Akatsuki Souken Lora (hentai)

It's a Lora trained on Akatsuki Souken style (using this guide). The safetensors(checkpoint) version is here. I did this Lora because many people asked me, upon using it, I still prefer my original checkpoint because I have higher control, but I won't deny that a Lora is more fun and being able to have more diversity.

Read this first before using it: the booru tags I used for training which you might want to use in prompts are here.

it's a highly customizable Lora, which means that, if you use a small prompt you might not get what you want, usually if you add something it should appear in the image, so if you don't add it it probably won't appear. It might be frustrating for some people, or cool for others, it is what it is.

Prompt: 1girl,solo,huge breasts,etc....,lora:AkatsukiSouken:1
Weight: between 0.7 - 1. (I prefer 1, you might change based on what model you use)
All booru tags that you can use for prompts for this Lora are here.

Prompt examples

Negative prompt: easynegative, badhandv4, lr, beard, curly, veiny, muscular, ((bodybuilder)), facial hair, mustache, lowres, fewer fingers, out of frame, jpeg artifacts, ugly, duplicate, morbid,mutilated, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers,

(I added young or speech bubble in my examples because certain checkpoints would generate unwanted material, depending on what checkpoint you will use, the negative prompt will probably be different)

Settings:

• Sampler: Euler a / DMP++ SDE Karras it's what I use

• Size: 512-512 (didn't test for other resolutions)

• Clip skip**:** 2

• Sampling Steps: 30

• Hiresfix upscaler: R-ESRGAN 4x Anime6B on 2x, steps: 20, denoise 0.5 (I didn't test higher or lower than 0.5).

• Cfg scale: 7 (6-11 I use), I tested even higher than 11 and in certain seeds it actually gives better results. It's good to go higher cfg scale if you want to respect the prompt more.

ChatGPT prompts, that's how I do my prompts.

Q/A
Q: What are the checkpoints(models) you're using in the examples?.
A: Beside anyhentai and cardology rest are trained by me and I didn't upload them on civitai.
Q: I used Lora but I don't get results like you?
A: It's most probably the safetensors model you use, you can notice in examples "anyhentai" produce the least amount of 'akatsuki souken' effect, yet others tend to produce more. The reason for that is because the other models are built to generally produce "plump women", so it's quite easy to stylize the eyes, breasts shape, etc. So you might want to add plump or mature woman or huge breasts, but add negative prompt (fat:1.3), so you don't get too fat stuff.

Useless NSFW info.
So if you used akatsuki souken model and you want that "manga panel" or "x-ray" effect on this Lora, when I edited the the images for Lora. I added "manga panel" for the images that have just a panel non-x-ray. Also I added "manga panels" tag if there were at least 2 panels (minimum 1 non-xray panel), I added "x-ray panel" if there is only 1 x-ray panel, and added "x-ray panels" for more than 2 x-ray panels. Usually If there are "2" x-ray panels, I would look at what the panel is showing and I would just write "cervix with lips" and "cervix sucking cock". Especially if that's what the x-ray panel would show. Go here and read all booru tags with: cervix,ovum,uterus, or whatever other tags was there, so you might want to bring that in your image (use ctrl+f). So "manga panel" tend to give body part of the woman, and x-ray panel give internal stuff. Also "cross-section" was added by waifu difussion for each image with internal stuff, so you might add that as well beside the x-ray panel tags.