454 images
Set is pretty good I think, the amount of images I'd personally keep is pretty high. Not that what I like is a good metric for what you like.
You'll never guess it, but images sets don't like generating if there are power interruptions. And it's not even just that my internet went out basically every time did actually have power. Thats the biggest part of the reason for the delay, even though previous issues aren't all gone. As such, billing is paused and will continue to be paused till the middle of April.
I tried combining workflows so that in theory images could come out fully finished. Didn't work. 24GB VRAM is not enough apparently to load 3 checkpoints and 1 onnx model 10 times. Like seriously, why does it have to reload every time and it doesn't clear itself till after the workflow is over? Even the nodes that unload models don't seem to know what is going on with that. Anyway. The end result in theory is better images, but about 25% less total images.
I'm currently working on a set up that will actually allow me to inject manually written prompts when they are relevant; rather than just hoping that when it rolls a pose or context LoRA it won't roll any other tag that conflicts and fuck up the image.