I often run wildcard generations overnight and then train (dreambooth) a random batch of not necessarily related images the next day. Sometimes I'll add cool Midjourney/etc images I've come across to the 20 or so images in the dataset. After these "mismash/garbage" models are trained I run them against folders and folders of old generations from similar "throwaway" models I've trained. Once I like some of the generations I train again and keep repeating the process. The prompts I use for this are just wildcard prompt scenarios created mostly w/ help of older chatgpt-2 style TXT continuation models. After a while I find things concepts I keep gravitating towards. This is one of those things....
I was at loss for what to call it, but since a few of the original images were intricate Midjourney V6 images of "over the top" football players wearing big metal helmets I had started w/ the trigger mjhelmet (aka Midjourney Helmet). The model itself became more of a powersuit generator. It may have some surprises based on the way it was trained over some time w/ other random images in there. If you're bored and this is your type of style maybe see what you can do w/ it running it on different base models.....I haven't posted much lately, so hope some find this one mildly amusing.
Trigger with "mhelmet artstyle" - (it was originally mostly images of metal helmets).
This is just for kicks so if you find it useful please share some images below. Lower the weight of the LoRA if needed to get more variety.
(BTW I write like shit so you know I didn't use AI to write this)