FaceChain, a program that generates “digital twin” lookalikes of a given user, has been introduced with potential use cases in healthcare, retail, and AI research. The software is a deep-learning toolchain that uses face biometrics and generative AI to create personalized portraits while preserving unique identity characteristics of individuals, and is explained by self-described IoT, AI, and ML enthusiast Devendra Bogati in a Medium post.
Developed by a team at Alibaba Group, the toolchain is a personalized portrait generation framework of pluggable components. The toolchain injects face models into the generation process of portraits, which improves label-tagging, data-processing, and post-processing, distinguishing FaceChain from DreamBooth and InstantBooth.
The software is trained by taking user-uploaded images and converts them into high-quality forward-facing training images through a series of face models. Then it uses face attribute and text annotation models in tandem with tag post-processing methods to generate labels for training images. The images and label data then fine tune the Stable Diffusion model to get the face LoRA model.
Source: Biometric Update