SEQUEL OUT NOW: Skirtlift
I was having a hard time prompting a woman lifting her shirt, so I made this LORA. It also works for men.
Sample images made with the model Liberty.
EDIT: some people asked how I made it. Here's a rough guide:
Workflow:
1. Scour the interweb for pictures that represent the theme (women pulling their shirts up). Try to get a range of bodytypes/outfits/locations (while keeping in mind that they need to still have enough of a consistent theme for the AI pick up on what's being trained)
2. Crop and tag images. It's better to crop manually, especially with smaller datasets, but I had over 100 images so I automated it and just deleted the bad crops. (Many tools available for this. I think I used the Smart Process extension for Automatic1111).
2a. Cut the tags that represent the theme ('flashing,' 'pulling back shirt,' etc.), replace with a universal tag ('shirtlift')
2b. Fix incorrect tags
3. Run a first LORA training (I used the Kohya LoRA Fine-Tuning Notebook by Linaqruf).
4. Test it with an uncomplicated prompt and see what's not working:
4a. Is it not grasping the theme? Cut images that are too much of a variation on the theme (here it was mostly shots where the shirt was too bunched up to for the AI to 'read' it as a shirt, and shots where other things were going on with the person's hands)
4b. Is it reproducing aspects that aren't essential to the theme? Cut some of the images that over-represent those aspects in the dataset and/or describe those aspects in the caption since the AI might be assuming they're part of the universal tag. (Here it was forcing some of the same facial features or bodytypes)
4c. Is it just being wonky in some indescribable way? Change the training settings. (This is still trial and error for me but there are some guides out there)
4d. Repeat step 3 with those changes
5. Repeat step 4 using more complicated prompts to see if it can carry the theme over to different styles and situations that weren't represented in the dataset.
5a. If it's having trouble reproducing something you really want, try to find examples of that thing to add it to the dataset.
6. Repeat step 5 until satisfied.
Settings
I ended up with 104 images after a couple rounds of trimming out unhelpful ones.
The settings were mostly the defaults in Linaqruf's Kohya Lora Fine-tuner notebook, but with these changes:
Model: SD1.5
mixed_precision: FP16
Network Dim + Network_Alpha: 64
LR_Scheduler: cosine_with_restarts
Dataset_Repeats: 2
Clip_Skip: 1
I don't know how important some of them are; I'm still trial-and-erroring this thing.
For more info, this is a helpful guide I learned from.
Description
Trained on SD1.5, 104 images, Network Dim 64, 10 Epochs, 2 Repeats.
FAQ
Comments (59)
"Sample images made with the model Liberty." So you generated images with shirt lifted with that model and trained lora with those generated images?
No, the samples I posted here were generated using the LORA on the Liberty checkpoint. I trained the LORA with images from the world wide web.
@revandie ooh, right. You used Liberty model + your lora to make posted here images. Sorry for confusion. :-D
just curious, what did you train the LORA against? I've been using Liberty but I find they look pretty good on other checkpoints too
@trelesngoling for the most part, if your inputs are real photos you want to train LoRAs on sd-1.5-pruned-emaonly and then you'll get clean results on whatever checkpoint you eventually use to generate outputs from the LoRA. I've only seen slightly worse results using other checkpoints for the training.
I've noticed that the sample and the downloaded file names are different, so the sample doesn't work as it is.
Someone should try to make a "sharking" clothes grab.
For the artistic value I mean. :)
LORA question -
I see the first one has: shirtlift <lora:shirtliftv1:1>
Are both necessary?
I guess I made it necessary? I could have trained it without needing a trigger word since it really only does one thing. But... I didn't.
@brandnewfakeuser OK, I just had no idea either way. Thanks for the lora, tho!
That's an auto1111 thing. Pretty sure <lora:shirtliftv1:1> means to use the shirtlift lora at an alpha of 1. Then putting shirtlift in the prompt will actually trigger it.
This is one of the best functioning LoRAs I have tried. I would love to know your process. What tool did you use to make it? Steps, settings ect. Thanks
I added a rough guide to the info. I hope it's helpful!
One downside if I may :
you chose it seems only training images with big boobs. In your example images even the guys get boobs.
I tried my usual positive (flat chest, etc) and negative (big breast, medium breast, small breast, etc.) prompts that usually give me 99.99% of flat chested girls. But as soon as I use your Lora : they all get boobs no matter what.
I'm sure it fits a lot of users ;) But given the quality of the Lora it's too bad that it takes over the breast prompt and mostly generates boobs.
Was just for info ;) Except that, it does work damn well.
Agree and would suggest doing "inpainting" for that DFC goodness. :)
Has not been my experience. Do (((big breasts))), (((huge breasts))), (((big tits))), in your neg prompt and don't mention the word breasts in your positive prompt.
Don't mention "small breasts" in your negative prompt. Only mention big and medium breasts. I guess it equates small breasts with flat chests.
Is there any video guide on how to use Kohya LoRA Fine-Tuning Notebook by Linaqruf or just any training?
Not sure if I can post links here. This guy said he was working on exactly this tutorial 20 days ago, and said keep an eye on his channel. YouTube @SECourses he may or may not have posted the video already
I'm trying to get this working on Deliberate Model: https://civitai.com/models/4823/deliberate.
It's based on SD1.5, so I'm assuming it should work but i'm not getting anything happening?
This looks awesome! Well done !bad thing is IM using the excellent mac app "Draw Things" to create, IM not sure how to enable this. Anybody else know if this is currently possible with DRaw Things?? or is there another full model that that this lora has been merged into? Thanks
Really isn't working for me neither with (shirtliftv1) nor doing it via the 'show extra networks' option in A1111, using liberty as the model and dpm+2M karrras
Works great, wonder id you could share a few of the captions you used for training?
@SleepyDaddy They're actually really bare-bones:
a redhead woman with glasses is sitting on a bed, shirtlift
a woman with a plate and a bag on a bus seat smiling, shirtlift
a woman with a long brown hair, shirtlift
a woman with very large breasts with a necklace on her neck, shirtlift
a woman at a music festival with other people, shirtlift
a woman in a blue sweater and a hat in the snow, shirtlift
@brandnewfakeuser Thank you for the info, as it seem like captions are the real art here! Curious if you have tried mixing in tags (like from wd-v1-4) with your captions at all.
How do I use it?
Any idea what you need to do to use "LoRA"? I've downloaded it but doesnt seem to do anything in NMKD?
You need to put it into the models/Lora folder of Automatic1111
Great work ! Would you want to train a model if i found you the images for asscracks ? Perhaps slight nipple slips or areola slips ?
Love this! Would also love to see one trained with tits having been pulled out of a top, from the neckline side. Like from a tanktop. Kept trying to inpaint something like that earlier and it absolutely refuses to deliver the kind of "top is squeezing the pulled out tits together"-result.
Or a "one boob out" version...
But this is a great alternative to kinda achieve the same.. thank you
To get girls with smaller breasts, you need to set the weight to 0.4
Hey a few days ago I saw this Lora and saw that there was an image of an older woman here and the prompt said something like american senator or senator, I need the all details for that one please.
The prompt, generation details, the models etc.
Thanks!
Will it work in inpainting using realistic vision model?
I cant seem to make it to work? Would love some tips..
When using this LORA, on img2img and on Inpaint both result in nothing related to flashing tits :(
What am I doing wrong?
Me too, this happens to me too” is the English translation of :(
Love this model, been trying to train my own using your loose guide but never getting consistant results, would you be able to share a snippit your training images/data?
I've tried to crop my data in so many differnt ways at this point but it still dosent seem to grasp the concept
cant get a good result.. trying with inpainting. tried a few different models.. could someone help me ?
this is most incrediblke thing i have ever witnessed
any person i want to flashj i just can
push of button
incfrebile
is it possible that instead of a shirt can i change with the other clothes? for example like sweater or hoodie maybe?
This is one of my favorite behavior Loras for the simple fact that it leaves the aesthetics of the image intact. Fantastic work.
How can I use this with img to img while keeping the initial image intact except for, the shirtlift?
Inpainting?
Doesn't works for me. Doesn't makes them pose, like ever.
what prompts were you using? what weights did you have for the lora and prompt keyword? was the lora in the right folder?
add the prompt "tits"
VERY impressive reliabilty. I test lots of LORAs and this is one of the absolute best.
Works in these models at least:
Unreal Realism, UniverseStable, PopOffMix V9 Ultra Realism, CyberRealistic Classic. Copax Realistic. BismuthMix, AbsoluteReality
Is shirtliftv3 and shirtliftv1 the same file?
Yes, some photo gen with shirtliftv3 has the same lora hashes as the shirtliftv1 photo. Both hashes are 5ce36857badc
dupeguru
The best Lora ever made with skirtlift. Hands down.
This lora is really good and versatile and doesn't ruin the image. Thanks!
now make a reverse, so I can make dress up a woman thats too naked. lol
love this
Request for the future: have an option for no hands lifting it, like say on all fours but it's already been lifted. :)
This is the absolute best. It does exactly it says it's gonna do, no more or less.
Msybe I am doint something wrong, because I am a beginner, but whenever I try to use it on a image of a girl it replaces her chest with male model.
This thing works perfectly, thanks.
I not tryied it yet, but already want to thank you for great guid, helped me a lot ! Grtz from Poland :)
Roughly 100 pictures later, I came to the conclusion that you get the best "shirtlift" photos by not using this Lora. 90% of the pictures end up glitched and some Checkpoints will by now generate the desired output better by default, in comparison with this fairly old Lora.
Details
Files
shirtliftv1.safetensors
Mirrors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
6693_shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv3.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirt-lift.safetensors
shirtliftv1.safetensors
C_shirtlift.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtlift_v1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtlift.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtlift.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtlift.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
ShirtliftALORA.safetensors
54.safetensors
shirtliftv1_brandnewfakeuser.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shrtlftlora.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1_brandnewfakeuser.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftv1.safetensors
shirtliftALORAFor_shirtliftv1.safetensors
shirtliftv1.safetensors






