Nota : Pony being based on SDXL, my pony / SDXL models can work on SDXL / Pony checkpoints... or not, and in sometimes surprising ways. You have to try.
Sometimes, a Pony/SDXL lora can even work on SD 1.5, but of course isn't made for good results like that.

This Lora is an envy of mine to make a Coco. Even if there is very good loras about her (and I use them), it's not the same as to make it myself.

Here's my version of Coco Bandicoot. Cleo is a character from a video game called Crash Bandicoot.

Trigger : The trigger is cocobandicoot.

But can works without. You try.

Change the word COLOR for the color you want, you silly goose ! Or just don't give color at all. Looks at my examples !
Honestly, look at the examples. Do you know the time it took me to do them on all those checkpoints ? I don't have a 600$ graphic card !

the word color is meant to be replaced by the color you want. Throw as much paint on the poor girl as you want to !

Prompting :

The points here are captioned (used in the training, meaning have strong effect when you use them in your prompt) :

COLOR hair : The color of the hair. By default, Coco has blonde, but can be purple (because some "bad/evil Coco" in the training images)

COLOR body fur, COLOR belly. Coco is HEAVILY orange with a beige belly that is nearly impossible to change, but I've still captioned it (a few training images with other colors)

COLOR eyes : By default, they are generaly green.

overalls. understand "naked overalls", too. Captioned "COLOR", but almost never works (Coco always wears a blue, after all)

COLOR shirt. this one works pretty well (just put the LORA weight at 0.75. At 1, the one she wears is generaly white).

hair ornament is often her flower. COLOR hair flower can be added (by default, its colors is blue, pink or red, sometimes mixed).

All images have been tagged with anthro and furry. The lora should works well with non-furry anime checkpoints, but you tell me.

goggles on head. Some checkpoints insist to give her those goggles, even when prompted in negative not to do so (despite me seriously captioning this).

Other prompt :

bangs (the hair on the face), sidelocks (the hair on the side), ponytail (the ponytail) are all captioned, as they are not present on all training images. Meaning they would be able to not appear if not prompted, or used in the negative prompt. Hats have a tendancy to mess with the bangs (as real hats do).

bangs, sidelocks and ponytails can also be used with COLOR, but the result is random, for some reason.

All images have also been captioned with anthro and furry. Will help on Anime checkpoints, maybe also on furries, you try. Particularly true on Pony.

Note for Pony :

Remember : With Pony, it's better to start your prompt with :

Positive (normal) prompt:

score_9, score_8_up, score_7_up, score_6_up, source_furry BREAK

The BREAK can be ignored or not. Depend of your needs and your checkpoint, honestly.

Negative prompt :

score_4, score_3, score_2, score_1

That can also help.

Note for SDXL :

anthro, furry in the prompt can be necessary on some checkpoints.

Tips:

Basically, the more detail you add in your prompt, the better Coco will be.

This is because the few number of images I could find for the training dataset.

clip skip can be 1 or 2. You should try.

Lora weight : between 0.7 and 1. depends of the checkpoint.

Trigger : The trigger is cocobandicoot. You can put no weight on it, or give it 1.1 or 1.2, particularly if you put a lower weight to the lora.

Also :

Using a weight of 0.75 is largely enough. Coco is a character without a lot of variations. and using 1 will make even her hair color unchangeable. It's better to weight more the triggers if she doesn't show well (like in my example in the snow.

Generation with DPM++ 2M / SDE and Automatic / Karras works well

20 - 25 sampling steps should be largely enough.

Upscaling upgrade the image a lot. I use R-ESRGAN 4x+ Anime6B, hires steps 0, Denoising strength 0.35, upscale by 2, and no refiner.

Older versions :

SD 1.5 - V1 : My first version.

You can choose it if you prefer, but it has a lots of errors in the captioning, and defects in the training images.