Lightricks releases LipDub: open-source IC-LoRA for 1080p lipsync in one pass

Lightricks has launched LipDub (Beta), an open-source IC-LoRA adapter built on their LTX video model. Unlike modular pipelines that process audio and video separately, LipDub performs joint audio-visual generation in a single pass, regenerating both speech and lip motion simultaneously while preserving the speaker's appearance, vocal identity, tone, and delivery. The beta outputs 1080p Full HD clips up to 8 seconds long, supports single-speaker scenarios, and validates five languages: English, French, Spanish, German, and Russian. The adapter is available on HuggingFace, with ready-to-use ComfyUI workflows and Python pipelines for integration into existing projects.

Practical applications include dubbing content into other languages, rephrasing or replacing dialogue in the original language, and streamlining talking-head generation workflows. LipDub is grounded in research from Lightricks and Tel Aviv University, detailed in the paper "Video Dubbing via Joint Audio-Visual Diffusion," which explains why joint generation outperforms traditional modular approaches. This is an early open-source beta—the team encourages the community to explore, break, and build with it before the official API ships. All resources, including links to the model, workflows, pipeline code, and documentation, are provided for developers to start experimenting immediately.