169 image
Mid set. Had to shave off a bunch of images thanks to my own folly.
So, I built this thing that adds texture to blank swathes of skin. Turns out I wasn't accounting for edge cases, like when generation shifts the image resolution, or when it decides the middle of a thigh is a face and excludes it from the mask. And let's not forget skipping color correction before merging pieces together. Or leaving the mask with no padding, so it generates at a ridiculous tiny size with zero surrounding context and fucks up the whole section. Good times. That's all fixed now, but the upcoming sets are going to need a lot of extra manual cleanup.
I also reorganized the workflow so rerouting around a node actually skips it instead of running it anyway and wasting my time. That was really just a prerequisite for being able to selectively disable steps based on what the prompt randomly rolls. Which I need because I'm still writing a massive catalogue of curated prompts it can cycle through and pull random elements from: weather, time of day, location, etc. Which I need because contextual and pose LoRAs don't work well when I just scrape their tags off the metadata and dump them in the prompt.
Remember when this was "throw words in a text box and re-roll till it's good"? Man, what am I doing with my life. I'm a walking cautionary tale. A year ago I didn't even know to start with hello world in Python, and now I'm using it for way more stuff than any sane person should. No idea how any of this relates to the set, I just started vibing. Anyway, the images are loosely grouped based on style rather than any notion of what I like or dislike. Seems for the best to stay that way.