Fluffy cute kittens that allow for a lot of modification (hopefully).

Version 3-XL:

DPM++ SDE Karras sampler seems to work best, but other ones should do as well. There is often an issue with the kittens not having any pupils, as well as other things, so in general I would include this as the negative prompt:

blur, blurry, bad paws, bad kitten ???????, creepy kitten eyes

This model was trained with captions as well, but they don't always work that great:

mother (kitten with its mom), looking up, jumping, cute paws, group, hands (sitting on human hands), walking, milk (drinking milk), head (sitting on someone's head), nice scenery, swimming, bucket (playing with a bucket), playful, in a boat, wanting a hug (supposed to look like it is reaching out to the viewer as if asking for a hug, does not work well), clothing, in a tub

I recommend this extension for A1111/ComfyUI (both for this LoRa, and in general):

GitHub - ahgsql/StyleSelectorXL: This repository contains a Automatic1111 Extension allows users to select and apply different styles to their inputs using SDXL 1.0.

GitHub - twri/sdxl_prompt_styler: Custom prompt styler node for SDXL in ComfyUI

Trigger tokens (3-XL): r4alc4ts / r4alc4ts kitten

Trigger tokens: fyuyK4 / fyuyK4 kitten

Version 2.1:

This one was trained on normal SD1.5, and finetuned for better quality (v2.1) and features more tags!:

cute paws: kittens more likely to hold their front paws up

group: will generate several kittens

grey fur, brown fur, white fur: will (hopefully) generate kittens with a specified fur color

nice scenery: better backgrounds (hopefully)

in a boat: puts them in a little boat

swimming: kitten/s swimming (sometimes looking like they are drowning)

mother: kitten and her cat mother (using "fyuyK4" instead of "fyuyK4 kitten" tends to give better results)

walking: kitten walking

jumping: mid-air of kitten jumping (can look really goofy/funny)

orange fur: now added to fur color choices

looking up: makes kitten look up (hopefully)

cartoonish: added recently, this is to make this LoRA more compatible with anime/cartoon style gens. But you should set the weight of this tag down to 0.6-0.8.

CLIP SKIP?: Depends, it is generally recommended when you create anime/cartoon style images (use the "cartoonish" tag). You just have to experiment and see what works best, I got good images both with and without Clip Skip 1.

____________________________________________________________________________________________

____________________________________________________________________________________________

Version 1.0: This LoRA was trained on Dreamshaper 8, not the standard SD1.5. However, the quality of images and editability will highly depend on which model you choose. Dreamshaper 8 does not necessarily produce the best images with this LoRA. I noticed that disney style models generally perform well. Looks like Euler a, DPM++ SDE / Karras, DPM++ 2M SDE / Karras, and sometimes DDIM work well.

Trigger token: fyuyK4 kitten

Additional keywords:

cute paws: kittens more likely to hold their front paws up

group: will generate several kittens

grey fur, brown fur, white fur: will (hopefully) generate kittens with a specified fur color

orange fur: not used as a token but orange kittens were part of the training, so this generally works well.

____________________________________________________________________________________________

____________________________________________________________________________________________

By the way, in case you are curious as to how I created the dataset for this LoRA, head over to the Bing image generator and enter this:

"absolutely outstanding image of fluffy cute kitten, sitting, adorable, cute little mouth slightly open, cute paws, nice scenery"