Spent a little time improving some pie stuff and training a LoRA file. I'll tidy it up and share it on soon as it's created with copyright free generated images.

I'll post two sets here. These first ones are straight up the LoRA file used with SDXL in the normal way.

This next set uses the workflow from my previous post here: https://umd.net/groups/open-gunge-lora-file-and-high-prompt-adh

It definitely introduces some more dynamic qualities to the scene.

In some ways, I like the first batch better. The "camera" is further back and we see more of the girls involved. just my 3 cents

I agree, I prefer the first ones. I learned something with the second lot that the Pixart Sigma model has no real idea what a pie in the face is, which makes it challenging as it relies on this part to create the scenes structure before it's passed off to SDXL and the lora files.