Descriptions:
The Veteran AI channel walks through the newly released official ComfyUI support for Wan 2.1 Infinite Talk, a lip-sync video generation system capable of producing infinitely long talking-head clips from a single reference image and audio input. The video compares the official release against previously popular community-built versions from Kijai, noting improvements in both generation quality and speed.
The tutorial breaks down two complete workflows — one for single-speaker and one for multi-speaker scenarios — hosted on RunningHub. Key technical details covered include the Wan 2.1 Image-to-Video 14B model paired with an acceleration LoRA, the placement of model patch files in the correct directory, audio encoding steps, and the critical “Wan Infinite Talk to Video” node. Generation settings such as 81-frame base video length, 9-frame motion reference, 25fps output, and a 4-step Euler sampler with CFG 1.0 are all explained.
The extension mechanism — which chains multiple 81-frame segments together to match longer audio clips — is covered in detail, including the “Trim Images” parameter used to avoid duplicate frames at stitch points. The video demonstrates outputs in both English and Chinese across dual-speaker and single-speaker scenarios, making it a practical reference for anyone looking to build production-quality lip-sync video pipelines in ComfyUI.
📺 Source: Veteran AI · Published February 02, 2026
🏷️ Format: Tutorial Demo
