I've done some playing around with Stable Diffusion and tried to create some scenes or characters from the stories. Due to the way LLM systems work, if they haven't got the subject in their data set, or it can't be imported as a LORA, then the outcomes can be problematic. So, for images from the imagination with no real grounding in reality, that's an issue. I really had to force the software to get what something close to what I was after. Just atrting to muck around with Control Net and other tools.
Most of these images are later outputs using improved versions of the software and different XL snapshots. No post work. It's amazing how far the software has come in just a few months: earlier versions - just months ago - were quite limited in resolution and output quality. Now, you can get some quite respectable results out of a home PC.
These are some of the images produced. They are close to what I imagine, but it's extremely difficult to stop the software going off and rendering something ridiculous. So, these are close, but by no means exact. There are always problems with extra limbs, objects, and oddly enough, lamps. The system has a built-in adversity to very dark scenes - it will always throw an extra lamp or light in to try and bring it up to an 'average' light level. That's very annoying.
