ComfyUI_EchoMimic

Why we included this project

This custom node is the practical route if you build talking-head or character animation inside ComfyUI and want the result to lip-sync to an audio track. It wraps the EchoMimic model family, V1 through V3, as native ComfyUI nodes, so you can drive portrait and semi-body animation from audio, pose landmarks, or a reference video without leaving your workflow graph. The maintainer has put real work into making the heavier V3 models fit on consumer GPUs: a flash variant renders 768x768 in about eight steps on 12GB VRAM, and block-offload options trim memory further. LCM and lightX2V LoRAs are supported for faster sampling, which helps if you are building avatar, dubbing, or character-animation pipelines. Setup is the trade-off: you place several model checkpoints in specific folders and install extra Python packages depending on which version you target.