New "modern" version
The original version is still good, but it based on the "classic" (think Night of Comet) cheerleader look that hasn't actually been used by cheerleaders for decades.
This new version is trained on the modern look. Choose whichever look you prefer. Strength of 1 gives a pretty standard result. Lower values give more creativity, usually reveal some midriff.
Keywords:
Modern version: moderncheerunifrom
Classic ("v1.0") version: cheeruniform
Info from Original ("1") version (still good!):
Fairly self-explanatory model. It is trained on over 400 adult cheerleader costumes (many of them sort of the "slutty Halloween" type). Using the phrase "cheerleader uniform" (or outfit or whatever) can also cause models to make the person wearing the uniform look like a stereotypical cheerleader, which may not be what you want. These were therefore trained with the keyword "cheeruniform" to focus entirely on the uniform and not the person wearing it. You must include the word cheeruniform (no spaces) for it to work.
The model is very strong. I included two images that show the same seed/prompt with various weightings. You can also use the LoRA to create cheeruniform-like clothing, such as the steampunk outfit I included.
I include things like ((words)) ((text)) ((logo)) in the negative prompt, but the awful lettering still creeps into a lot of the results.
"I feel a little under-dressed, but at least I'm not dressed like a slutty cheerleader, right?"
Description
The "classic" cheerleader look, generally reserved for old movies and Halloween costumes today, but common in the mid to late 20th century.
FAQ
Comments (7)
Doesn't stable diffusion already make cheerleaders pretty well just by prompting it? How does this improve on that?
do you think every single model has a cheerleader concept? if not then this lora helps
Very model dependent. Also, even with models that do cheerleaders well, it can be difficult to prompt for a cheerleader outfit without also having the woman in the uniform trend towards the stereotypical cheerleader look, which may not be what you want.
I'm not an expert but in my experience since October, the less you leave up to chance (RNG, seed) the better.
There are oddities with token length vs CFG scale, for example. Seems like the less tokens the better. So anything that removes uncertainty is like removing variables. Leaves more tokens to focus on other things, such as say the background of where this cheerleader is appearing.
It does, but like anything with vanilla sd it's hard to get a consistent look, or something that doesn't mix or bleeds colors
Everytime I prompt cheerleader (without Lora), the girl would always bring pom poms, even when I put 'pom poms' on negative. Ive tried it on almost 10 models.
most models make them look like "stripper cheerleaders" rather than actual cheerleaders. Even the other LORAs available are more stripper cheerleaders. This one is like a realistic college cheerleader uniform from a real university.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.