I previously had a bunch of variants, but I archived them. I am reworking and reducing the count. My plan is to consolidate the creations into two posts. This one which is "normal", and the future second which is like.. doujins.

If you like my work and would like to leave a review, I mean.. feel free to. I do this because I'm a yuge ??????' dege- I mean for fun. Yeah.

Lora Information

Dataset is mostly from Inuyasha: The Final Act. Depending on the version, dataset consists of 100 or 120 hand picked and hand edited images. Mix of colored backgrounds and tagged backgrounds. Both generally are equally split between the two outfits.

SangoPonyV1: 1250 steps (100 images @ 5 epochs, 10 repeats, 4 batch size)
SangoPonyV2: 1080 steps (120 images @ 9 epochs, 4 repeats, 4 batch size)

Both use Adafactor. Second uses slightly higher LRs. Both are trained on Pony Diffusion V6 XL. The Trigger Word is Sango.

Locally trained with LoRA Easy Training Scripts after referencing certain things. I messed around with Prodigy and Adafactor. I'm sure there are better LoRAs out there for this, but me, this is as good as I can get it. Now I do have a doujin version coming eventually.

Prompt Usage

Example Prompt 1 (Kimono)

Sango, 1girl, solo, brown eyes, brown hair, black hair, long hair, low-tied long hair, makeup, bangs, kimono, pink kimono, skirt, green skirt, standing, cowboy shot, anime coloring, retro artstyle, outdoors, day

Example Prompt 2 (Bodysuit / Armor)

Sango, 1girl, solo, brown eyes, brown hair, black hair, ponytail, makeup, bangs, bodysuit, black bodysuit, shoulder pads, thighs, standing, cowboy shot, anime coloring, retro artstyle, outdoors, day

Note 1: These are examples. Use body angles, expressions.. use the score tags. I tried to tag the Hiraikotsu via "giant boomerang" but it's inconsistent. It's kinda hard to get a full body, standing image with sandals, in good quality. This is what I want to achieve.

Note 2: My LoRAs are designed to be le horny. You can probably achieve vanilla stuff with the right prompts/checkpoint. Mine are.. not really.. yeah.

Note 3: The second version features an additional 20 images, improvements to the dataset, ~10 images of Miroku, additions of stuff like "bodysuit", etc.

Use a Lora Weight of 1.

Recommended models to use for this LoRA

I ran X/Y/Z plots for a few checkpoints I have and or use. Sometimes. See the gallery. I wish I enabled the sub image option. You look at the image info or look below at resources.

Can you guess which one I like the most and use the most? Anyway, cheers.