Luke Wood is a Machine Learning Specialist and a Software Engineering Generalist. Currently, Luke's focuses are in making KerasCV a powerful and expressive library to solve common Computer Vision tasks and publishing high quality research in top Machine Learning conferences. Luke currently works full time at Google on the Keras team, and is pursuing his Doctorate in Machine Learning at UC San Diego under Peter Gerstoft.
Over the last 6 months text to image models such as DallE-2, ImageGen, and StableDiffusion have taken off. These models can achieve previously unfathomable levels of photorealism, illustrate entirely new images from a single text prompt, and even paint your favorite animal in the art style of Picasso. The potential of these models is nearly limitless.
This talk walks through the architecture and theory enabling these models to generate novel yet coherent images, explores some more advanced uses of text to image models, and lastly shows you how to get started generating images using KerasCV, the most optimized implementation of StableDiffusion available to date.
Thanks Stephan for your flexibility on this! I'm REALLY excited about this topic.
Check out the tutorial I launched today to prepare:
You should be able to generate your own image in <5 minutes on Colab using this guide. Up next, hussling to finish a fine-tuning guide so I can include the theory of that in the talk.