DreamBooth is a new approach for 'personalization' of text-to-image diffusion It fine-tunes a pretrained text-to-image model using only 3-5 user provided images of a subject. This is achieved by binding a unique identifier with the specified subject. Once the subject is embedded in the output domain of the model, the unique identifier can then be used to synthesize fully-novel photorealistic images of the subject contextualized in different scenes.

Key features:

• Able to to fine-tune a text-to-image model using only 3-5 images.
• Able to synthesize fully-novel photorealistic images of the subject.
• Can also synthesize images with specified viewpoints for a subject.

Developed by Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein and Kfir Aberman - Google AI Research.

Image source