There’s plenty of software with AI and plenty of uses for it. But my interest peaked when the image generators gained popularity. Finally AI could really help me do something I never mastered, visual art!

I’m partially sighted and managed to get to the point where I can do semi-decent photography. That is, within boundaries. I can generally judge mentally what the outcome will be, but through the viewfinder I cannot actually judge it well enough. Drawing, painting … ? Well, many years of classes in school taught me those are skills I do not possess nor will I ever master.

Yet, now AI can do it for me?!

Online engines are cute but limited. I got myself an invite for DALL-E 2 but quickly figured out that I wasn’t interested in paying for the amount of research I’d like to do. Midjourney? Same.

So I landed on Stable Diffusion. Unlike the others, it is Open Source. Slight bump in the road as I realised that my Radeon wouldn’t do CUDA. But as it’s Open Source, it didn’t take long at all for an alternative to pop up. You can now run on Radeon just fine using ONNX. Speed isn’t really at the level it could theoretically be, but beggars can’t be choosers. There’s people out there running it via ONNX on CPUs which is even slower.

Why is running locally so important? Mastering prompt writing and tuning parameters takes hundreds of images. And when provided with the opportunity, expect to make thousands.

For this blog I wanted something with an eye so I had Stable Diffusion spit out a few dozen images across a few prompts. Within minutes I landed on something useful. Is it art? I guess that’s in the eye of the beholder. I’d stop in a gallery and give it some time, so it is for me. It’s unique and won’t be exactly repeated unless someone uses the exact same parameters, prompt …