LTX-2.5 (video model)
Related tutorialLTX-2.5 is an AI video generation model that produces short video clips from a text prompt or a starting image. It generates motion, camera movement and visual detail, and produces synchronized sound to match the action on screen rather than leaving the clip silent. Lightricks released it with open weights on 11 August 2026.
Video models like this one are more demanding than image models, since they generate many consecutive frames that need to stay consistent with each other.
What version 2.5 added
- Native multi-shot scenes. The model can render connected shots, wide then medium then close-up, in a single generation, keeping the character, the setting, the lighting and the voice consistent across the cuts.
- A new diffusion video decoder, which cuts the artifacts that earlier versions produced in high-motion scenes.
- A fine-tuned Gemma 4 12B text encoder with a dedicated prompt enhancer, so complex prompts with several subjects are understood from shorter instructions.
- Automatic duration, the clip length being inferred from the action described instead of set by hand.
- Native 4K HDR, high dynamic range being part of the pipeline rather than a post-process.
In tendre.AI, LTX powers all video generation, in two versions: Turbo for fast, lighter drafts and Full for the finished, higher quality clip. Either one renders locally on your own NVIDIA GPU for free, or in the cloud on credits when you need more power or have no compatible card. Sound is generated natively alongside the picture, in sync with the action.