Wan2GP tutorial: replace a person with LTX 2.3

This tutorial explains person replacement in Wan2GP using LTX 2.3 and IC-LoRA Control. The presenter edits the source video's first frame to replace himself with Einstein, then uses that image with the original clip as a motion reference. He stresses that the edited image should closely match the first frame to reduce distortion.

The local AI workflow covers updating the web interface, selecting the model and uploading both inputs. The presenter uses version 1.1 of the selected model option and suggests trying Q4 if video memory is insufficient, though he does not confirm its memory savings. He identifies the RF05 control LoRA as a 654 MB download and demonstrates a weight around 0.7, with 0.6 discussed as another setting.

Speech comes from a text prompt that describes the voice and puts the spoken words in quotation marks. The presenter explains that the generated voice differs from his own. For the roughly 17-second example, he selects 25 FPS, enters 440 frames and keeps the resolution setting at 720. He warns that too little time can make the speech rush, and that separate prompt lines trigger separate batch generations.

The presenter reports a generation time of 3 minutes 40 seconds for the 17-second result. His review notes restrained movement and teeth that appear fused together. The example provides a concrete setup and its visible limitations, without establishing performance on other hardware.