NO ACTUAL LEAKED PHOTOS WERE USED FOR TRAINING. I condemn photo leaks and victim-baiting. Every image used in dataset was sourced from public platforms like Reddit, Pinterest, Instagram and RedGIFs, where the girls voluntarily shared their content.
Then why did I call the lora that? Well, if you've seen "the scandalous leak" in 2014, you'll notice that this LoRA captures a similar style - but not the actual content. This, and the controversy xD
Trained on/for LUSTIFY OLT, compatibility with other checkpoints is not guaranteed. FLUX one in plans. 161 pic total, 6 repeats per image, 6 epochs.
The most important trigger word is "amateur photo" - it captures most of the intended style. For an extra touch, try adding "film grain" as well, since many of the grainy images were tagged that way.
Recommend strength: 0.7-1.1 (depends on your prompt and preferences)
Other tags:
1) "taking a mirror selfie" - pretty obvious.
2) "holding her phone in her hand" - present in most mirror selfie pictures, not required for these.
3) "face out of frame" - I've captioned most of the pics where the face wasn't fully visible with this.
4) "focus on the woman's torso and upper thighs" - pretty obvious as well.
5) "light leak" - doesn't work the way it should, probably needs more training, but gives a cool effect nonetheless.
TO DO:
1) Improve composition variety (although it's pretty decent already)
2) Improve face variety (I'm yet to see a serious problem with faces tho)
3) To stay alive
Description
Trained on "LUSTIFY OLT"
FAQ
Comments (80)
hey, lora doesn't seem to activate, doesn't change anything even on a scale of 10, even with key fames :)
key frames? what are you using it for? that's not a video lora
@coyotte I meant the trigger word
@pinek878 I've just downloaded the lora from this page to see if the upload has corrupted, but no, it works fine on my PC. I'm really puzzled right now. Do you have any errors in console?
@coyotte no, it just doesn't change anything, I use A1111
@pinek878 okay, it really doesn't work on a1111. idk why. trying to fix right now
@coyotte
@pinek878 please check now, i've updated the file
@coyotte
it works, thanks :)
@pinek878 no, thank YOU for letting me know about the problem!
THERE WAS A BUG WITH THE FILE.
IT'S FIXED NOW, PLEASE REDOWNLOAD THE FILE IF YOU GOT IT WITHIN THE FIRST 30 MINUTES OF ORIGINAL UPLOAD.
Please make a hair style lora , theres no good loras for that kind of thing
It's a very nice style you've captured!
Thank you!
Awesome style, always wanted something like this for xl, btw , do you think you could do more variations of the same "style"?, something like candid moment of it, or even older resolutions
I can recommend https://civitai.com/models/375657/vhs-xl for VHS style, it's wrongly listed as a sd1.5 lora.
@DonutsDelivery interesting, thanks youu
not really sure what you mean. can you give some example images of what you have in mind?
@coyotte sorry for late response, like some VHS era type leak
@P_Universe sure, I'll add some VHS pics to the dataset
this might also work as a negative xD
Yeah, using it with a negative value might make the image cleaner (just a guess, I haven't tested it), but it'll also censor the hell out of your model 💀 cause there were a lot of nudes in the dataset
@coyotte Right. I tried to target just the most outer part of the model and it actually seem to work. the layers that is just in charge of fine details.
@DonutsDelivery really interesting way to use the lora 🤣 but hey, whatever works
Any chance we get something similar with a more modern (iphone) quality? 🙌
Ditch the "film grain" tag and lower the strength and you'll get at least 80% there. There was a lot of iPhone photos in the dataset, so it knows the look already
Hey, have you tried it? how much strenght would that be? 0.5?
The king is back
Thanks, just what I was looking for, gonna try
Hii, can I use this with Lustify Endgame? I feel like endgame is best model so far.
Can your reupload the older non-fixed?
Nice, i love this style
There is an upscaler, which add grain(a ton), if its your taste
https://openmodeldb.info/models/1x-Filmify4K-v2
very sparly with denoise
My fav is: very good for skin, preserve the realistic look
https://openmodeldb.info/models/1x-ITF-SkinDiffDetail-Lite-v1
Didn't know about the first one, thanks I will take a look
Getting interesting results using Instant ID + RealVis 4.0 lighting : ). I've been using some regular prompts like Super Panvision 70 and certain others that I don't need to anymore after this lora.
Looking forward to your future interations.
Hey, nice LoRA, but saying you 'sourced' images from Reddit, Instagram, Pinterest, and RedGIFs doesn’t exactly scream originality. Anyone can scrape public platforms for a photorealistic dataset and train a model—hardly a flex. Did you verify the rights for those images, or are we just trusting the 'voluntarily shared' vibe? 🤔 Curious to hear how you ensured this is all above board.
shut up blud
@Nemesis18 What a hilariously puerile response which only demonstrates your low intellect lmao
@CabbageBatman shut up monsieur
@nomad42023 Try again.
nigga no one cares
this the type of dude who reminds the teacher about the homework
That's not the point. The point is that it was not trained on revenge porn.
You really out here arguing ethics on a nsfw ai image generator...get a grip or get out
@gomibrev3 exactly
First time in the world of AI?🤣
@CabbageBatman nigga talking like he got 1000IQ
Nice bait, keep trying buddy.
What do U mean by light leak
Photography term. Google it to see the example
More absolutely magnificent work! You, Nutbutter & Subtle are the 3 gods of AI image gen realism. Your time, skill and efforts have taken SDXL to another level to Flux IMO. Appreciate your work so much!
Amazing, groundbreaking, hope it will be used only ffor appropriate private use in the future
i will love you forever if you make a FLUX version of this and lustify :)
I had a trained FLUX version at one point, but it just wasn't good - NSFW focus of this lora doesn't play nicely with flux's nonexistent knowledge of such stuff. Might give it another shot for V2
This is amazing, does anyone know of any other ones like this? Not necessarily for nsfw but for like scenes / normal photos?
Mind sharing the dataset? I would love to bring this to pony & illustrious for my own use. I can train it myself. It'd be a shame to lose the dataset overtime as ai models advance.
I plan to release my training/image parsing/dataset editing and filtering scripts, so that people can collect datasets just like the one I've used for this lora 😌
coyotte Bless you kind sir!!
Why anyone needs a crap shit dataset? Just download some million cheap looking pics from the web.
Like someone said, since this is very unique, would you mind sharing the dataset?, planning to bring this one to chroma unless you thought of that already
Actually I was going to make a chroma version when v50 releases, which wouldn't take long at this point
I plan to release my training/image parsing/dataset editing and filtering scripts tho, so that people can collect datasets just like the one I've used for this lora 😌
@coyotte I was actually lurking your profile wondering if you had released any Chroma tool and saw this comment. I hope its still planned, I like your works a lot, and this lora is amazing, I'm eager to see what can it do on Chroma!
Very good looking model
I have to use put in lora folder? Because it's there and doesn't work
What model are you trying to use it with? It is only for SDXL currently
download the model the open comfyui folder>models folder>lora folder(place it in here) then refresh the comfyui page
Using forge, png samples from this page do not copy data into the UI, and copying by hand yields inferior results. Only happens with this model. Any tips? Thanks
This is my favorite lora to make realistic photos, its very difficult to keep things SFW so interested in the new version!
hope you make a return to train loras/ for QWEN and ZImage with this dataseet.
the dataset from this lora will be baked in into my finetune based on Z-Image Base
@coyotte are you still on it?
This is amazing! Looks so real
What are the best settings for this addon
For example if I use it with endgame
What are the setting I need to use
Any guidance please !!
This is mostly meant for OLT, because Endgame had that kind of training data in it.
For OLT - use triggerwords from description of the lora, 30 steps, dpm 2m sde + karras, cfg 3.5
Its been a year bro
Details
Files
leaked_nudes_style_v1_fixed.safetensors
Mirrors
leakcore_xl.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leak_core.safetensors
leak.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
coretest.safetensors
1439962.safetensors
leak_core.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
amateur_photo,film_grain.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
leaked_nudes_style_v1_fixed.safetensors
SDXL_LEAKED_NUDES.safetensors
leaked_nudes_style_v1_fixed.safetensors
leak_core.safetensors
leaked_nudes_style_v1_fixed.safetensors
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.