BERRYBRITE IL

I had to remake this because the original got messed up somehow. (I assume Civit decided to do something to it without any warning or reason.)

Anyway, BERRYBRITE is back on Illustrious, and I took this opportunity to fix some of the issues from the first attempt. Thus this new version is a completely different model from the last attempt, so feel free to delete the previous two versions I made for Illustrious if you happened to download them.

I actually like how this new v1 turned out, so I might not even update it unless some error I missed pops up.

As with the XL and Pony versions, this model isn't meant to capture specific characters, but the general styles, mainly the original Rainbow Brite and Strawberry Shortcake character styles, with a few extra styles thrown in for variety.

Like in the first attempt I made at this model the styles and tags have changed a little.

First we have the Rainbow Brite style, which now uses the 'rbrt' tag.

Next is the Strawberry Shortcake style, which uses the 'stsc' tag.

Third, the pixar-esque/toon style, which I just call ToonMix, which uses the 'tmix' tag.

Then we have a style which was hidden in the XL and Pony models, SuperHero, which uses the 'sprh' tag. (The Pony/XL loras just use a 'SuperHero' tag.)

And finally a new style that was added to Illustrious, which I call Mini-Cute, using the 'mnct' tag.

As you can see there are two variations for each character here. This is because, without running the training multiple times (which causes a bit of overtraining), the styles all have slightly taller, more natural proportions, than the actual styles. So to get around this, you can simply add 'chibi' to the prompt. I figured this fix works well enough that I don't need to train again, and risk that overtraining. Of course you can play around with the strength/weight of 'chibi' to get what you want, these examples just have it at base (1.0) strength, but I find 'chibi' set to strength 0.1 gives the best results. (I'll post a comparison separately)

This model was trained using the Illustrious-XL v2.0 checkpoint, but all generations are made using WAI-NSFW-illustrious-SDXL v14 and Fooocus.

All of my models may be used for your own training and/or merges.