Cute & Rainbow Mix Pony Redux
These models are intended to add colorful styles to your art so the tag isn't required, but adding the "Cute" tag to your prompt will help create cute chibi-like characters.
(This is just a repost to fix an error, if you've already downloaded one of the models from the last post you are fine to keep using them as the Loras were not changed.)
This is my attempt to remake the 8 Cute and Rainbow style Loras I released for pony before, the originals can be found here https://civarchive.com/models/446108 (I do plan to remove the original models in the future though)
My reason for doing this is simply because I admit the original models were not trained properly and used style merges that were a bit broken. However, these new models had to be made using recreated data sets, so the styles are not exactly the same, though they are a much less broken overall.
I also didn't add either of the BerryBrite or Care Bears Loras in a merge this time, so those can be used with these Loras without issue.
I major note is that the more colorful backgrounds from the previous models aren't as apparent without prompting for things like "rainbow" or "colorful", so the sample images mostly have simple backgrounds to show this change.
That said, even with the previous models there wasn't necessarily anything more colorful between the Cute & Rainbow Loras, beyond the more colorful loras used in the merges. I simply kept those names for comparing these new versions.
Cute Mix Redux v1-14 (CuteMixPRv1-4)
Rainbow Mix Redux v1-4 (CuteMixRRv1-4)
I also remade the Illustrious Cute Mix (https://civarchive.com/models/1521926/cute-mix-il) models for Pony and added them, and I have already remade these Pony loras for Illustrious (to be released later).
ILLUSTRIOUS Pony Remake v1-4 (CuteMixPDIRv1-4)
I hope these models work out better than the originals and people enjoy them as much as those.
All 8 versions were trained using the ponyDiffusionV6XL checkpoint.
All generations are made using ponyDiffusionV6XL and Fooocus, and upscaled to 2048x2048.
All of my models may be used for your own training and/or merges.