LTX-2.5: LTX's Latest AI Open-Source Foundation Model | LTX
Lightricks has released LTX-2.5, the latest version of its open-source video foundation model, available with open weights that users can fine-tune and run on their own hardware. The release marks a meaningful step for the LTX model family, which has positioned itself as a production-friendly alternative to closed, API-only video generation services.
The headline capability of LTX-2.5 is multi-shot scene generation in a single inference pass - meaning the model can produce a sequence of distinct shots that flow together as a cohesive scene, rather than requiring separate generations that are later stitched together. This approach can help maintain visual consistency across a scene and reduces the manual effort typically involved in assembling multi-clip sequences.
Beyond generation from scratch, LTX-2.5 supports editing of real, existing footage. This puts it in a category of models that blur the line between pure generation and post-production tools, allowing users to modify captured video rather than replacing it entirely. The model also supports export to EXR, a high dynamic range image format widely used in professional film and visual effects pipelines. EXR support is relatively rare among generative video models and signals an intent to fit into established production workflows rather than exist purely as a creative prototyping tool.
As an open-weight model, LTX-2.5 can be fine-tuned on custom datasets and deployed locally, which matters for studios and developers who have data privacy requirements or who want to specialize the model for a particular visual style or domain. Lightricks has built a commercial product layer on top of the LTX model family, but the open release means the underlying weights are accessible to the broader research and development community. The combination of multi-shot coherence, footage editing, and professional export formats makes LTX-2.5 one of the more feature-complete open video models available at this stage.
