ConsistentID is a portrait generation system designed to maintain fine-grained identity preservation using a diffusion model architecture. It integrates FaceParsing and FaceID information into the generation pipeline to achieve high identity fidelity without sacrificing diversity or text controllability. The system allows rapid customization within seconds without requiring additional LoRA training, and has been published in TPAMI 2026. Models are available for both SDv1.5 and SDXL versions on HuggingFace.