First attempt at making a lora, based on Ryu(u) from danmachi.
I tried making a tag for her "shorts", "Ryu_b0tt0ms" but I don't think the dataset had enough images for it to learn what it was / was not on both models.
Hopefully I can add other outfits in future versions.
For Anima:
I genned images locally, face detailed most.
40 steps,
sampler: euler
scheduler: simple
cfg: 4
haven't tested other settings.
my workflow should still be embedded in the images, though they are a mess.
For Kera 2:
I genned images locally, face detailed most.
8 Steps
sampler: euler ancestral,
scheduler: sgm_uniform.
CFG: 1
Trained on kera 2 raw
Gen on Kera2 Cat tower, since I liked the anime outputs.
Trained only on tags, since it was already done for anima and a so far unreleased illus model.
may try alternating captioning for future versions.
Workflow used should be in the images.
similar prompts used for previews to compare outputs of models
