Ltx · v2.5 · depuis 11. August 2026 · 4× · vu le 14 août 2026
91
Momentum
LTX-2.5 is an open-weight "world model" from LTX (spun out of Lightricks) for joint video and audio generation from text, image, and video inputs. It is built on a 22-billion-parameter diffusion transformer architecture with native multi-shot generation, automatic clip duration, and native 4K HDR output. The model can be used locally (ComfyUI, Hugging Face) as well as via an API with Fast and Pro variants, and is designed for fine-tuning on custom data.
Historique du momentum
16.05.14.08.
Fonctionnalités
| Fine-Tuning | Yes, pretrained base checkpoint built specifically for fine-tuning and self-hosted deployment |
| Generation Time | 10-second clip from image in 6.8 seconds on 2x GB200 GPUs at 720p |
| License | Open-weights license (LTX-2.x Community License), free for orgs under $10M ARR, larger companies need a commercial license |
| Max Resolution | 4K (Fast variant up to 4K, Pro variant up to 1080p) |
| Max Video Length | Fast variant: 2–20 seconds; Pro variant: 2–10 seconds |
| Platform | Open weights on Hugging Face, native ComfyUI integration, LTX API, LTX Desktop |
| Price | API: $0.09/sec (720p), $0.15/sec (1080p), $0.19/sec (2K), $0.37/sec (4K); free for organizations under $10M ARR |
| Release Date | August 11, 2026 |