10 Feb, 16:00, online
Nikita Detkov, PhD student
Fine-tuning of large text-to-image models: where all these AI-generated profile pictures came from
Last year, it was a boom in text-to-image models. People have seen beautifully generated images of an astronaut riding a horse, argued on the art ethics, and tried to understand how it can be used in the real world. But even if those models do understand the correspondence between words and visuals and produce high-quality results from most of the text prompts, they lack the knowledge of how... you look like, or your dog, or the style of your favorite underground artist.
In this seminar, we will touch on the text-to-image model basics and explore the ways of fine-tuning those models so they can understand new concepts and generate new concepts with high fidelity in new contexts.
