Technology of Stable Diffusion: Generating Art from Text

During this session, we will discuss the rapidly evolving state of the art in the ML text-to-image field. We'll start with an introduction to what has gone on recently with DALLE2, dalle-mini, and then stable diffusion, all rapidly rising to prominence. Next, how "diffusion" models are trained, and how they work. Then, we'll go over the techniques we can use to exploit these models in all sorts of areas (such as inpainting), as well as future possibilities and ethical dilemmas.

Finally, I will review some brief example code demonstrating how to call the stable-diffusion api programmatically and receive an image from a given prompt.

This is a hybrid event, so if you cannot join us in person at the Microsoft Reactor in Redmond, then you can catch this live or recorded at [https://aka.ms/EastsideAIMLIoT](https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Faka.ms%2FEastsideAIMLIoT&data=05%7C01%7CJim.Bennett%40microsoft.com%7C8b96cd2d7d3041492a6308dab6a85a9c%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C638023131442654058%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=n5SgxjbnnIGe%2BhrY1gLryCazZSvhl8bZIqBmTCpOqcM%3D&reserved=0)

Speaker

Ben is a software engineer at Mosaic ATM, and he has mainly been working on Mosaic's "accelerate" platform for drone footage analysis.
After he graduated from UW in 2018, his work focused on computer vision and ML, but also full-stack coding and implementation of the business logic.

In his spare time, he is the creator of the "sobelstoryboard" web app. He also enjoys writing screenplays and drawing artwork now and then.