Hi there ! (For V2 details, please read below in green.)Here is a LoRA trying to copy pictures of clothed/unclothed women in one single image, separated by a vertical line. It is not able to nudify an existing picture of a person, as it creates both clothed and unclothed parts.It's intended to work with realistic SDXL checkpoints, as all my dataset was made of real photos. But the LoRA seems to work with anime checkpoints (with some unpredictable results), and even with Pony (with even more unpredictable results, as the one in the example images).If you want to use this concept with anime checkpoints, you may use this lora instead (not mine, I've not tried it though) : <a target="_blank" rel="ugc" href="https://civarchive.com/models/522623/before-and-after">https://civarchive.com/models/522623/before-and-after</a> Edit : and I'm stupid, I've not searched for other LoRAs already in CivitAI before making this one... Now you can find other links at the bottom of this description, some of them may work better than mine.My LoRA doesn't work perfectly : the vertical line is sometimes weird, and the number of frames is not always following the prompt. You may need to try several times with several seeds. Also, there are some words you should use in the prompt.All the images above were generated using ADetailer for the faces, and nothing else (the hands are quite often weird). If you use the same prompt with the same seed and parameters, you should get the same images as me.<h3 id="v2-:-chaotic-screencap">V2 : Chaotic screencap</h3>This version seems quite unstable, i'm not really happy with it. But I may not be able to play with it for the next weeks, so here it is anyway.I've added some pictures with the "movie_screencap" tag in the dataset. This tag may be the main addition for this V2. You can use something like : "beforafter, 2frames, movie_screencap, amateur, photo of [...]" For me, this tag seems to produce interesting results. You may want to try using it. I feel that, surprisingly, the "movie_screencap" tag works well with the "amateur" tag.(The pictures in the example post were generated with ADetailer (face_yolov8n) and nothing else.)The new dataset contains slightly more shyness too : if you use the "movie_screencap" tag, the characters may seem a little more embarrassed. You can add something like "she is smiling" or "looking seductively at the camera" to make things look right.In addition, the "embarrassed" tag usually produces unpredictable hands reactions, like covering breasts/vagina, going under the character's back, above her shoulders, or other. I've tried to add the tags "covering vagina", "covering breasts", but the final image rarely works as intended, especially because the hands/fingers look quite bad. The "hands behind her back" and "hands on her hips" (and "hands behind her head") tags should work quite often. You may need to try with several seeds to get good results. Sorry for that !I've added many more "3frames" images in the dataset. With V1, the LoRA tried quite hard to get images with only 2frames, so it was quite hard to get 3frames but you could get 2frames easily. Now with V2, a "2frames" prompt may result in a 3frames images a little more often, but a "3frames" prompt should result in a 3frames image very more often. I guess it's better overall. ("4frames" is still often confused with "3frames".)I've added some new prompt examples at the end of this page too.Have fun !<h3 id="prompting-:">Prompting :</h3>The activation tag is : beforafterYou also should add the [number of frames]. Example : "2frames" is the safest choice. I've tried to make the LoRA work with "3frames", "4frames" and even "5frames", but it rarely works as prompted.I made the dataset as follow (and you may get better results if you follow this structure) :<ul><li>photo (/selfie) of a [age] woman with [length and color] hair</li><li>she is wearing [top clothes] with [bottom clothes]</li><li>(rare/abundant) pubic hair / shaved pubes</li><li>she is [standing / sitting] [description] in front of [background]</li></ul>"Selfie mode" special words to add to your prompt : "she is holding a [color] phone" and (if needed) "she is looking at her phone".In addition, you can use :<ul><li>"[amateur / professional]", which would adapt the image quality</li><li>"photo flash", especially in association with the "amateur" and "selfie" tag</li><li>"same pose before/after", which actually seems to (highly) improve the background consistency if you want it.</li><li>"chest tanlines", "pubic tanlines"</li><li>"bracelet", "navel piercing", "earrings", "ring", "necklace" seem to work but are quite unpredictable as well : with "bracelet", your character now can have cuffs, two watches, other jewelry, ...</li><li>"portrait" or "close-up portrait" instead of "photo". In this case, if you just want a picture of an upper body, you can add something like "pretty face" or eye color, to let the model focus on the face. And don't write anything about pants, legs nor feet.</li></ul>(I've tried to make the LoRA learn some caption like "she is looking seductively at the camera", "she is shyly smiling", "looking cute" or other words. But they don't seem to work very well, so I suppose you can just ignore them.)Well.. Sorry if this seems complicated. Of course, nothing is really necessary after "beforafter" and "number_frames". You can see in the image examples that you can have a good result with just "beforafter, 2frames, a pretty woman is sitting on the train".<h3 id="some-examples-:">Some examples :</h3>"beforafter, 2frames, selfie of a 25 years-old woman with medium black hair with two short braids, she is wearing a short blue dress with white flowers drawn on it, shaved pubes, holding a grey phone, she is looking at her phone, she is smiling shyly, looking cute, she is standing in her bedroom in front of her bed""beforafter, 2frames, amateur, photo of a 25 years-old woman with tied black hair, she is wearing a brown sleeveless top with grey shorts, pale skin, she is standing outside next to black waters in front of a waterfall, pubic hair, earrings, she looks like Lara Croft from Tomb Rider""beforafter, 2frames, amateur, close-up portrait of a 28 years-old woman with long blonde hair, pretty face, photo flash, in her living room, she is wearing a white top with a batman symbol on it""beforafter, 3frames, photo of a 23 years-old woman with very long blonde hair, she is wearing a red t-shirt with short blue hotpants, shaved pubes, same pose before/after, she has big breasts with pale erect nipples, she is standing in front of blue curtains, chest tanlines, necklace, various facial expressions" (which may not get 3 perfect frames)"beforafter, 3frames, photo of a 26 years-old woman with medium black hair, asian woman, she is wearing a white skirt and a white sleveless top, she is in her living room next to a glass table, photo flash, abundant black pubic hair, she has brown nipples" (same)"beforafter, 2frames, photo of a 32 years-old woman with medium black hair, Asian woman, she is wearing a white dress, pink bench, small breasts, pubic hair, she is smiling to the camera"(V2) : "beforafter, 3frames, movie_screencap, amateur, close photo of a 37 years-old woman with long black hair, she is wearing a pink shirt with short sleeves and a beige skirt, pubic hair, she has big breasts with pale nipples, she is standing in her living-room in front of a white wall next to a brown sofa, embarrassed, earrings"(V2) : "beforafter, 2frames, movie_screencap, amateur, close portrait of a 44 years-old woman with long black hair, she is wearing a grey shirt with a white long skirt, covering vagina, she is standing in in front of a white door next to a beige wall, embarrassed"<h3 id="other-similar-loras-you-can-find-on-civitai-:">Other similar LoRAs you can find on CivitAI :</h3><ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/487151/front-unclothed-x-steps">https://civarchive.com/models/487151/front-unclothed-x-steps</a> (for SDXL),</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/568803/front-unclothed-x-steps">https://civarchive.com/models/568803/front-unclothed-x-steps</a> (for Pony),</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/69576/onoff-concept-fantasy-konyconi">https://civarchive.com/models/69576/onoff-concept-fantasy-konyconi</a> , <a target="_blank" rel="ugc" href="https://civarchive.com/models/14502/cloth-uncloth-lora">https://civarchive.com/models/14502/cloth-uncloth-lora</a> , <a target="_blank" rel="ugc" href="https://civarchive.com/models/42214/onoff-onoff-lora">https://civarchive.com/models/42214/onoff-onoff-lora</a> , <a target="_blank" rel="ugc" href="https://civarchive.com/models/42370/concept-onoff">https://civarchive.com/models/42370/concept-onoff</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/10703/clothed-vs-unclothed">https://civarchive.com/models/10703/clothed-vs-unclothed</a> (for SD1.5),</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/776854/on-off-for-flux">https://civarchive.com/models/776854/on-off-for-flux</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/649762/onoff-for-flux-onoff">https://civarchive.com/models/649762/onoff-for-flux-onoff</a> for Flux</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/120105/onoff-lora-for-sdxl-and-pony">https://civarchive.com/models/120105/onoff-lora-for-sdxl-and-pony</a> for several models</li></ul>

Files with SHA256: 483c1c8f8fb595c34c1a3bd7031479f4760d21979dc5f31052843f0bf0b067f7