Generate videos with NVIDIA’s Cosmos-Predict2 2B Video2World, generating physically accurate, high-fidelity, and consistent video simulations. Cosmos-Predict2-2B-Video2World is part of NVIDIA’s powerful diffusion-based world foundation model suite designed for future video frame prediction. It takes a text prompt + an initial image and generates the next few frames—perfect for simulating physical environments or scene continuations with temporal coherence. Released under a commercially-friendly license, it supports modular variants tailored to speed, resolution, and attention needs.
| Variant | Resolution | Frame Rate | Feature |
|---|---|---|---|
| Default | 720p (1280×704) | 16 FPS | Best overall quality and smoothness |
| 720p_10FPS | 720p | 10 FPS | Balanced size minus speed |
| 480p_16FPS | 480p (832×480) | 16 FPS | Faster generation with higher framerate |
| 480p_10FPS | 480p | 10 FPS | Very fast generation, lightweight |
| 720p_16FPS_NATTEN | 720p | 16 FPS | Uses sparse attention (NATTEN) for efficiency |
| 720p_10FPS_NATTEN | 720p | 10 FPS | Efficient and compact, with NATTEN speedup |
NVIDIA benchmarked Cosmos‑Predict2‑2B across various GPUs. For the 720p @ 16FPS default:
Join our distributed GPU compute network. Help us make AI accessible, scalable
and secure for designer, developers and start-ups.