LTX 2.3 LipDub in ComfyUI: multilingual dubbing tutorial

This tutorial explains how to use LTX 2.3 LipDub IC-LoRA in ComfyUI to dub generated video into another language. The speaker compares English source clips with French, Spanish, Korean and German outputs, describing changes to mouth movements and lower-face expressions while the character and scene remain consistent.

The setup uses updated ComfyUI LTX Video custom nodes and a two-stage distillation workflow. LTX Set Audio Reference Tokens supplies speaker identity context from reference audio encoded through the LTX audio VAE. The walkthrough loads video frames, matches the latent length to the clip and swaps width and height for portrait footage. The demonstrated first sampling stage uses half-size dimensions and eight steps, followed by 2x upscaling.

Prompts identify the character, target language and quoted dialogue. Whisper supplies a transcript of the source speech; the presenter then translates that text before inserting it into the prompt. A French example compares dialogue alone with a prompt that includes laughter. A Spanish example assigns separate lines to two characters across camera changes.

Steady-camera examples make the facial changes easier to compare. The presenter notes shading and color shifts in the Korean clip's final frames and suggests higher-quality sampling could help. That test leaves the high-resolution group disabled, so it does not establish the workflow's best achievable quality.