返回工作流

LTX 2.3 - Lipdub LoRA + Voice Clone

Upload a driving video and prompt the desired dialogue to generate a lipdub of your video using LTX 2.3. The workflow then clones your dubbed audio back to the original character's voice using Chatterbox. It is a fully open source workflow for changing the speech of your character from a driving video.

下载工作流在 Cloud 中试用

Input

Describe the person and the scene, and write the new line in quotes after 'saying:'.

A clip of one person talking, with its audio. Their face and voice are reused.

First-pass height; the output is twice this (544 → 1088).

Default: 544

First-pass width; the output is twice this (960 → 1920).

Default: 960

Advanced settings

What to avoid.

Leave empty for a new result each run. Enter a number to repeat one.

Run

Output
00:00
Example

试试一个示例

关于此工作流

This workflow builds a full, open‑source lip‑dub pipeline around LTX 2.3 and Chatterbox voice conversion. You provide a "driving" video of a person speaking, plus the dialogue you want them to say. LoadVideo ingests your clip, and GetVideoComponents extracts frames, FPS, and the original audio. Your typed dialogue flows from a PrimitiveStringMultiline prompt through RegexReplace to clean punctuation and spacing, then into the LTX 2.3 Lipdub node (the custom node with ID 1e1eaad5-a949-4d3e-9a68-694e64a936a0).

That node applies a Lipdub LoRA finetune to LTX 2.3 to retime mouth shapes to your script while preserving the subject’s identity, head motion, and scene context.

For audio, the same dialogue drives a temporary dub track that’s then passed to FL_ChatterboxVC. Using the original video’s audio (from GetVideoComponents) as a voice reference, Chatterbox VC re-voices the dub so it matches the speaker’s timbre and vocal traits. Finally, CreateVideo assembles the edited frames and cloned audio at the source FPS, and SaveVideo writes the final lip‑synced render. Note the dimension rule from the MarkdownNote: set input width × height to half your intended final size (for example, 960×544 in gives ~1920×1088 out).

常见问答