LTX-2.5 launches as open-weights video model with native ComfyUI integration and Nvidia GB200 speedups
AI video model releases
LTX-2.5 launches as open-weights video model with native ComfyUI integration and Nvidia GB200 speedups
LTX-2.5, an open-weights video and world model from LTX (spun out of Lightricks), generates a 10-second 720p clip in 6.8 seconds on two Nvidia GB200 superchips, versus 23.7 seconds through LTX's own managed API.
The model is free for organizations under $10M in annual recurring revenue, is available on Hugging Face, inside ComfyUI via a day-one launch partnership, and through the LTX API; LTX claims its models have passed 33 million downloads.
New features include a diffusion video decoder for high-motion detail, native multi-shot generation that keeps characters and voice consistent across cuts, a custom Gemma 4 language backbone with a prompt enhancer, a pretrained checkpoint for physical AI and robotics, and a distilled model that runs locally on Nvidia RTX GPUs.
LTX-2.5 Fast costs $0.09 per second ($0.90 for a 10-second clip), cheaper than Veo 3.1 ($4.00), FLUX 3 Video ($1.70), and HappyHorse 1.0 (~$1.82), but pricier than Google's Veo 3.1 Lite at $0.50 per 10-second clip.
LTX's marketing claim of one-eighth the cost of rivals does not hold up against published per-second rates from Veo 3.1 Fast, Gemini Omni Flash, or Veo 3.1 Lite, though its one-seventh render-time claim is roughly supported when self-hosted on GB200 hardware.
cho.sh