It's on our long-term roadmap to offer other services such as scientific computing, monte carlo simulation, AI, etc. One issue that arises from Deep Learning tasks such as stable diffusion is that the various devices in the network need to communicate periodically, especially when training models, which presents problems when the devices are scattered across the globe.
Yeah, I can see that distributed generation of a single image would be challenging, but from what I've seen it's very common to generate dozens or hundreds of independent images, then have a human select the best ones.