>>14245680i have a few different AIs i use.... but for this stuff, There are not enough kot edits to train purely on them, so the best results come form my customized fork of VQGAN which is modified to take a text prompt and a starting image example and then use a pretrained model such as Imagenet_16384 and it does it's best to replicate the starting image in the style of the text prompt.
for instance: this one the starting image I gave it was a large template of kot head only, and then I had to dial in the text prompt as to best describe what the image was starting with, and what the output should be.
>prompt: "Feline facial expression which can be interpreted in many ways depending on context in the styles of gothic stone gargoyle"this is a basic way you can prompt the AI, and when I get it dialed in fully the text prompt also includes instructions as to weights and instructions as to when and how much of each part of the prompt the AI should prioritize.
