The civitai community is amazing and I learned a lot here. This is my small contribution in gratitude. All data set images for this FLUX LoRA were hand made and are unique. This LoRA is for those who likes women dressed:)
This FLUX LoRA was trained for trigger sentence "nipples under clothes women dress style", but with just "nipples under clothes" it works amazingly fine. I do not recommend to change LoRA strength from 1.0 (by default), better play with your prompts instead and you will get all you want, I hope. The only thing this LoRA doesn't able to do is generate one girl with nipples near other girls without. This can do the 2.0 version, I will post it some day if I see this one is well received.
Try to start with "Woman with nipples under clothes denoted..."
Please, like the model and post your works if you get good results, it costs nothing for you, but lets me know if i am on the right way.
If you see the nipples do not appear, check the right writing of the keyword sentence, try to move the keyword sentence closer to the beginning of the prompt or use the full sentence "nipples under clothes women dress style". If you are trying to use this LoRA with other 3D or anime style LoRAs, you may need to increase the strength up to 1.6.
P.S.: my last generation experience shows that this LoRA not only reveals the nipples under clothes but also improves a lot the general appearance of standard FLUX women making them to look more natural, even in the cases when the nipples are not visible.
This LoRA is completely free of charge for non-commercial use, but if you want to support me for any reason and motivate me to continue my work you can buy me a coffee https://ko-fi.com/cocosus
You can find some of my exclusive LoRAs on my Tensor.art account https://tensor.art/u/857428067369388114 , like “Sand Drawing” LoRA, for example: https://tensor.art/models/866823207397679395/Sand-Drawing-V1-BEST
Description
FAQ
Comments (15)
it works good, thanks
Thank you for uploading to the gallery!
works so much better than any of the other nipple or pokies Flux loras!
Thank you for appreciate, i always try to make quality content using handmade datasets and captions. Please, post in the gallery the images you get👍
What is the best VAE / Text Encoder to use with this?
Anyone. For ClipL I personally prefer https://huggingface.co/zer0int/CLIP-GmP-ViT-L-14/blob/main/ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors then T5 GGUF version and VAE standard for FLUX. But it works well with any.
If you don't mind me asking are these the VAE's you use for must of your Lora's
@williamdavies1966764 VAE is always the same, original of Black Forest Studio
I cannot get this to work. What am I doing wrong? Tested Lora at 1.0 or 0.5, with exact trigger "nipples under clothes women dress style" and also with partial trigger "nipples under clothes". It always makes a discolored areola area or nipples peak straight through the clothes.
Further testing shows if I use no other lora's, it works decently. When I start adding another lora in addition (character), it doesn't work at all; they become discolored areolas through clothes instead of nipple bumps under clothes. Bumping strength higher than 1, like 1.6 per your notes, has better chance of working, but it changes the character likeness a lot. Any tips, please?
@Cyph3r Working with other LoRAs will depend a lot of that LoRAs quality. 75% of LoRAs here on CivitAI are low rank LoRAS (specially character LoRAs) of bad quality made very fast with small datasets of small size (512x512) images, they are not flexible at all and are difficult to work with. If you find good quality LoRAs (you can find good ones on CivitAI too) it will be no problem to combine them with my LoRA. As a tip for difficult cases, try to describe with a lot of details the upper body clothes, type of fabric, etc. For example: “Natalie Portman woman with nipples under clothes denoted wearing a soft coral-pink blouse made of thin synthetic fabric that is smooth and slightly shiny, etc.” Other tip, get and try CLIP-L Model “ViT-L-14-TEXT-detail-improved-hiT-GmP-TE-only-HF.safetensors”. But here is no universal solution, just look for good quality LoRAs to combine. And you can check the prompts of examples in the model gallery.
@Cocosus Awesome tips! Thank you so much. May I ask you another question? I'm starting to create my own loras and trying to get top quality. 1024x1024, 1:1, 9:16, 16:9. 50-300 images. Starting to separate face and body pics with additional specific triggers for them. What would you say creates a high quality lora like you suggested; I want to take them to the next level beyond what the 75% of loras on here are like, as you say.
@Cyph3r LoRAs for FLUX are the world apart. Small dataset of 40 – 75 images (1024x1024) LoRA takes me one-two weeks on RTX 4080 Super (16 Gb VRAM) to get decent results on 128 rank. The clue is to simplify a lot; you can’t go training multiple concepts at the same time like on SDXL. You need to prepare a data set as small as it possible but of the most quality consistent images. What most affects the result is the dataset, training parameters are not so relevant. For good flexible LoRA the recipe is: good quality consistent 1024x1024 dataset, 128 rank and a lot of days of training. Good Luck!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.