This is a LoCon model and may require additional plugins for your software.

A LoCon trained on Pony V6 using 250 images generated by my Girl Next Door SD checkpoint. The training set pulled prompts from Danbooru that contained solo portraits of women. The style should mainly change the look of women's faces while leaving men, backgrounds, bodies, etc. mostly the same. The original Girl Next Door also uses a high contrast art style, which has carried through quite well. Overall, it adds a clean and vibrant "western anime" style on top of Pony's unique character.

While trained against Pony V6, I have had much better results with Autism Mix SDXL Pony. It will work with Pony, but images are very inconsistent. If using with Pony, recommend also mixing in another style such as 2.5d or smooth anime at a lower percent (40-60).

All sample images use Autism Mix. All sample images were generated with the prompt template:

POS
score_7, score_5, kawaii, 
<< description of scene >>,
<< booru tags >>

NEG
loli, 3d, embedding:xl-bad-hands-5-SDXL

All sample images use a Strength of 0.8, Face Detailer, and 1.25 latent upscaling using ComfyUI efficiency node suite. Each sample image has an embedded Comfy workflow. I do not know how these settings translate to A1111 or other tools.

There are no keywords, but the training set used woman rather than 1girl. Recommend prompting with woman when you would normally use 1girl.

General tips:

• Pony tends to like muscular women. If too much, add negative muscular_female.

• The original Girl Next Door sometimes has scary-freakish-gigantic-teeth. Negate with round_teeth.

• Remove 3d from the negative prompt for a realistic anime style.

• As with the original Girl Next Door, the bad-hands-5 embedding has a significant influence on the style. I like the result, so always keep it in the negative. It also appears to have a non-negative to positive impact on hand quality.

• Looks much better at low CFG. Between 3 and 5.

• I use clip weight of 0 for this model.