MENU

Over 100 Commercially Usable Illustrations! Introducing 6 Latest Models of Stable-Diffusion

This site uses affiliate links.
目次

Do You Like Sisters Who Are Eager to Undress?

Party Season is Here! Be Careful Not to Drink Too Much.

Hello, this is kojirom.
This time, I ended up with a lot of models and images that didn’t make the cut, and ultimately, the number of illustrations I could publish was just over a hundred… which now feels a bit strange. When I was drawing by hand, even managing to upload one a day was a challenge.

Anyway, when downloading models, I highly recommend visiting the official page. They have recommended prompts and detailed instructions, which are very educational.

Here is the link. ↓

Colorful – v4.6
Colorful – v4.6 | Stable Diffusion Checkpoint | Civitai
[Author = recoilme]

hamix – v1.0
hamix – v1.0 | Stable Diffusion Checkpoint | Civitai
[Author = terracottahaniwa]

LightsushiReal – v1.1
LightsushiReal – v1.1 | Stable Diffusion Checkpoint | Civitai
[Author = showywharf]

Rusalka – v0.5.1
Rusalka – v0.5.1 | Stable Diffusion Checkpoint | Civitai
[Author = Dolka]

The SR WonderMix – V3
The SR WonderMix – V3 | Stable Diffusion Checkpoint | Civitai
[Author = Lorim]

Ultimate Hentai – v2.0
Ultimate Hentai – v2.0 | Stable Diffusion Checkpoint | Civitai
[Author = Eddieddg1]

Actually, when I tried following the instructions on the page, I got some interesting results, so I’m thinking of publishing them maybe tomorrow.

All illustrations are free.
Feel free to take them as you please!

Colorful

ColorfulXL has been released. We are grateful for the feedback and examples of work from our users, which have been motivational.

In version 4.6, we created and trained with an ‘ugly’ dataset and merged it with additional differences. This slightly improved the anatomy, but resulted in overly saturated colors. We then retrained with ‘good images’ and merged using cosine similarity, which led us to discover that high-quality images could be generated with fewer steps.

In version 3.7, we achieved training speeds up to 10 times faster than SDXL. Normal results were obtained after 4 epochs, and we focused on training with a multi-aspect ratio for portrait and landscape generation. The optimal resolution is 768×960, and using negative prompts significantly reduces the image quality.

I couldn’t create SDXL overnight, but I am satisfied with the results.

(Summary of the Site)

“This is an advertisement.”

FANZA Illustration Collection (R18)

hamix

Here’s the English translation of your text:

“We recommend using tags such as an10, easynegative, verybadimage, verybadimagenegative_v1.3 for negative prompts.

(Summary of the Site)

LightsushiReal

This is a realistic synthesis of Darksushi and Maijikmix. You can create anything from anime to realistic portraits.

Credits for the original models are here: https://civitai.com/models/24779/dark-sushi-mix-mix https://civitai.com/models/43331?modelVersionId=176425

Recommended Settings

Sampling Steps: 15-30 (use more steps only as a last resort for very complex images, as excessive steps tend to ruin the image)

CFG Scale: 5-12

Optional Settings

Hires. Fix

Upscaler: 4x-UltraSharp

Hires Steps: 15-20

Denoising Strength: 0.20-0.40 (the stronger it gets, the clearer the image, but the original quality is lost)

Upscale Ratio: 1.5 – 2.0

You should use Adetailer for a better face

Adetailer:

Adetailer Model: face_yolov8n.pt

Inpaint Denoising Strength: 0.25 – 0.40

ADetailer Steps: 15 – 20

Works with most portraits and dresses with Lora. You only need to adjust Lora’s settings.

I am new to Stable Diffusion and Civitai, generating images on a regular PC. I have an AMD card, which did not support Stable Diffusion, so I created it with CPU only. It takes 45 minutes to 1 hour per image. I hope to create my checkpoint in the future. For that, I need support from the community. Whether you can support me depends on all of you. It would be a great support. Enjoy. Support here: https://ko-fi.com/showywhart

(Summary of the Site)

〈AD〉

Rusalka

The new version is not an upgrade but an alternative option, with its own art style.

A simple anime model based on multiple models.

Generate sample images using CLIP SKIP 1.

Use HIRES. FIX and ADETAILER on all images.

Incorporates VAE (kl-f8-anime2).

Enhance details using ‘intricate and complex details’.

Thanks to Skumerz’s model, using it as a starting point.

(Summary of the Site)

Recommended Article

The SR WonderMix

In the V3 version, the upscaler was changed to 4x-Ultrasharp

Aimed for a more realistic model, but not completely real

The results are not bad

Can generate from cute and sexy women to muscular beauties

Capable of creating backgrounds and male characters

Usable for NSFW content, but LoRa is more suitable

Most effective for portrait images

Detailed settings and examples of prompts are shown

Hope you enjoy it, and please post your generated images

(Summary of the Site)

〈AD〉

Ultimate Hentai

This work was created by combining Hassaku and Koji made by Ikena. For more details, please see his model pages for Koji and Hassaku.

(Summary of the Site)

Some illustrations may have incomplete details such as hands and fingers. We kindly ask for your understanding.

When using these images, please feel free to crop or modify them as needed.

We encourage you to fully express your creativity and use these illustrations with a free and imaginative approach.

The illustrations featured on this site are available for commercial use.

I look forward to seeing you again!

Illustrated by DALL-E 3

〈AD〉

目次