






The other day I was trying to work on this tron-inspired scifi theme in stable diffusion and I wanted to see if I could take the theme and create a fake promo/ad out of it. I recently trained a Lora model of a new synthesizer groovebox I picked up, it's not the best AI training job, but I was having pretty good luck turning plain boxes into the synthesizer so.. I figured if I adjust my prompt to include "carrying a medium sized cardboard box" then I could generate the groovebox in place of the box and I'd have a neat little cyberpunk style ad. It didn't work, at all. I rendered a bunch of pics on several models and the size of the box varied WILDLY, like there is no such thing as a "medium cardboard box" in the AI brain. This actually ended up being an important reminder on HOW things are referenced when applying AI to real photos (which I plan on doing soon) - If I want to have the model holding something or, as previously mentioned, have a big head piece, I'm gonna need to use SOMETHING as a physical reference for the AI.. otherwise it just wont know where or how to shape things for different angles. This reminded me of an AI video that popped up on reddit that had a guy wearing a towel hanging loosely on his head while he whipped it like hair, then turning himself into a hot anime babe using AI. The fricking TOWEL was the reference for the hair, and it worked almost flawlessly in VIDEO, surely it'd work great in stills. I need to remember to think like this *prior* to going into a planned AI shoot. With that, enjoy this hilarious collection of sexy cyberpunk postal workers carrying boxes with failed attempts at typescript.
Don't ask how the first model is holding her's up.