Cutesy-Co Version 2.0 is here! This is a massive re-vamp to the original LoRA in many ways:<ul><li>The LoRA is now trained on NoobAI-XL EPS 1.1, similar to <a target="_blank" rel="ugc" href="https://civarchive.com/models/1063417/hannah-mahjong-soul-noobaiillustrious">Hannah v3</a>. Read Hannah's description for more details on what this means if you're mostly an Illustrious user, but generally this LoRA should be useful with most of your favorite Illustrious merge checkpoints, although you'll probably need to turn the weight up. You'll also be able to use it on Civit's generator with the "Advanced Mode" setting.</li><li>The dataset was updated with more images from more artists that I felt fit into the the original two buckets of <code>cutesy-one</code> and <code>cutesy-two</code>. This was enough to change the absolute style of the LoRA compared to v1.0. While it still personally meets my intended spirit (and I have some examples in the sample gallery on how to get the v1.0-specific styles if you want them), v1.0 is probably still useful for its more specific styling.</li><li>With the dataset update came more tags to play with. Besides <code>cutesy-co</code>, <code>cutesy-one</code>, and <code>cutesy-two</code>, try some more common aesthetic tags like <code>aliasing</code>, <code>painterly</code>, <code>lanky</code>, <code>thick lines</code> , <code>thick thighs</code>, and <code>3d</code> (maybe soft-weight that one) to get at some of the more specific styles within the LoRA.</li><li>A good bit of the training data features character sheets. If you're making landscape orientation images and notice your character showing up multiple times, try <code>solo</code> in the positive and <code>multiple views</code> in the negative.</li></ul>As before, the dataset has a huge bias towards female anthro furries. You can generate other subjects/objects and get the styles, but the strength of the LoRA will fade the further you get away from the training data. It similarly remains a relatively weak model: if you’re using it with a checkpoint that already has styles built in, it’ll be additive rather than overpowering those styles. I’ve suggested some models whose built-in styles pair well with it below, including some of which are routinely available for on-site generation.Have fun with it! If you make anything with this LoRA, please post it below.<hr />Version 1.0 description:I took a couple of my favorite artists (and some private LoRAs I've had sitting on my computer for a while now, you've definitely seen me generate images with prior versions of this) and made a style LoRA out of em. This is a cutesy, cartoony style that is primarily intended for female anthro furries, but you can use it for other characters with some messing around.The LoRA has three activation tags:<ul><li><code>cutesy-co</code> is the style in its entirety, made up as a combination of the two tags below. <code>cutesy-one</code> is definitely the dominant style, with <code>cutesy-two</code> mostly contributing shading and some 2.5d-ness rather than the drawing style.</li><li><code>cutesy-one</code> is the thick outline simple cartoon style. If you just use this, you lose some shading.</li><li><code>cutesy-two</code> is a more detailed kemono-style that has a bias towards chubbier characters, NSFW, and more flowy-styling. Note that this tag is extremely strong and tends to overfit very quickly, so use it at lower weights.</li></ul>Take a look at the sample images to see the various weights I use to get different stylings. This is definitely a LoRA to not just set and forget at 1.0! If you generate anything with the style, post it below!

Files with SHA256: ac38c1ccdda8b34ec73cd58add038a1664c987ba20fc4b7e7b3ae99a4955037a