Sub Light Aesthetic -- (yeah, it makes no sense)
This is my attempt to replicate a certain style, with lighting and depth of field and all sorts of camera tricks in the dataset. Whether or not the model actually picks up on those traits is yet to be seen.
Version details:
Light: your standard Pony-based model, trained on 179 images
F: Didn't realize that Flux LoRA models were so small (I initially had it on a separate page after training it; was expecting something in the 12GB range for some reason lol).
H: Heavy model (Pony-based), trained on 3212 images. Lots of images that are similar with subtle changes, such as facial expressions and poses. This was a test to see how many images I could squeeze into a LoRA's dataset. There are plans to see if I can't make a Flux LoRA/Model with this very dataset, in order to combat the skittles effect that vanilla Flux produces and get some good anatomy.
Description
Heavy version. Trained on 3212 images. Due to the sheer size of the dataset and total steps involved, this model might be considered overbaked in a few areas, thus recommend a weight of 0.8--at least with the on-site generator on Civitai; you might get away with a weight of 1 on LimestoneYen.