vLLM-Omni for high throughput, and Cosmos Framework for research. Optimize your generative AI workflow.

- Tags: deep-dive
- Published: 2026-06-14

### [How to Implement and Disable Guardrails for Content Filtering in Cosmos 3 Generation](/NVIDIA/cosmos/implement-disable-guardrails-content-filtering-cosmos-3)

Implement and disable Cosmos 3 guardrails for content filtering using the cosmos_guardrail package. Control safety features per-request or server-wide for generated media.

- Tags: how-to-guide
- Published: 2026-06-14

### [Cosmos 3 World Models Failure Modes and Limitations: A Technical Guide](/NVIDIA/cosmos/failure-modes-limitations-cosmos-3-world-models)

Explore Cosmos 3 world models common failure modes including temporal inconsistency and physics issues. Learn about essential guardrails and validation for production.

- Tags: technical-guide
- Published: 2026-06-14

### [How to Configure Tensor Parallelism and CFG Parallelism for Cosmos 3 Super Model Inference](/NVIDIA/cosmos/configure-tensor-parallelism-cfg-parallelism-cosmos-3-super)

Learn to configure tensor parallelism and CFG parallelism for Cosmos 3 Super model inference. Optimize inference speed by sharding models across GPUs and accelerating model guided diffusion.

- Tags: how-to-guide
- Published: 2026-06-14

### [Supported Action Dimensions for Different Robot Embodiments in Cosmos 3 Action Models](/NVIDIA/cosmos/supported-action-dimensions-robot-embodiments-cosmos-3)

Explore supported action dimensions for Cosmos 3 robot embodiments. Discover how raw action dimensions from 9D to 57D enable single MoT models for diverse robotic applications.

- Tags: api-reference
- Published: 2026-06-14

### [How to Optimize Inference Latency for Cosmos 3 Text-to-Video Generation with vLLM-Omni](/NVIDIA/cosmos/optimize-inference-latency-cosmos-3-text-to-video-vllm-omni)

Slash Cosmos 3 text-to-video inference latency under 8 seconds by optimizing with vLLM-Omni on multi-GPU Hopper systems. Learn how now!

- Tags: performance
- Published: 2026-06-14

### [How the Mixture-of-Transformers (MoT) Architecture in Cosmos 3 Unifies Autoregressive Reasoning and Diffusion Generation](/NVIDIA/cosmos/how-does-mixture-of-transformers-mot-architecture-work-cosmos-3)

Discover how NVIDIA Cosmos 3's Mixture-of-Transformers (MoT) unifies autoregressive reasoning and diffusion generation using a single transformer for advanced multimodal applications.

- Tags: deep-dive
- Published: 2026-06-14

### [Diffusers vs vLLM-Omni for Cosmos 3 Generator: A Complete Integration Guide](/NVIDIA/cosmos/comparing-diffusers-vs-vllm-omni-for-cosmos-3-generator)

Explore Diffusers vs vLLM Omni for NVIDIA Cosmos 3 Generator. Choose the best runtime for research or production scale serving with this complete integration guide.

- Tags: comparison
- Published: 2026-06-13

### [Troubleshooting vLLM DeepGEMM Errors with Cosmos 3: A Complete Guide](/NVIDIA/cosmos/troubleshooting-vllm-deepgemm-errors-with-cosmos-3)

Troubleshoot vLLM DeepGEMM errors with Cosmos 3 by disabling DeepGEMM via an environment variable. Learn how to resolve compatibility warnings and maintain inference performance.

- Tags: how-to-guide
- Published: 2026-06-13

### [Cosmos 3 Guardrails Configuration: How to Disable Safety Features in NVIDIA Cosmos](/NVIDIA/cosmos/cosmos-3-guardrails-configuration-and-disabling-safety-features)

Learn to disable NVIDIA Cosmos 3 guardrails. Explore three methods: per-request, command-line flag, or server-wide YAML configuration to manage safety features effectively.

- Tags: how-to-guide
- Published: 2026-06-13

### [Using Cosmos 3 Reasoner for Video Captioning and Temporal Event Localization](/NVIDIA/cosmos/using-cosmos-3-reasoner-for-video-captioning-and-temporal-event-localization)

Leverage the Cosmos 3 Reasoner for advanced video captioning and temporal event localization. Get structured text or JSON outputs via an OpenAI-compatible API.

- Tags: how-to-guide
- Published: 2026-06-13

### [Configuring Tensor Parallelism for Cosmos 3 Super on Multi-GPU Setups](/NVIDIA/cosmos/configuring-tensor-parallelism-for-cosmos-3-super-on-multi-gpu-setups)

Discover how to configure tensor parallelism for Cosmos 3 Super on multi-GPU setups. Reduce memory usage by sharding model weights and synchronizing activations with the tensor parallelism flag. Learn more.

- Tags: how-to-guide
- Published: 2026-06-13

### [Cosmos 3 Action Policy Models for Robot Manipulation: Architecture and Deployment](/NVIDIA/cosmos/cosmos-3-action-policy-models-for-robot-manipulation-tasks)

Explore Cosmos 3 action policy models for robot manipulation. Convert vision-language prompts into executable robot trajectories with the NVIDIA/cosmos repository. Deploy DROID arm joint commands from scene images and text inst...

- Tags: architecture
- Published: 2026-06-13

### [Implementing Forward Dynamics with Cosmos 3 for Robotics Simulation: A Complete Technical Guide](/NVIDIA/cosmos/implementing-forward-dynamics-with-cosmos-3-for-robotics-simulation)

Implement forward dynamics with Cosmos 3 for robotics simulation. This guide shows how to use the forward_dynamics action mode to predict future visual observations, enabling physics-aware simulations.

- Tags: how-to-guide
- Published: 2026-06-13

### [How to Use Cosmos 3 for Text-to-Video Generation with Synchronized Audio Output](/NVIDIA/cosmos/how-to-use-cosmos-3-for-text-to-video-generation-with-synchronized-audio-output)

Learn to use Cosmos 3 for text-to-video generation with synchronized audio. Discover the Mixture-of-Transformers architecture and how to enable sound via Cosmos3OmniPipeline or vLLM-Omni API.

- Tags: how-to-guide
- Published: 2026-06-13

### [Cosmos 3 vs Cosmos 3 Super: Choosing the Right Model Size for Your Use Case](/NVIDIA/cosmos/cosmos-3-vs-cosmos-3-super-choosing-the-right-model-size-for-your-use-case)

Compare Cosmos 3 Nano and Cosmos 3 Super model sizes. Find the right NVIDIA Cosmos model for your use case, from rapid prototyping to production ready visual fidelity.

- Tags: deep-dive
- Published: 2026-06-13

### [How to Deploy Cosmos 3 Reasoner Using vLLM for Production Inference](/NVIDIA/cosmos/how-to-deploy-cosmos-3-reasoner-using-vllm-for-production-inference)

Deploy Cosmos 3 Reasoner with vLLM for production inference. Install the plugin, launch the server, and expose an OpenAI-compatible endpoint for efficient multimodal processing.

- Tags: how-to-guide
- Published: 2026-06-13

### [How to Integrate Cosmos 3 Generator with Diffusers for Research and Development](/NVIDIA/cosmos/how-to-integrate-cosmos-3-generator-with-diffusers-for-research-and-development)

Integrate NVIDIA Cosmos 3 Generator with Diffusers for research and development. Use Cosmos3OmniDiffusersPipeline for text to image, video generation without Docker or custom CUDA kernels.

- Tags: how-to-guide
- Published: 2026-06-13

### [How to Resolve libxcb.so.1 Import Failures on Headless Servers for NVIDIA Cosmos](/NVIDIA/cosmos/how-to-resolve-libxcb-so-1-import-failures-headless-servers)

Fix libxcb.so.1 import errors on headless servers. Install essential X11 libraries with apt-get to resolve Cosmos pipeline loading issues.

- Tags: how-to-guide
- Published: 2026-06-12

### [How to Integrate Cosmos 3 with Custom Robot Embodiments: DROID, UMI, and Bridge Guide](/NVIDIA/cosmos/how-to-integrate-cosmos-3-custom-robot-embodiments-droid-umi-bridge)

Integrate Cosmos 3 with custom robot embodiments like DROID UMI and Bridge. Learn to specify domain_name and raw_action_dim for action tokenization.

- Tags: how-to-guide
- Published: 2026-06-12

### [Cosmos 3 Generation Failure Modes: Common Limitations and Architectural Constraints](/NVIDIA/cosmos/cosmos-3-generation-failure-modes-limitations)

Explore common Cosmos 3 generation failure modes including temporal flickering and misalignment. Understand limitations stemming from its diffusion architecture and architectural constraints.

- Tags: deep-dive
- Published: 2026-06-12

### [How to Use Temporal Localization for Event Detection in Videos with NVIDIA Cosmos](/NVIDIA/cosmos/how-to-use-temporal-localization-event-detection-videos)

Learn temporal localization for event detection in videos using NVIDIA Cosmos. Explore multimodal rotary position embeddings and MoT backbone for timestamped JSON event segments.

- Tags: how-to-guide
- Published: 2026-06-12

### [How to Configure CFG Parallelism and Ulysses Sequence Parallelism in NVIDIA Cosmos](/NVIDIA/cosmos/how-to-configure-cfg-parallelism-ulysses-sequence-parallelism)

Learn how to configure CFG parallelism and Ulysses sequence parallelism in NVIDIA Cosmos for efficient model serving. Optimize performance with vLLM Omni.

- Tags: how-to-guide
- Published: 2026-06-12

### [How to Use the Describe Anything Feature for Attribute Captioning in NVIDIA Cosmos](/NVIDIA/cosmos/how-to-use-describe-anything-feature-attribute-captioning)

Learn how to use the Describe Anything feature in NVIDIA Cosmos to generate structured JSON captions for image subjects. This guide explains the capability JSON file and inference module for attribute captioning.

- Tags: how-to-guide
- Published: 2026-06-12

### [How to Enable Sound Generation with Video Output in NVIDIA Cosmos](/NVIDIA/cosmos/how-to-enable-sound-generation-with-video-output-cosmos)

Discover how to enable sound generation with video output in NVIDIA Cosmos. Learn to set enable_sound=True or generate_sound=true for dynamic multimedia experiences.

- Tags: how-to-guide
- Published: 2026-06-12

### [Why `torch.cuda.is_available()` Returns False After Installing PyTorch: A Technical Guide](/NVIDIA/cosmos/torch-cuda-is-available-false-after-pytorch-install)

Fix torch.cuda.is_available() returning False after PyTorch install. Learn how to resolve driver CUDA version mismatches for seamless GPU acceleration.

- Tags: how-to-guide
- Published: 2026-06-12

### [How to Configure Tensor Parallelism for Cosmos3-Super on Multiple GPUs](/NVIDIA/cosmos/how-to-configure-tensor-parallelism-for-cosmos3-super-on-multiple-gpus)

Configure tensor parallelism for Cosmos3-Super across multiple GPUs. Set tensor parallel size to your GPU count to reduce memory usage on each device.

- Tags: how-to-guide
- Published: 2026-06-12

### [Reasoner vs Generator Runtime Surfaces in NVIDIA Cosmos: Architecture and Usage](/NVIDIA/cosmos/difference-between-reasoner-and-generator-runtime-surfaces-in-cosmos)

Understand the difference between NVIDIA Cosmos Reasoner and Generator runtimes. Explore their distinct attention mechanisms and diverse output capabilities for multimodal applications.

- Tags: architecture
- Published: 2026-06-12

### [How to Implement Chain-of-Thought Reasoning with the Cosmos Reasoner Using the Format Instruction](/NVIDIA/cosmos/implement-chain-of-thought-reasoning-reasoner-format-instruction)

Effortlessly implement chain-of-thought reasoning with NVIDIA Cosmos Reasoner. Master the format instruction for structured, detailed answers directly from the Qwen-3-VL message format. Get started now.

- Tags: how-to-guide
- Published: 2026-06-06

### [How to Handle Out-of-Memory Errors When Running Large Models Like Cosmos3-Super](/NVIDIA/cosmos/handle-out-of-memory-errors-running-large-models-cosmos3-super)

Tackle out-of-memory errors with Cosmos3-Super. Learn to implement tensor parallelism, CPU offloading, and adjust timeouts for seamless large model deployment.

- Tags: how-to-guide
- Published: 2026-06-06

### [Cosmos 3 Resolution, Aspect Ratio, and Frame Rate Settings: Complete Configuration Guide](/NVIDIA/cosmos/recommended-resolution-aspect-ratio-frame-rate-cosmos-3-use-cases)

Optimize your Cosmos 3 performance with recommended resolution, aspect ratio, and frame rate settings. Discover ideal configurations for various use cases and unlock the full potential of NVIDIA/cosmos.

- Tags: how-to-guide
- Published: 2026-06-06

### [How to Use the OpenAI-Compatible API with Local Media Files via `--allowed-local-media-path` in NVIDIA Cosmos](/NVIDIA/cosmos/use-openai-compatible-api-local-media-files-allowed-local-media-path)

Learn to use the OpenAI compatible API with local media files in NVIDIA Cosmos. Set the --allowed-local-media-path flag to enable file:// URL access for your media.

- Tags: how-to-guide
- Published: 2026-06-06

### [How to Integrate Cosmos 3 with Custom Robotics Frameworks Using Action Inputs](/NVIDIA/cosmos/integrate-cosmos-3-custom-robotics-frameworks-action-inputs)

Integrate Cosmos 3 with custom robotics frameworks using action inputs. Leverage its OpenAI-compatible REST endpoint to stream joint commands to any HTTP-enabled framework.

- Tags: how-to-guide
- Published: 2026-06-06

### [Common Failure Modes and Limitations of Cosmos 3 Outputs: 7 Critical Issues](/NVIDIA/cosmos/common-failure-modes-limitations-cosmos-3-outputs)

Discover common failure modes of Cosmos 3 outputs including temporal flickering, unstable motion, and depth errors. Learn about limitations caused by its Mixture-of-Transformers architecture.

- Tags: deep-dive
- Published: 2026-06-06

### [How to Optimize Inference Latency and Throughput for Production Generator Workloads in Cosmos 3](/NVIDIA/cosmos/optimize-inference-latency-throughput-production-generator-workloads-cosmos-3)

Optimize generator inference latency and throughput in Cosmos 3. Leverage vLLM-Omni, multi-GPU tensor parallelism, and concurrent requests for faster production workloads.

- Tags: performance
- Published: 2026-06-06

### [How to Use Video-to-Video Transformation with Prompt-Guided Conditioning in vLLM-Omni](/NVIDIA/cosmos/use-video-to-video-transformation-prompt-guided-conditioning-vllm-omni)

Learn video-to-video transformation with prompt-guided conditioning in vLLM-Omni. Regenerate frames using POST requests to the sync endpoint.

- Tags: tutorial
- Published: 2026-06-06

### [What Is mRoPE and How Does It Encode Spatial-Temporal Structure in Cosmos 3?](/NVIDIA/cosmos/what-is-mrope-encode-spatial-temporal-structure-cosmos-3-modalities)

Discover mRoPE in NVIDIA Cosmos 3 a 3D rotary position embedding that encodes spatial temporal coordinates to unify language images video audio and robot actions in a single transformer.

- Tags: deep-dive
- Published: 2026-06-06

### [How to Enable CFG Parallelism in Cosmos 3: Run Positive and Negative Branches on Separate GPUs](/NVIDIA/cosmos/enable-cfg-parallelism-positive-negative-branches-separate-gpus)

Enable CFG Parallelism in Cosmos 3 to run positive and negative branches on separate GPUs. Slash generation latency by half using --cfg-parallel-size 2 with no extra memory costs.

- Tags: how-to-guide
- Published: 2026-06-06

### [Supported Action Dimensions for Robot Embodiments in NVIDIA Cosmos](/NVIDIA/cosmos/supported-action-dimensions-robot-embodiments-camera-motion-egocentric-humanoid)

Explore supported action dimensions for robot embodiments in NVIDIA Cosmos. Discover 6-DOF camera motion to 10-DOF UMI robots for your projects.

- Tags: api-reference
- Published: 2026-06-06

### [How Prompt Upsampling Works in the NVIDIA Cosmos 3 Generator: Default Parameters Explained](/NVIDIA/cosmos/cosmos-3-generator-prompt-upsampling-how-it-works-default-parameters)

Discover how NVIDIA Cosmos 3 uses prompt upsampling to expand short scene descriptions into detailed prompts. Learn about default parameters including max_tokens, temperature, top_p, and seed.

- Tags: deep-dive
- Published: 2026-06-06

### [How to Troubleshoot `libxcb.so.1` Import Failures on Headless Servers with NVIDIA Cosmos](/NVIDIA/cosmos/troubleshoot-libxcb-so-1-import-failures-headless-servers)

Troubleshoot libxcb.so.1 import errors on headless servers. Install essential X11 libraries with apt-get to fix NVIDIA Cosmos pipeline issues in Linux or Docker.

- Tags: how-to-guide
- Published: 2026-06-06

### [Why NVIDIA Cosmos First Run Inference Takes Long (And Why It's Not a Hang)](/NVIDIA/cosmos/long-inference-times-first-run-pipeline-hung-troubleshooting)

NVIDIA Cosmos first run inference takes time due to large model downloads and diffusion. Learn why this expected behavior isn't a pipeline hang.

- Tags: performance
- Published: 2026-06-06

### [How to Configure Action Conditioning for Robot Embodiments in NVIDIA Cosmos](/NVIDIA/cosmos/configure-action-conditioning-robot-embodiments-droid-umi-autonomous-vehicles)

Configure action conditioning for robot embodiments like DROID or UMI in NVIDIA Cosmos. Learn JSON array dimensionality for 10-D robots, autonomous vehicles, and egocentric motion.

- Tags: how-to-guide
- Published: 2026-06-06

### [vLLM-Omni vs Diffusers for Generator Inference in Cosmos 3: Architecture, Performance, and Code Examples](/NVIDIA/cosmos/difference-vllm-omni-vs-diffusers-generator-inference)

Compare vLLM-Omni and Diffusers for generator inference in Cosmos 3. Discover performance differences, architectures, and code examples for production and research.

- Tags: deep-dive
- Published: 2026-06-06

### [How to Use Forward Dynamics for Robot Trajectory Prediction with Action Conditioning in Cosmos 3](/NVIDIA/cosmos/use-forward-dynamics-robot-trajectory-prediction-action-conditioning)

Learn to predict robot trajectories using forward dynamics and action conditioning in Cosmos 3. Achieve physics-aware predictions without simulators and enhance your robotics projects.

- Tags: how-to-guide
- Published: 2026-06-06

### [Which CUDA Version Should I Use and How to Match It With Your NVIDIA Driver for Cosmos 3](/NVIDIA/cosmos/recommended-cuda-version-nvidia-driver-match)

Find the right CUDA version for Cosmos 3 and NVIDIA drivers. Match PyTorch or vLLM CUDA builds to your driver using COSMOS3_UV_GROUP to avoid runtime failures.

- Tags: how-to-guide
- Published: 2026-06-06

### [How to Disable Safety Guardrails in Cosmos 3 Generation Requests](/NVIDIA/cosmos/disable-safety-guardrails-cosmos-3-generation-requests)

Learn how to disable safety guardrails in Cosmos 3 generation requests using JSON, CLI, or config files for unrestricted creative output. Maximize your AI model's potential.

- Tags: how-to-guide
- Published: 2026-06-06

### [Why `torch.cuda.is_available()` Returns False After Installing PyTorch and How to Fix It](/NVIDIA/cosmos/torch-cuda-is-available-false-after-pytorch-install-fix)

Fix `torch.cuda.is_available()` returning False. Learn why and how to reinstall PyTorch with the correct CUDA version to enable GPU support in your NVIDIA/cosmos project.

- Tags: how-to-guide
- Published: 2026-06-06

### [How to Configure Tensor Parallelism for Cosmos3-Super on Multiple GPUs to Reduce Memory Usage](/NVIDIA/cosmos/configure-tensor-parallelism-cosmos3-super-multiple-gpus-reduce-memory)

Configure tensor parallelism for Cosmos3-Super on multiple GPUs with vllm serve. Set the --tensor-parallel-size flag to shard weights and reduce VRAM usage. Optimize your LLM deployment now.

- Tags: how-to-guide
- Published: 2026-06-06

### [Difference Between Reasoner and Generator Runtime Surfaces in Cosmos 3](/NVIDIA/cosmos/difference-between-reasoner-and-generator-runtime-surfaces-cosmos-3)

Discover the key differences between Cosmos 3's Reasoner and Generator runtime surfaces. Understand how Reasoner generates text and Generator creates vision, audio, and action outputs.

- Tags: deep-dive
- Published: 2026-06-06

### [How the Mixture-of-Transformers (MoT) Architecture in Cosmos 3 Unifies Reasoning and Generation](/NVIDIA/cosmos/how-does-mixture-of-transformers-mot-architecture-work-in-cosmos-3-for-reasoning-vs-generation)

Explore the Mixture-of-Transformers (MoT) in Cosmos 3 uniting causal reasoning and multimodal generation with AR and DM transformer heads via a shared backbone and mRoPE.

- Tags: architecture
- Published: 2026-06-06

### [Cosmos 3 Troubleshooting: Fix libxcb.so.1 Cannot Open Shared Object File](/NVIDIA/cosmos/cosmos-3-troubleshooting-libxcb-so-1-cannot-open-shared-object)

Troubleshoot Cosmos 3 import errors by installing missing libxcb.so.1 libraries. Resolve Shared Object File issues quickly and get your pipeline running.

- Tags: how-to-guide
- Published: 2026-06-05

### [How to Use Image-to-Video Generation with Conditioning Frames in Cosmos 3](/NVIDIA/cosmos/how-to-use-image-to-video-generation-with-conditioning-frames-cosmos-3)

Learn to generate video from images using conditioning frames in Cosmos 3 through Pipelines, APIs, or CLI. Explore advanced image-to-video techniques with NVIDIA's Cosmos 3 framework.

- Tags: how-to-guide
- Published: 2026-06-05

### [How to Disable Safety Guardrails in Cosmos 3 Generation](/NVIDIA/cosmos/how-to-disable-safety-guardrails-in-cosmos-3-generation)

Learn how to disable Cosmos 3 safety guardrails for single requests, CLI, or server launches. Gain full control over your Cosmos 3 generation model. Fast and easy.

- Tags: how-to-guide
- Published: 2026-06-05

### [How to Run the Cosmos 3 Reasoner with vLLM for Video Understanding](/NVIDIA/cosmos/how-to-run-reasoner-with-vllm-for-video-understanding)

Learn how to run the Cosmos 3 Reasoner with vLLM for efficient video understanding. Get an OpenAI-compatible API for multimodal inference with video URLs.

- Tags: how-to-guide
- Published: 2026-06-05

### [Cosmos 3 Supported Resolutions and Frame Rates: Complete Technical Guide](/NVIDIA/cosmos/cosmos-3-supported-resolutions-and-frame-rates)

Discover Cosmos 3 supported resolutions and frame rates. Explore 256p, 480p, 720p resolutions and up to 30 FPS for your NVIDIA/cosmos projects. Get the technical details.

- Tags: technical-guide
- Published: 2026-06-05

### [How to Use Forward Dynamics for Action-Conditioned Video Prediction in Cosmos 3](/NVIDIA/cosmos/how-to-use-forward-dynamics-for-action-conditioned-video-prediction)

Learn to use forward dynamics for action-conditioned video prediction with NVIDIA Cosmos. Generate future video rollouts from robot commands or camera poses.

- Tags: tutorial
- Published: 2026-06-05

### [How to Fix torch.cuda.is_available() Returning False in NVIDIA Cosmos](/NVIDIA/cosmos/troubleshooting-torch-cuda-is-available-returning-false-with-cosmos)

Fix torch.cuda.is_available() returning False in NVIDIA Cosmos. Resolve CUDA version mismatches by reinstalling PyTorch with the correct --torch-backend flag. Get CUDA working now.

- Tags: troubleshooting
- Published: 2026-06-05

### [How to Configure Tensor Parallelism for Cosmos 3 Super Model](/NVIDIA/cosmos/how-to-configure-tensor-parallelism-for-cosmos-3-super-model)

Learn how to configure tensor parallelism for the Cosmos 3 Super model. Boost performance by sharding the 64B parameter model across multiple GPUs using vLLM.

- Tags: how-to-guide
- Published: 2026-06-05

### [How to Use Cosmos 3 for Robot Action Prediction: Complete Policy Server Setup Guide](/NVIDIA/cosmos/how-to-use-cosmos-3-for-robot-action-prediction-policy)

Learn to set up the Cosmos 3 policy server for robot action prediction. Stream predicted actions to clients for simulation or real world execution.

- Tags: how-to-guide
- Published: 2026-06-05

### [How to Set Up vLLM-Omni for Video Generation with Cosmos 3](/NVIDIA/cosmos/how-to-set-up-vllm-omni-for-video-generation-with-cosmos-3)

Learn to set up vLLM-Omni for video generation with Cosmos 3. Utilize the official Docker container and its OpenAI-compatible API for seamless multimedia creation.

- Tags: how-to-guide
- Published: 2026-06-05

### [Text-to-Video Generation with Cosmos 3 Using Diffusers: Setup and Inference Guide](/NVIDIA/cosmos/how-to-run-text-to-video-generation-with-cosmos-3-using-diffusers)

Generate videos from text using Cosmos 3 and Diffusers. This guide covers setup and inference, allowing you to create videos locally on your GPU with simple text prompts.

- Tags: how-to-guide
- Published: 2026-06-05