LSDub for ComfyUI
Copyright 2026 Humanjava Enterprises (Vergel Evans)
https://github.com/HumanjavaEnterprises/comfyui-lsdub
https://lsdub.vidbnch.app

This product includes software developed by Humanjava Enterprises.

If you redistribute this code or a work derived from it, please keep this NOTICE file with your distribution
and credit LSDub with a link to the repository above.

Third-party (nothing below is vendored; models are downloaded by the user or on first use):
- Mel-Band RoFormer architecture (Wang et al., 2023), re-implemented here from the paper. The vocal
  checkpoint it loads is MIT-licensed and is not included in this repository.
- MediaPipe Face Landmarker model (Google, Apache-2.0) and YuNet face detector (OpenCV Zoo, MIT): downloaded
  on first use, checksum-verified.
- DINOv2 (Meta, Apache-2.0), Florence-2 (Microsoft, MIT), SAM 2.1 (Meta, Apache-2.0; the sam2 package),
  Whisper weights via faster-whisper (MIT), SoundSwallower + CMUdict (BSD), g2p (MIT): loaded at run time.
- Generator models run through ComfyUI: LTX-2.5 (Lightricks, LTX-2.x Community License Agreement) with its Gemma 4 text encoder (Apache-2.0), Qwen-Image-Edit-2511
  (Alibaba Qwen, Apache-2.0), ACE-Step 1.5 (MIT).
- Optional external tools, not included and not installed by this pack: Rhubarb Lip Sync and the Montreal
  Forced Aligner. LSDub uses them only if you install them yourself and point it at them.
- The demo assets (released separately, Apache-2.0) are original: synthetic characters generated with LTX-2.5
  and Qwen-Image-Edit, a song generated with ACE-Step, and a lip kit generated by LSDub with Qwen-Image-Edit.
