Stable Diffusion Web App

An AI Tool

What if a dog and a frog were friends? What does an alien with a mustache look like? What about a biker gang of grandmothers in the style of Lisa Frank?

Thanks to modern technology, we no longer have to imagine these things. We can have a computer  imagine them for us! How? Well, with machine learning of course.

Stable Diffusion is a text-to-image machine learning model that allows you to generate “realistic” images from only a text prompt. A lot of discussion has been going on lately about “AI art” and several models have been released in the past several months(i.e. DALL-E, Midjourney) that are creating quite a stir, but both are behind a paywall. 

The models and weights were released as an open source version of Stable Diffusion which meant anyone could run this type of model on their own hardware. Our own Ben Benjamin was really interested in learning more about this new open-source approach and began playing around and testing it. He was quickly able to get the model running on a local server which meant that we could use this technology in-house without a big paywall in front of it.

But what should we do? Enter Alex Iten. Together we talked about the various applications of this kind of technology and how we might as a company leverage it for various projects and installations. But there is still a lot to learn. So what we decided to do was to first create an internal tool where all VT employees can experiment and play with this amazing new tech. 

Our Stable Diffusion Toolkit is a cloud based app that communicates with a “beefy” server in our office that allows VT employees to generate weird and strange images to their heart’s content. Not only is this a fun and inspiring activity, it is also a way for us to crowdsource our experimentation to figure out strategies to prompt crafting with this new model. With the help of the incredibly creative minds of our team, we hope to figure out fine-tuned approaches to image generation that we could later leverage as part of our installations and client work.

From generating detailed, imaginative visuals for creative decks to rendering entire environments for installations, the applications are endless and untapped.

But for now, we’ll just ask the important questions, like, what would it look like if Mikhail Gorbachev was vaping in a casket?