LTX-2.3-22b-IC-LoRA-LipDub Magically Redubs Videos With a Text Prompt

    
        By vramkickedin    
     | 
    
            May 15, 2026 at 9:23 pm        
    
     | 
    
        2 min read

Lightricks released LTX-2.3-22b-IC-LoRA-LipDub, a new open-source adapter that replaces speech and lip motion in existing videos. The adapter works with the LTX-2.3-22b video model to generate synchronized audio and lip movements from a text prompt and an input video. It keeps the original speaker’s appearance and vocal tone while changing what they say.

Lightricks developed the LipDub IC-LoRA and published the research paper “JUST-DUB-IT: Video Dubbing via Joint Audio-Visual Diffusion” detailing the joint audio-visual approach. The team released this version as an open-source beta before launching a commercial API, allowing the community to test and build with it. The project aims to let users rephrase dialogue, dub videos into another language, or create talking-head clips without altering the rest of the footage.

Key features for creators

Key Features

Generates full HD 1080p lip-synced clips.
Handles up to 8-second video segments.
Supports single-speaker videos in five languages.
Uses a single IC-LoRA for joint audio-visual generation.
Works with ComfyUI and native Python pipelines.
Available under the LTX-2 community license.

Privacy-minded creators and small agencies can run the model locally on consumer GPUs, keeping sensitive footage off cloud servers. The pipeline integrates into existing ComfyUI workflows or Python scripts, making it accessible to non-developers and power users alike. By regenerating only the lip region, it preserves backgrounds and overall scene quality while updating spoken content.

Developer notes

This is an early beta, released for community testing before the official API, so users may encounter bugs or rough edges. The current version supports only a single speaker and clips of up to eight seconds; longer or multi-speaker videos are not yet handled. Future updates may add multi-speaker support and longer durations based on feedback gathered during this open-source phase.