Descriptions:
Nerdy Rodent covers the v1.1 update to LTX 2.3, an open-source AI video model from LightTricks, and demonstrates a full video extension workflow inside ComfyUI. The update delivers improved audio quality and a different visual aesthetic compared to the prior release, along with a revised Union ControlLoRA that must be downloaded separately to maintain compatibility with the new version.
The tutorial benchmarks four model variants on an RTX 1390 GPU: KJ’s FP8 conversion (the host’s recommended option for balancing performance and quality), the official BF16 checkpoint from LightTricks’ Hugging Face repo, and Unsloth’s GGUF options ranging from an 8GB Q2 to full BF16 size. Documented render times give viewers a practical planning reference: stage one runs in 33 seconds, stage two in 49 seconds, and video extension takes approximately 5 minutes at full resolution on this hardware—with 4090 and 5090 owners expected to see significantly faster results. The FP8 and BF16 variants produce near-identical visual quality, while GGUF files run slower and deliver somewhat lower audio fidelity.
The video extension workflow uses KJ’s LTX Audio Video Mask node, which allows users to define a start and end point for seamless clip continuation. The host demonstrates both generating new footage from scratch and extending pre-existing video, with the v1.1 model showing noticeably cleaner transitions, brighter colors, and better audio coherence than its predecessor across all tested quantization formats.
📺 Source: Nerdy Rodent · Published April 24, 2026
🏷️ Format: Tutorial Demo
