About
An attempt ad making Quarter Cup Bra using the Civitai Training service
What is Quarter Cup Bra?
From Wikipedia,
A type of shelf bra, sometimes referred to as a cupless, open-cup, half-, or even quarter-cup bra. An underwire fashion design that offers minimal breast coverage, supporting only a portion of the underside of the breast, pushing the breast upward, and leaving the nipple and areola uncovered. Suitable for erotic purposes or when a woman would otherwise want to go braless.
Usage
Invoke the LoRA in the way you would normally use a LoRA. I've found that it doesnt always play nicely with other LoRA (your mileage will vary), but works fine with Textual Inversion.
Use the Trigger word QuarterCup, works with Bra's, Corset or Lingerie. There was a very limited image dataset used, so you may find that poses and subject breasts start to all look the same.
You may also find that inpainting works well, and may give you more flexibility that txt2img
--Update
Suggestion from user fansay,play around with the modify (nipple) in the positive prompt
Description
FAQ
Comments (12)
How did you successfully train anything with Civitai's training service? I tried with 21 images and 500 buzz and it was awful. This looks really good in comparison.
I used around 50 images, and trained it as a concept - maybe I got lucky
@potbelliedconan Oh so I guess i had too few images... good to know! I'll see if i can improve my success rate later.
@potbelliedconan How does one go about tagging for outfits? do you tag everything except the outfit so that the things you tagged are changeable, but not the clothes?
@xSatouMatsuzakax I used BLIP to do the heavy lifting of tagging. My case is fairly straightforward because stablediffusion knows what a bra is, I'm introducing a concept that is variant of a known concept.
My tags were Activation Word followed by the BLIP output. So my tags looked like "Quarter Cup, A woman wearing a bra"
@potbelliedconan Oh? It's that easy? I was thinking about making a lora for some very specific looking patterned socks. Would it be like:
rainbow socks, a woman wearing rainbow socks
Perhaps?
Are you supposed to do it with no other backgrounds? Just a blank background? How does it work if they are wearing different outfits but the socks always stay the same?
Could a person try just making a lora for inpainting purposes and only zoom in on the legs or something from different angles?
@xSatouMatsuzakax this is an actual caption from one of mine
QuarterCup, a woman in a white bra top is sitting in a chair,
Different outfits with the socks staying I think would actually help. I didn't modify any background - don't understand SD enough to say what that would achieve.
@potbelliedconan Good to know! I'll be sure and try it out and see where this gets me. Thank you! I'll also try out your lora one day when I get around to it.
It worked more reliably for me with the additional tags bra, {nipples}
good point, will test and update the guidance
What are squigglies "{ }" for again? I've only used square brackets [ ] and parenthesis ( ) :o
that's a perfect lora model