Pending benchmarksTracking since Nov 2, 2026
Is LTX-2.5 state of the art?Pending benchmarks.Announced Aug 11, 2026; no independent results yet.
LTX's open-weights video and audio generation model, which it calls an open world model: synchronized video and audio from text, image and video inputs, with native multi-shot scenes, a diffusion video decoder and a Gemma 4 12B text encoder. Weights on Hugging Face under the LTX-2.x Community License, plus the LTX API.
Announced facts
The details so far
- Lab
- Lightricks
- Announced
- Aug 11, 2026
- API
- Live · ltx-2-5-pro · LTX API models ltx-2-5-pro and ltx-2-5-fast, billed per second of video
- Open weights
- Published · Hugging Face Lightricks/LTX-2.5 (gated: accept the license to download), with a Diffusers pack in Lightricks/LTX-2.5-Diffusers; day-one ComfyUI support
- Parameters
- 22B diffusion transformer (plus a separate Gemma 4 12B text encoder)
- Type
- Diffusion transformer for joint video and audio generation (text, image, video and audio inputs)
- Checkpoints
- 22B dev (trainable) and distilled transformers in bf16, Comfy int8 and NVFP4 (distilled), with video and audio VAEs, a duration head and x2 spatial and temporal upscalers
- Text encoder
- Custom Gemma 4 12B with projections
- New in 2.5
- Native multi-shot generation, diffusion fidelity rendering, a diffusion video decoder, a prompt enhancer and an optional duration predictor; most LTX-2.3 LoRAs run unchanged, per LTX
- License
- LTX-2.x Community License Agreement (dated Aug 11, 2026): free for entities under $10M annual revenue; larger entities need a paid commercial license
- Developer
- LTX, the world-model business of Lightricks Ltd. (licensor)
- Earlier models
- Separate from LTX-2 (19B), already in the catalog
Tags marked unsure are not stated on the lab's own pages.
Next steps
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
Self-reported
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: self-hosted on two GB200 GPUs at 720p, it generates a 10-second clip in 6.8 seconds.
Receipts
Sources
LABLTX announcement on X: Introducing LTX-2.5x.com/ltx_io/status/2087255203489755243 · Aug 11, 2026LABLTX-2.5 model page (ltx.io)ltx.io/model/ltx-2-5LABLTX-2.x Community License Agreement (GitHub)github.com/Lightricks/LTX-2/blob/main/LICENSE-2_x · Aug 11, 2026REPOLightricks/LTX-2.5 on Hugging Facehuggingface.co/Lightricks/LTX-2.5REPOLightricks/LTX-2 on GitHubgithub.com/Lightricks/LTX-2PRESSVentureBeat: LTX-2.5 open weightsventurebeat.com/technology/ltx-2-5-can-generate-a-10-second-ai-video-from-an-image-in-just-6-8-seconds-on-nvidia-superchips-and-its-open-weightsLABLTX on X, Nov 2 post about LTX-2.5x.com/ltx_io/status/2117380491422642477 · Nov 2, 2026Tracking continues until an independent board lists it.