LTX-2.5 launches open video model with native ComfyUI support

LTX has released LTX-2.5, an open-weights video and world model designed for faster generation, local deployment and integration into creative workflows. Carl Franzen reports for VentureBeat that the model is available through Hugging Face, the LTX API and ComfyUI, where it launches with native support.

The release adds native multishot generation, allowing users to create several connected shots in one output while aiming to maintain consistent characters, scenes and voices. LTX-2.5 also introduces a diffusion video decoder intended to reduce artifacts in fast-moving scenes and improve details such as faces and text.

According to LTX, the model can generate a 10-second, 720p image-to-video clip in 6.8 seconds on two Nvidia GB200 chips. That result relies on high-end data-center hardware. The company says the same task took 23.7 seconds through its API at 1080p. These and other comparisons with competing models are vendor-reported figures and have not been independently verified.

Built for local and enterprise workflows

LTX says LTX-2.5 can run on GPUs with at least 16GB of VRAM and can be deployed on premises, at the edge or through an API. The company also highlights a distilled version that is designed to lower computing costs and memory requirements.

For content teams, the open-weights approach could be significant. It allows companies to test and adapt the model on their own infrastructure, keeping sensitive footage and proprietary assets inside their systems. Organizations with less than $10 million in annual recurring revenue can use the model free of charge. Larger companies require a commercial licence.

LTX positions the model beyond conventional video production. Potential uses include visual effects, animation between keyframes, computational photography, robotics training and interactive virtual worlds. The ComfyUI integration is central to that strategy, giving users a node-based environment to combine LTX with other generative AI tools and turn experiments into repeatable production workflows.