DreamBooth
A fine-tuning technique that teaches diffusion models to generate specific subjects from a few photos.
Overview
DreamBooth, introduced by Google in 2022, is a technique for personalizing text-to-image diffusion models to generate images of a specific subject (person, pet, object) given just 3-5 reference photos. It fine-tunes the entire model using a unique identifier token bound to the subject.
Key Details
A key innovation is the prior preservation loss, which generates class-specific images during training to prevent the model from forgetting how to generate the broader category. DreamBooth enables personalized AI art, virtual try-on, product visualization, and creative applications. Combined with LoRA, it can be trained efficiently with minimal compute.