Since the recent advent of text-to-image diffusion models for high-quality realistic image generation, a plethora of creative applications have suddenly become within reach. I will present my work at Google where I have attempted to unlock magical applications by proposing simple techniques that act on these large text-to-image diffusion models. Particularly, a large class of these applications can be unlocked using personalization by finetuning, starting with our popular work on DreamBooth where we can learn a subject’s appearance and generate that subject in different contexts and with different semantic modifications. My presentation will include a deeper dive into our recent works ZipLoRA, RealFill, RB-Modulation and our latest work Magic Insert.

Nataniel is a Research Scientist at Google and the lead author of DreamBooth, which was selected for a Best Paper Award at CVPR 2023. His main research interests revolve around generative models, and he has authored other works in the areas of controllability and personalization of diffusion models, including StyleDrop, ZipLoRA, and HyperDreamBooth. He obtained his PhD from Boston University, his Master’s from Georgia Tech, and his Bachelor’s from École Polytechnique in Paris. Prior to joining Google, he also interned at Apple, Amazon, and NEC Labs.