LTX-2.5 brings open video generation to local NVIDIA hardware
LTX has released LTX-2.5, an open-weights world model designed for video generation, real-time applications and physical AI. It is optimized for local inference on NVIDIA RTX GPUs and NVIDIA DGX Spark, with reduced VRAM requirements. The model is available on Hugging Face, in ComfyUI and through the LTX API, while its code and documentation are published on GitHub.
The system supports native multishot generation, producing sequences with greater consistency in characters, scenes and voices. LTX also introduced a new diffusion video decoder for reducing artifacts in high-motion footage, a Gemma 4 language backbone, a prompt enhancer and a physical-AI checkpoint intended for robotics fine-tuning. A distilled version targets faster, lower-cost deployment.
In LTX’s image-to-video benchmark, a 10-second clip took 6.8 seconds on-premises using 2x NVIDIA GB200, compared with 23.7 seconds through the LTX API. Listed closed alternatives required 52 to 398 seconds. LTX-2.5 is free for organizations with less than $10M in annual recurring revenue and carries no per-generation fees for local use, according to the article.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.