V3

V3 has 15x the database of v2

It took a while for civitai to process and cost a few hundred extra buzz.

I added the images from the first 2 versions and all I've produced over the last few months.

I did some extensive comparisons with all 3 loras side by side with different Pony checkpoints.

In all 3 the skin tones are about the same.

V3 works well at a lower strengths. I've been using around 0.5 for most checkpoints.

V3 is definately the best for NSFW images.

V2 looks close to v1 but faces do come out thinner and more glamour shot looking. Both work well with realistic photo type images, especially portraits.

Most of the images I had saved were the hyperrealistic type like the ones in the examples. Most were NSFW images so in comparison v3 images came out with sexier clothing and poses. It also does full body images very well without much prompting.

I've noticed V3 does hairstyles fairly well, afro puff pony tail and pigtails came out consistantly. Large tight curl afros don't look great in any of the versions so I may make an afro lora in the future.

In comparing the 3 I also noticed that women in V3 come out looking a bit older. Pony tries to make most images 18 year olds so I'm always trying to age up women. It's noticeable if you look at my profile images.

I'm happy with the new version. The example images were easy to produce, probably because the lora knows my preferences.

I was shocked at how quickly the donation goal was met in v2. I appreciate everyone that contributed, It motivated me to produce this new version quickly.

Thank you to everyone that contributes buzz to this version, I'm planning some FLUX loras and I think I'm gonna need it.

V2

This has a much larger dataset.

Works best at about .80 but can be bumped up to around 1.5 if you're still getting lighter skin.

The darkest of the models were tagged with "very dark skin", so give that a try.

Works great paired with my Box Braids Lora just dial them down a bit .35 to .75

I used quite a few studio glamour shots so that will show in comparison to v1.

Mostly portraits so it does those the best.

This was intended for, and works great with, PONY but helps get some amazing images with SDXL.

Trigger words:

black woman

black girl

dark skin

V1

I was having a difficult time getting black women with the Pony Checkpoint so I decided to try and make a Lora. This is my first attempt making a lora and I think it works fairly well.

I tested it out on several pony checkpoints and it does have an effect on all that I tested.

If the Checkpoint model would produce a black woman then they would usually have very light skin. In these cases a model strength between 1 and 2 would produce a darker complexion.

For the Checkpoints that wouldn't produce a black woman without many prompts, a strength between .5 and 1 would work.

This was made with photos so a checkpoint like realPony will work best.