N(Ai)ggy – Stable Diffusion Prototypes: Iteration 4

Setup for N(Ai)ggy Iteration 4:

• Txt2Img Prompt – cute green bird with a blunt curved beak and a strap around its head that covers one eye, wearing a pirate hat, using a compass and a map, pixar, Rembrandt, Richard Kane Ferguson, warhammer 40k, comic book, 2D, standing on the deck of a pirate ship

• Tuning Details – Steps: 30, Sampler: DPM++ 2S a Karras, CFG scale: 14 (previously 5), Seed: 4173043286, Face restoration: CodeFormer, Size: 512×512, Model hash: 6ce0161689, Model: v1-5-pruned-emaonly

Still didn’t get my eyepatch… but I 1/20 of them DID have it’s view obscured! Both eyes… but I’m getting closer. I added 2D, which I believe helps it focus more on the outlines of the forms, rather than trying to work with depth perception. The thing that made the biggest difference in this set is the change of the CFG scale. The CFG scale is a factor that determines how much the generated images will follow the prompt. This I bumped this up significantly, by 9, and it really narrowed in on the language of the prompt, and produced small green birds, most of which, are clearly pirates…

…but still no eyepatches!!! 🦜☠️🏴‍☠️