Additional Examples from Cosmos Model Repos
This page provides links to inference and post-training examples from the official Cosmos model repositories. These examples complement the end-to-end tutorials in the cookbook with comprehensive guides for model usage and customization.
Cosmos Predict
Cosmos Predict 2.5 (Latest)
For the latest Cosmos Predict 2.5 model documentation, visit the Cosmos Predict 2.5 Repository.
Inference with Pre-Trained Cosmos Predict 2.5 Models
- Inference Guide: Video generation with Text2World, Image2World, and Video2World capabilities
- Auto Multiview Inference Guide: Multi-camera view generation for autonomous vehicle applications
Post-Training with Cosmos Predict 2.5 Models
- Video2World Post-Training for DreamGen Bench: Humanoid robot trajectory generation using the DreamGen benchmark
Cosmos Predict 2
Inference with Pre-Trained Cosmos Predict 2 Models
- Text2Image Inference: Generating high-quality images from text prompts
- Video2World Inference: Generating videos from images/videos with text prompts (single/batch processing, multi-frame conditioning, multi-GPU inference, prompt refiner, rejection sampling)
- Text2World Inference: Generating videos directly from text prompts (single/batch processing, multi-GPU inference)
Post-Training with Cosmos Predict 2 Models
- Video2World Post-Training Guide: General guide to the Video2World training system
- Video2World Post-Training on Cosmos-NeMo-Assets: Post-training on Cosmos-NeMo-Assets data
- Video2World Post-Training on Fisheye-View AgiBotWorld-Alpha Dataset: Post-training on fisheye-view robot videos from the AgiBotWorld-Alpha dataset
- Video2World Post-Training on GR00T Dreams GR1 and DROID Datasets: Post-training on GR00T Dreams GR1 and DROID datasets
- Video2World Action-Conditioned Post-Training on Bridge Dataset: Action-conditioned post-training on Bridge dataset
- Text2Image Post-Training Guide: General guide to the Text2Image training system
- Text2Image Post-Training on Cosmos-NeMo-Assets: Post-training on Cosmos-NeMo-Assets image data
Cosmos Transfer
Cosmos Transfer 2.5 (Latest)
For the latest Cosmos Transfer 2.5 model documentation, visit the Cosmos Transfer 2.5 Repository.
Inference with Pre-Trained Cosmos Transfer 2.5 Models
- Inference Guide: Multi-control video generation with depth, segmentation, LiDAR, and HDMap conditioning
- Auto Multiview Inference Guide: Multi-camera view generation for autonomous vehicle applications
Post-Training with Cosmos Transfer 2.5 Models
- Post-Training Guide: General guide for custom control modalities and domain adaptation
- Auto Multiview Post-Training for HDMap: Multi-view autonomous driving scenarios with HDMap control
Cosmos Transfer 1
Inference with Pre-Trained Cosmos Transfer 1 Models
- Cosmos-Transfer1-7B Inference: Multi-GPU support
- Cosmos-Transfer1-7B-Sample-AV Inference: Multi-GPU support
- Cosmos-Transfer1-7B-4KUpscaler Inference: 4K upscaling with multi-GPU support
- Cosmos-Transfer1-7B Inference (Depth): Depth-based control
- Cosmos-Transfer1-7B Inference (Segmentation): Segmentation-based control
- Cosmos-Transfer1-7B Inference (Edge): Edge-based control
- Cosmos-Transfer1-7B Inference (Vis): Visual-based control
- Cosmos-Transfer1pt1-7B Inference (Keypoint): Keypoint-based control
- Cosmos-Transfer1-7B-Sample-AV-Multiview Inference: Multi-view generation
Post-Training with Cosmos Transfer 1 Models
- Cosmos-Transfer1-7B Post-Training: Depth, Edge, Keypoint, Segmentation, and Vis controls with multi-GPU support
- Cosmos-Transfer1-7B-Sample-AV Post-Training: LiDAR and HDMap controls with multi-GPU support
- Cosmos-Transfer1-7B-Sample-AV-Multiview Post-Training: Multi-view LiDAR and HDMap controls with multi-GPU support
Post-Training Cosmos Transfer 1 Models from Scratch
- Cosmos-Transfer1-7B Post-Training: Depth, Edge, Keypoint, Segmentation, and Vis controls with multi-GPU support
- Cosmos-Transfer1-7B-Sample-AV Post-Training: LiDAR and HDMap controls with multi-GPU support
- Cosmos-Transfer1-7B-Sample-AV-Multiview Post-Training: Multi-view LiDAR and HDMap controls with multi-GPU support
Cosmos Reason 1
For the latest Cosmos Reason 1 model documentation, visit the Cosmos Reason 1 Repository.
Post-Training with Cosmos Reason 1 Models
- Cosmos Reason 1 Post-Training Example: Complete post-training guide for vision-language reasoning tasks