SD seems to love big breasts. But not all people prefer them too. Even most of the NSFW Lora's for SDXL only produce big ones. This is a Lora for really small breasts.
SD 1.5 Version: https://civarchive.com/models/285031?modelVersionId=320676
Read the "About this version" to get the best results.
I have nothing against 5* ratings, but please refrain from doing so if this LoRA has obviously not been used...
I recommend a strength between 08 and 1.25.
I strongly recommend the "flat chested" versions. Other versions are experimental.
Hopefully it is clear that this LoRA may not be used to create underage or otherwise illegal content
Feel free to share the images you created with this LoRA if you like. I'm excited to see what others here have done with it.
There are not many tutorials for training NSFW stuff. So I am absolutely no expert either.
For this LoRA everything in the image was tagged except for the breasts. For the head I made it easy and it was either not on the pictures or I just tagged it with "head". Not much different than for SFW stuff.
Description
trained on 170 images. For this version i recogment a strenght of 1.2
FAQ
Comments (8)
why is this 200MB?
this is a pretty normal size for a loRA
not really. 32 or 64 dim is plenty for 99.9% of concept/characters loras. 128 is only needed for styles.
Try and retrain with 32dim. See if you can tell a difference. If not, congratulations; you just cut the thiccness by 400%.
it's trained with 32. Here are all the parameters:
{
"unetLR": 0.0004,
"clipSkip": 1,
"loraType": "lora",
"keepTokens": 0,
"networkDim": 32,
"numRepeats": 3,
"resolution": 1024,
"lrScheduler": "cosine_with_restarts",
"minSnrGamma": 5,
"targetSteps": 1530,
"enableBucket": false,
"networkAlpha": 16,
"optimizerArgs": "weight_decay=0.1",
"optimizerType": "AdamW8Bit",
"textEncoderLR": 0.00004,
"maxTrainEpochs": 12,
"shuffleCaption": true,
"trainBatchSize": 4,
"flipAugmentation": false,
"lrSchedulerNumCycles": 3
}
Because smoll tits = big mb size.
@petardapocpocc I'm an idiot. I forgot this was XL.
Sorry!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.