NVIDIA NVlabs, open-sourced 2026-05 under Apache 2.0. A 2.6B world model: one image plus a camera trajectory yields 60 seconds of 720P. It is the only one of this batch that fits a consumer card — the distilled build with NVFP4 quantisation renders a 60-second clip in about 34 seconds on a single RTX 5090. The trade-off: it is built for camera motion, not complex character performance.
Weights are free, hardware is not: you buy the GPU up front (one-time estimate, may rise) and still pay for power and time. Six of the twelve are Apache 2.0 (HappyHorse-1.0, SANA-WM, Bernini, MAGI-2, LingBot-Video, LongLive 2.0); LTX-2 and LTX-2.5 use community licences and SANA-Video uses the NVIDIA Open Model License. Hardware runs from one RTX 5090 to 8× H100. See all 12 →
The distilled build with NVFP4 renders 60 seconds of 720P in about 34 seconds on a single RTX 5090.NVIDIA NVlabs · github.com/NVlabs/Sana
| # | Platform / model | Layer | VRAM · setup | One-time hardware |
|---|---|---|---|---|
| — | OSSSANA-WM ↗NVIDIA · 2026-05 · Apache 2.0 · 2.6B | Open source | 32 GBDistilled + NVFP4 · one 5090 | ≈ ¥35,000 · one-time1× RTX 5090 32GB · estimate, may rise |