I've seen a lot of people building with Stable Diffusion over the past few weeks - it's clear that application of this technology is still in the nascent phase, as people try to determine what is both viable and useful. Here are a few interesting examples that I've seen so far:
Building on his previous experiment This House Does Not Exist, Pieter Levels has built InteriorAI. You can take a photo of any room, upload it and remodel it in a variety of styles to get interior design inspiration. He's in the process of adding new features to it, and tweeting about the updates he's making in real-time which is very interesting to learn from.
Describing themselves as 'AI-powered stock photos', I think the product has the potential to be much more than that. It allows you to type in any prompt, and generate images based on it. Over time this builds up a very large database of images, which you can then license to users in the same way that existing stock image platforms do (but with significantly lower costs). Something like this could be a better alternative to Google Images in many cases in a few years time, when you're not looking for photographs.
I had considered building this same idea a few weeks ago as a side project, but ultimately decided not to. I struggled to see where the defensibility was besides the underlying technology.