Descriptions:
AI Search delivers an in-depth look at LTX 2.3, the latest version of Lightricks’ open-source video generator, which is notable for having native audio generation built directly into the model. The video combines personal benchmark comparisons against the previous LTX 2 with a complete step-by-step local installation walkthrough on Windows 11 using an RTX 5000 ADA with 16 GB of VRAM.
The comparison tests cover a range of challenging scenarios: high-action fight scenes, text-to-video generation of a samurai ambush, the classic Will Smith spaghetti prompt with spoken dialogue, Japanese anime characters with lip-synced dialogue, and a K-pop group singing and dancing. Across all tests, LTX 2.3 shows measurable improvements in motion coherence, facial consistency, and limb stability compared to version 2, with audio quality also improving — particularly on explosion effects and non-English speech. Some static noise artifacts remain in dramatic audio scenarios.
For installation, the video covers the WGP (WToGP) platform, which supports as little as 6 GB of VRAM through memory offloading, with reports of LTX 2 running on as little as 2 GB VRAM given sufficient system RAM. The walkthrough includes installing Git on Windows, cloning the WGP repository, and navigating through Python 3.11 setup — a practical reference for anyone wanting to run LTX 2.3 locally for free and offline without cloud dependencies.
📺 Source: AI Search · Published March 11, 2026
🏷️ Format: Hands On Build
