LTX-2.5 Pro

LTX-2.5 Pro is a high-fidelity video foundation model developed by Lightricks. It is the quality-optimized variant of the LTX-2.5 family, designed to generate synchronized video and audio in a single pass. Unlike its speed-focused counterpart, the Pro variant prioritizes visual fidelity, supporting resolutions up to 1080p and durations between 2 and 10 seconds. It is built on a 22B-parameter Diffusion Transformer (DiT) architecture and is released with open weights for local execution and fine-tuning.

The model introduces several architectural advancements, including Diffusion Fidelity Rendering, which dynamically allocates compute based on scene complexity to preserve detail in high-density areas. It utilizes a custom Gemma 4 12B text encoder to improve prompt adherence and a dedicated Diffusion Video Decoder that replaces traditional VAE stages to deliver sharper textures, legible on-screen text, and more natural motion with reduced artifacts.

A significant capability of LTX-2.5 Pro is native multishot generation. This allows the model to produce multiple connected scenes in a single output, maintaining character identity, lighting, and environmental consistency across cinematic cuts. It also features a Prompt Enhancer that automatically expands short, simple descriptions into detailed cinematic instructions and an optional Duration Predictor that can automatically determine the optimal frame count based on the action described.

For optimal results, users should leverage the model's ability to handle complex camera directions and multi-subject actions within a single prompt. When using the image-to-video mode, providing a high-resolution reference image helps the model establish better initial consistency. The Pro variant is particularly suited for professional finishing workflows, supporting high-dynamic-range (HDR) output and frame rates up to 50 fps.