Glossary term
Glossary term
Multimodal AI
Fine-tuning technique that teaches a diffusion model to generate a specific subject (person, object, pet) from 3-30 training images.
Google's original DreamBooth paper (2022) showed that 3-5 images of a dog could teach Stable Diffusion to generate that specific dog in any context - spawning commercial applications for personalised portrait generation.
Astria.ai uses DreamBooth to create professional headshots from selfies - a user uploads 15 photos, a custom LoRA is trained in 20 minutes, and 100 professional-style portraits are generated in diverse settings.
MYNUFACE and similar services use DreamBooth fine-tuning to generate consistent product photography - an e-commerce company uploads 10 product images and generates hundreds of lifestyle shots without photo shoots.