Cosmos-Predict2: World Foundation Model for Physical AI by NVIDIA

Cosmos-Predict2 is NVIDIA's world foundation model for physical AI, offering text-to-image and video-to-world generation in 2B and 14B variants.

C

Cosmos-Predict2

VideoWorld Foundation ModelText-to-ImageVideo-to-WorldPhysical AINVIDIA

NVIDIA's world foundation model for Physical AI, specializing in future state prediction. Offers text-to-image generation and video-to-world generation (producing visual simulations from video inputs). Available in 2B and 14B parameter variants with multiple resolution/FPS configurations.

Hugging Face (Comfy-Org)GitHubNVIDIA Docs
DeveloperNVIDIA
Release Date2025-06
ArchitectureDiffusion Transformer
Model Sizes2B, 14B
LicenseNVIDIA-Open-Model-License
CapabilitiesText-to-image, video-to-world simulation at 480p/720p

Guides and workflows related to this model series.

No articles found.