I’ve been making images with Stable Diffusion, and the part worth passing on is the bill.
I put $50 on RunPod, a GPU cloud, and made dozens of images very quickly. Total spent: $0.23. Here’s how it works, in RunPod’s own write-up — you rent the graphics card and run the Stable Diffusion interface on it.
For anyone who hasn’t tried it: Stable Diffusion is a text-to-image model. You type a description, and it paints the picture. The prompt is the craft, and the same few words rearranged can make the difference between a keeper and a dud. Stability AI released it publicly in August under what it calls “a permissive license that allows for commercial and non-commercial usage,” and at launch it wanted about 6.9GB of video memory, with NVIDIA chips recommended. The code and the model weights are public. You can run it at home if you own a decent graphics card — an optimized version, in Wikipedia’s summary, “can run on most consumer hardware equipped with a modest GPU” — and renting one is the shortcut if you don’t.
That openness is the whole reason for the price. Because anyone can run the model, what you’re paying for is silicon by the minute, and a minute of a rented GPU turns out to be worth a lot of pictures. Nobody’s selling you the images. They’re selling you the electricity, more or less.
The model keeps moving, too. Version 2.1 came out on December 7, nine days ago.
Twenty-three cents for dozens of images. Give it a try.
