I’ve been working on my syntax, my commands, my interaction with my new toy, stable-diffusion. I’m like a few hundred thousand others tonight, all working away at how best to communicate with this software to get the images we think we might want, but we won’t know until the machine comes back with them, completed. The one included below took about 20 seconds to generate and another 30 to resize and smooth out. The initial image isn’t big, 512×512 and very grainy so some post processing is needed to bump it to 2024x.
I’m jealous of my input phrase and will keep it guarded for now as this is terrifyingly awesome. It is unlikely that this combination of phrase and software would generate this image again randomly and as only I have the key and it’s my secret, for now.
These systems have their limits and the user community is finding them. Trying to make realistic looking hands or bodies in general is difficult if not impossible with the state of the craft today. The upswing in new models being generated daily by thousands of users may change all that, but for now, it isn’t easy to get a good grip with a set of hands without 8 fingers.