Lightricks, open-sourced 2026-08 under a community licence (not Apache 2.0 — commercial use needs a separate licence). Weights are free; the cost is all hardware: a discrete GPU with 12GB VRAM is the floor, 24GB is comfortable (full FP16 wants 24GB+; on a 12GB card you need INT8 quantisation plus offload and lose more than half the speed). For serious work look at a 4090 / 5090-class card — more VRAM means less quantisation hacking.
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 →
In cases of GPU memory constraints, consider --quantization fp8-cast --offload {cpu, disk}.LTX-2 GitHub README + packages/ltx-pipelines/docs/optimization.md
| # | Platform / model | Layer | VRAM · setup | One-time hardware |
|---|---|---|---|---|
| — | OSSLTX-2.5 ↗Lightricks · 2026-08 · community licence · 22B + 12B encoder | Open source | 24 GB12GB only with FP8 + offload | ≈ ¥24,000 · one-time1× RTX 4090 24GB · estimate, may rise |