Stable Diffusion 19 Apr 23

A sort-of-sketchy, sort-of-painterly update today, with pictures generated by AnythingV4 in combination with a pair of Loras to handle the processing for the lineart and watercolour styles. Only the most egregious anatomical errors were sent to inpainting for correction, in keeping with the sketchy nature of the images.
It’s interesting how- when the bot chose to include colour- that Yellow seemed to be the favoured one. I’m not sure if AnythingV4 was trained on more yellow content, or that’s it’s just the most likely colour to slip past the lineart Lora’s (I’m assuming) inclination for grey scale output. Maybe the algorithm just wants all outwear to be yellow? I don’t know, and I doubt I ever will! 😉