A²BFR: Attribute-Aware Blind Face Restoration

A²BFR is an attribute-aware blind face restoration model built on FLUX.1-dev. It restores low-quality face images while allowing text prompts to guide facial attributes such as smiling, eyeglasses, hairstyle, and other semantic changes.

This repository hosts the released A²BFR LoRA checkpoint for inference.

• Code: MediaX-SJTU/A2BFR
• Base model: black-forest-labs/FLUX.1-dev
• Checkpoint: default.safetensors