Hentai Manga Style
This is a embedding i trained on just hentai images. Therefore you will get a lot of open mouths, ahegao, sweating, saliva etc.. from just using the embedding.
There is no Trigger word
All training images were nsfw so even if you try to go for something SFW you have to properly build your prompt. The lineart gets weaker the longer your prompt is. In other words, the longer the prompt is, the weaker the 2D style gets. You can reinforce the embedding with prompts like:
monochrome, lineart, comic, cross-section,
If you want to keep the embedding relative simple use the above tags, but if you want to get the manga panels etc.. then use:
monochrome, lineart, comic, cross-section, manga \(object\), left-to-right manga,
A good negative prompt is:
(worst quality, low quality:1.4), zombie, watermark, username, patreon username, patreon logo, (extra fingers, deformed hands, polydactyl:1.5), censored, bar censor, child, loli,
This embedding is a hit or miss. As i said earlier, you have to build your prompt properly if you want to get something specific, else it will throw random NSFW stuff at you.
The example images were not all first generations. Some are inpainted, some just went through hires fix, and some just img2img.
Here is my recommended setup:
First of all USE a VAE!!! I have gotten comments from people telling me that they are getting washed out colors and artifacts when using my models, that is because you are NOT using a VAE.
Two of my favorite VAEs are:
kl-f8-anime: https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt
Download one or both, put the file in your stable diffusion folder -> models -> VAE
Then go to the webui and settings -> stable diffusion -> SD VAE and choose the VAE from the drop down. Some models have integrated VAE, my DarkRevPikas model does not, so you have to use one. The "Automatic" option does not work.
You should also consider using Stable Diffusion Dynamic Thresholding (CFG Scale Fix) extension, if you wanna use higher CFG scale without getting artifacts.
Txt2img:
512x768 or 768x512 or 512x512
steps: 30-60
CFG: 10-12 With CFG scale fix set mimic CFG scale 6
Without the extension use CFG 6-8
Img2img
Basically whatever your pc can generate, i use:
896x1344 or 1344x896 or 1024x1024
steps: 30-60
CFG: 10-12 With CFG scale fix set mimic CFG scale 6
Without the extension use CFG 6-8
Denoise strength: 0.5
Hires fix:
Hires steps: 10-20
Denoising Strength: 0.3-0.45
Upscale by: 1.5-2
Description
First version
Leans towards line art 2D style with NSFW images.
FAQ
Comments (14)
"The file may be malicious, so the program is not going to read it."
Please check your process, I'm getting malicious warnings.
What.. The file has been verified. Maybe it got corrupted when downloading? Could you try again and see if that helps?
Please try downloading it again. Apparently the file structure was changed, instead of "archive/data" it was saved as "filename/data". Also i had to delete a "byteorder" file within the data folder. There should not be any problems now.
Maybe there is some problem with this textual inversion.
After adding in embeddings folder i don't see it in my SD, but another embeddings work well.
Don't understand why.
Please try downloading it again. Apparently the file structure was changed, instead of "archive/data" it was saved as "filename/data". Also i had to delete a "byteorder" file within the data folder. There should not be any problems now.
@BixBit11 I tryed the updated file and it works now, thanks :).
[Edit: Works fine with the updated/re-uploaded file]
Sadly, it refuse to load on Automatic1111:
Error verifying pickled file from /stable-diffusion-webui/embeddings/style/hentai-manga.pt: Traceback (most recent call last): File "/stable-diffusion-webui/modules/safe.py", line 135, in load_with_extra check_pt(filename, extra_handler) File "/stable-diffusion-webui/modules/safe.py", line 82, in check_pt check_zip_filenames(filename, z.namelist()) File "/stable-diffusion-webui/modules/safe.py", line 74, in check_zip_filenames raise Exception(f"bad file inside {filename}: {name}") Exception: bad file inside /stable-diffusion-webui/embeddings/style/hentai-manga.pt: MangaStyle-1000/byteorder The file may be malicious, so the program is not going to read it. You can skip this check with --disable-safe-unpickle commandline argument.
I would not recommand to people to use the "--disable-safe-unpickle" argument (it will put you at risks if you load a malicious ".pt" file).
Don't know if this one is really malicious or not, it's maybe just a training setting issue leading to the creation of an unsafe file. But not taking the risk to load it as it is.
Hoping to see a working and safe one in the future :).
I tryed to download the updated version of the ".pt" file, it works perfectly now.
Thanks for the quick fix :).
@Rauks i'm glad you could confirm that it works now. Yeah i just posted a message in discussion regarding the problem. If you are interested, please read it since it might help others with the same problem when training their own embeddings.
!!Important!!
If you downloaded the embedding within the 4 hours it was posted, it will not work for you, and you might be getting a "The file may be malicious" warning.
This is because after updating automatic1111 to the newest version the file structure of the embedding has been changed. Instead of being "archive/data" it changes to "filename/data" and adds another file "byteorder" which corrupts the file.
The problem is now fixed so if you're reading this, downloading the embedding now should not cause any problems.
I'm sorry for the confusion.
And for those that have made their own embeddings that they can't use or train and are getting the same "The file may be malicious", you have to open the file, change the name directory from "filename/data" to "archive/data" and delete the "byteorder" file in the data directory.
Thanks for looking into it.
Note: Compatibility problems occur with the model.
I want to generate an erotic manga, but the AI will mysteriously misinterpret it as "I should just make a fertilized egg!" and it mysteriously misinterprets it.
Where is the demand?
I could not test how compatible this embedding is with Lora's since after updating the SD, Lora's stopped working for me.
Your problem might come from two reasons. One is that the embedding is not compatible with a Lora you might be using. Two, this embedding was trained on images that had alot of intercourse, therefore you get a lot of the spread vagina, puffy vagina etc etc.. Because in the training images their vaginas was spread open and penetrated by a penis. Thats why you might be getting something that looks like a egg since SD thinks this should be something like a penis penetrating the vagina. Hope you understand what I mean.
I am already working on another version, removing what I can so that it won't focus so heavily on nsfw stuff, but more on creating panels and keeping stuff more or less consistent.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.