All LTX2.3 Dynamic GGUFs + workflow out now!

The AI optimization company Unsloth has published a significant release for the open-source community: all Dynamic GGUF variants of the LTX-2.3 model are now available on Hugging Face. This release focuses on the LTX-2.3, a 2.3 billion parameter model, converted into the efficient GGUF format used by popular local inference engines like llama.cpp. The 'Dynamic' designation is crucial—it signifies that the model files have been optimized by selectively upcasting the numerical precision of important neural network layers. This technique can lead to better output quality and stability during inference without requiring the full computational cost of running the entire model at a higher precision, making it a smart trade-off for performance on consumer hardware.

Accompanying the model files is a complete, reproducible workflow for ComfyUI, a popular node-based interface for Stable Diffusion and other generative AI pipelines. Instead of a static diagram or complex configuration file, the workflow is embedded directly within an MP4 video file hosted in the same Hugging Face repository. Users simply download the video and open it with ComfyUI; the software reads the embedded workflow data and reconstructs the entire node-based graph used to create the video's output. This 'workflow-in-a-video' method provides an intuitive and immediate way for users to replicate advanced image generation techniques, experiment with the new LTX-2.3 Dynamic models, and understand how to integrate them into their own projects, significantly lowering the barrier to entry for cutting-edge model experimentation.