Created at the request of https://civarchive.com/user/badlyfe, who also provided the dataset.
A LoRA designed to create characters lifting or tugging on their bikini bottoms/panties. Works from the front or from behind. As this was trained on real images, adding "realistic" to your negative prompts may improve anime-styled results.
Trained on animefull-final-pruned for 1664 steps.
Description
FAQ
Comments (6)
Just realized these don't have metadata - oops! Here's the prompts:
"masterpiece, best quality, 1girl, clothes pull, wedgie, cameltoe, highleg, lifted by self, blonde hair, navel, print bikini, yellow bikini, leopard print, bikini lift, looking at viewer, braid, large breasts, thick thighs, gold earrings, blonde hair, looking at viewer, smile, grin, outdoors, beach, twin braids <lora:thongwedgie_128dim_nai:1>",3030570218,512,768,DPM++ 2M Karras,7,20,"00002-3030570218-masterpiece, best quality, 1girl, clothes pull, wedgie, cameltoe, highleg, lifted by self, blonde hair, navel, print bikini, yel.png","animal ears, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
"masterpiece, best quality, 1girl, clothes pull, wedgie, from behind, bedroom, adjusting panties, panty lift, pink panties, brown hair, short hair, brown eyes, topless, glasses, freckles, huge ass, thick thighs, cowboy shot, looking back, looking at viewer, open mouth, :o, (spoken question mark:1.2), blush <lora:thongwedgie_128dim_nai:1>",3606502909,512,768,DPM++ 2M Karras,7,20,"00003-3606502912-masterpiece, best quality, 1girl, clothes pull, wedgie, from behind, bedroom, adjusting panties, panty lift, pink panties, brown.png","animal ears, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
"masterpiece, best quality, 1girl, clothes pull, wedgie, cameltoe, highleg, lifted by self, blonde hair, navel, print bikini, yellow bikini, leopard print, bikini lift, looking at viewer, braid, large breasts, thick thighs, gold earrings, blonde hair, looking at viewer, smile, grin, outdoors, beach, twin braids <lora:thongwedgie_128dim_nai:1>",3030570218,512,768,DPM++ 2M Karras,7,20,"00004-3030570218-masterpiece, best quality, 1girl, clothes pull, wedgie, cameltoe, highleg, lifted by self, blonde hair, navel, print bikini, yel.png","animal ears, realistic, shaded face, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
"masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, from behind, bedroom, highleg panties, adjusting panties, panty lift, pink panties, brown hair, short hair, brown eyes, topless, glasses, freckles, huge ass, thick thighs, cowboy shot, looking back, open mouth, :o, (spoken question mark:1.2), blush <lora:thongwedgie_128dim_nai:1>",3030570218,512,768,DPM++ 2M Karras,7,20,"00003-3030570219-masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, from behind, bedroom, highleg panties, adjusting panties.png","animal ears, realistic, shaded face, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
"masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, 1girl, solo, highleg, swimsuit, bikini, bikini lift, cameltoe, hip focus, thick thighs, side-tie bikini bottom, orange bikini, large breasts, outdoors, beach, (head out of frame:1.2) <lora:thongwedgie_128dim_nai:1>",3030570218,512,768,DPM++ 2M Karras,7,20,"00004-3030570219-masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, 1girl, solo, highleg, swimsuit, bikini, bikini lift, cam.png","animal ears, realistic, shaded face, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
"masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, chun-li, solo, large breasts, simple background, brown hair, underwear, panty lift, huge ass, sweat, from behind, white thighhighs, double bun, back, thick thighs, topless, thong, blue panties, bun cover, (facing away:1.2), backboob <lora:thongwedgie_128dim_nai:1>",3775841931,512,768,DPM++ 2M Karras,7,20,"00005-3775841931-masterpiece, best quality, 1girl, clothes pull, wedgie, lifted by self, chun-li, solo, large breasts, simple background, brown h.png","animal ears, realistic, shaded face, (low quality, worst quality:1.3), (loli, petite, aged down, child:1.2), watermark, artist name, censored"
I seen the dataset days ago (is the same of the bounty, right?) and I was shocked by "how bad" are the images: bad quality, stange and extremely confused poses etc...
How can SD learn something usable from this stuff?
Now I'm testing it but... the result is very scarce, with the LORA ignoring the prompt at all or, in rare cases, producing just a girl touching her panties, like in the picure of Arte_I.
Sometimes the girs is pulling the strings, but they are wrong with the rest of the costume...
May be you can use some good result from the LORA itself to improve the dataset of this LORA for a future re-train?
In this case I will help, posting any good result I can get...
Sorry for the bad images, I've never trained a LoRa before, so I have no clue what would make a good data set. :)
@badlyfe Not your fault! You probably collect all the stuff about avaiable around...
piggy backing on this, the prompts don't work at all.
something is wrong with this one, i want it to work so bad
getting this error with photon.
RuntimeError: Error(s) in loading state_dict for CLIPTextModel:
size mismatch for text_model.encoder.layers.0.self_attn.k_proj.lora_A.default_0.weight: copying a param with shape torch.Size([32, 768]) from checkpoint, the shape in current model is torch.Size([2, 768]).
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.