LTX releases open-weight LTX-2.5 video world model with multishot generation and 6.8-second 720p clips · cho.sh
LTX releases open-weight LTX-2.5 video world model with multishot generation and 6.8-second 720p clips
AI video model releases
LTX releases open-weight LTX-2.5 video world model with multishot generation and 6.8-second 720p clips
LTX-2.5 is an open-weights video world model released on 11 August 2026; its weights are on Hugging Face and are free for organizations with under $10 million in annual recurring revenue.
LTX says its self-hosted LTX-2.5 Pro generates a 10-second 720p image-to-video clip in 6.8 seconds on two NVIDIA GB200 GPUs, while its API result is 23.7 seconds.
The model generates a full sequence as one native multishot output to retain character appearance, environment, lighting, and voice across cuts, rather than stitching independently generated clips.
LTX-2.5 adds native 4K HDR output, a RAW workflow, automatic duration prediction, and a beta mode for editing existing footage; LTX states a 16GB VRAM minimum and support for any GPU, edge deployment, on-premises use, or API access.
LTX attributes changes to a diffusion video decoder, a custom Gemma 4 12B text encoder, and Diffusion Fidelity Rendering, which uses temporally compressed latents and adaptive high-fidelity keyframes; its quality and speed comparisons are vendor-run preliminary results.
cho.sh