C-strings (strapless underwear) for anime models.
Description
Trained on Anything v3, but works on most 1.5-based anime models. Works with men too, albeit somewhat inconsistently. Demo images were created with Counterfit, Cetus, and AbyssOrangeNSFW.
FAQ
Comments (14)
Could you please give us a link to lora:NAI-merunyaa?
It's in here: https://mega.nz/folder/33JBHQQZ#UvpkVuWwT43p8j37OsyWog
Thank you so much for sharing loras. I was wondering if I could try this vivid artistic style as yours example images, but my result is so much worse. Could you tell me where my mistake is?
Model: CetusMix [e911995387]
Lora: <lora:c-string2:1>,<lora:NAI-merunyaa:1>
Prompt and seed pasted from your images.
@lynx If I had to guess, I'd say you probably aren't running Hires Fix, which will give you a gigantic quality boost. I set the size to 896x512, and then scale it 2x with hires fix (25 steps, SwinIR upscaler, .45 denoise strength). You may not be able to replicate these images exactly because I run them in batches of 4, and DPM SDE++ Karras seems to generate different images depending on whether you're batching and what the first image is. However, you should get a similar level of quality (these images were cherry picked, but most of them were pretty good).
If hires fix doesn't fix your problem, post your image somewhere along with a screenshot of your settings screen and I'll see if I can help you with it.
@elbarto526 sure, thanks! it's here https://civitai.com/gallery/130619?reviewId=18365&infinite=false&returnUrl=%2Fmodels%2F9972
The thing is I have pasted all configs/metadata from your example so it is totally identical (resolution, highres fix, denoise, seed, etc,). But it still missing your vivid colors and artistic style somehow. Batch generation does not make any difference.
My automatic1111 is up to date.
Model hash: CetusMix [e911995387]
VAE: vae-ft-mse-840000
Lora: <lora:c-string2:1>,<lora:NAI-merunyaa:1>
@_Envy_ Hello again! I'm not sure if this is the right place to message you, but I have a question, and I don't know anyone better than you when it comes to training lora with focus on specific item. My problem with lora training is that I don't understand how to get it to focus on the part of the suit I want. When I train lora it starts to affect absolutely everything, including the style, the character face and his costume, but not the costume specifically.
hello, very good, your loRA wanted to know how you do it, do you have a link where you learned from, thank you very much
https://www.youtube.com/watch?v=N4_-fB62Hwk&t=0s (also see part 2, linked from the description)
@elbarto526 thanks you are my idol hahaha
how do I replicate the orange lighting you did on the cover illustration, its insane
Hey! Can i ask what model is used for all the images you generated? The lighting and coloring looks unbelievably good and i thought it might have to do with the checkpoint
CetusMix v2
Left a review. I had a lot of fun with this LoRA and will definitely be using it again in the future.
In case you ever decide to improve this with a new version, something I think could be a new feature is other types of bottomwear to go with the c-string. Right now the best options to use this with are wearing almost nothing below the waist or an upskirt (either a wind lift or hands lifting).
I bet this is probably difficult to find training data for, but if you do decide to update this I'd like to see support for crotchless/assless pants, and skirts/dresses that have no front (as opposed to being lifted up). I have examples of the skirt/dress one in the images of my review that I was able to make with a lot of inpainting; I haven't been able to get crotchless pants that look good.
Even if you don't decide to update, this LoRA still works very well as it. Thanks for making it.
Can you do one for XL. Trying to make it work in Starlight, but alas, it will not
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.