Various usable model data specialized in image generation AI 'Stable Diffusion' Summary



Stable Diffusion , an image generation AI, is a ``latent diffusion model'' that generates images by removing noise. There are many forked models that specialize in generating specific images. I tried to summarize the specialized model that was derived from such Stable Diffusion and its characteristics and generation examples.

Stable Diffusion Models
https://rentry.org/sdmodels

The results of actually generating images with the same prompt, number of steps, and CFG scale with multiple models and seed values are as follows. The models are Stable Diffusion v1.4, Waifu-Diffusion v1.2, Trinart Stable Diffusion, Hentai Diffusion, Zack3D_Kinky v1 from the left. Although the general composition and color are similar for each seed value, you can see at a glance that the pattern is quite different depending on the model.



◆Waifu-Diffusion
``Waifu-Diffusion'' is a fork model of Stable Diffusion, and is trained on a dataset `

`Danbooru2021 '' consisting of more than 4.9 million SFW images (which can be viewed without problems at work) posted on the two-dimensional image site Danbooru. increase. You can see how it feels to actually generate an image using Waifu-Diffusion by reading the following article.

Summary of how to use model data 'Waifu-Diffusion' specialized for drawing illustrations with image generation AI 'Stable Diffusion' - GIGAZINE



Waifu-Diffusion has the following versions at the time of article creation. At the time of writing the article, v1.2 is the official version, and v1.3 is the beta version only. In addition, 'epoch' indicates the number of times the parameter is adjusted by repeating learning with the dataset, and the higher the epoch number, the better the accuracy can be expected.
・Waifu-Diffusion v1.2
・Waifu-Diffusion v1.3 beta epoch03
・Waifu-Diffusion v1.3 beta epoch04
・Waifu-Diffusion v1.3 beta epoch05
・Waifu-Diffusion v1.3 beta epoch06
・Waifu-Diffusion v1.3 beta epoch07
・Waifu-Diffusion v1.3 beta epoch08

◆Trinart Stable Diffusion
Trinart Stable Diffusion is a Japanese model. ' Trin-sama AI ', which is a model of AI for novel generation AI, has been specialized in two-dimensional illustrations such as anime and manga, taking over the art style of Stable Diffusion v1.4 as much as possible. It is an improved version.

naclbit/trinart_stable_diffusion_v2 Hugging Face
https://huggingface.co/naclbit/trinart_stable_diffusion_v2

An example of an image actually generated by Trinart Stable Diffusion



At the time of writing the article, three types of models have been published according to the number of learning steps (the number of parameter updates per iterative learning).
・

trinart2_step60000.ckpt
・trinart2_step95000.ckpt
・trinart2_step115000.ckpt

◆gg1342_testrun1_pruned.ckpt
It is a model trained with 280 NSFW (extreme content that should be viewed with caution at work) images and 80 SFW images by fictional characters, and is suitable for generating live-action images.

The model can be downloaded from below. Some models, including gg1342_testrun1_pruned.ckpt, are distributed by Torrent, and a client such as BitTorrent is required to download them.
・(BitTorrent required) gg1342_testrun1_pruned.ckpt

◆Hentai Diffusion
Hentai Diffusion is a fork model of Waifu-Diffusion v1.2, trained on 150,000 images uploaded to 2D image sites Rule34 and Gelbooru . Hentai Diffusion has repositories hosted on Hugging Face and GitHub, but GitHub was disabled at the time of writing the article because it violated the terms of use.

Deltaadams/Hentai-Diffusion at main
https://huggingface.co/Deltaadams/Hentai-Diffusion/tree/main

GitHub - Delcos/HentaiDiffusion
https://github.com/Delcos/HentaiDiffusion

Below is an image generated by Hentai Diffusion.



There are two types of models at the time of article creation.
・RD1212.ckpt
・RD1412.ckpt

◆Bare Feet / Full Body b4_t16_noadd
Bare Feet / Full Body b4_t16_noadd is a model trained on a carefully selected dataset of images containing bare feet and full nudity. Images actually generated with Bare Feet / Full Body b4_t16_noadd have been posted on the image sharing site Imgur.

Imgur: The magic of the Internet
https://imgur.com/2sJGz3j

The model is distributed by Torrent in fp16 and fp32 versions.

・(BitTorrent required) bf_fb_v3_t4_b16_noadd-ema-pruned-fp16.ckpt
・(BitTorrent required) bf_fb_v3_t4_b16_noadd-ema-pruned-fp32.ckpt

The developer said, ``I have noticed that normal SD and other models have many problems with drawing limbs. I created a dataset and checkpoint focused on the Bare Feet / Full Body b4_t16_noadd that seems to draw better than the other models when it comes to bare feet, and as a bonus it also draws the genitalia quite well It looks like it can, and it seems to work as a general NSFW image generation model as long as you keep the originality-focused 'stylize' low.It's not 100% consistent, but there are 100 decent images out of 100 that work. It was 0-1, now it's like 5. This model is still a work in progress, but I plan to finish it.Also, Bare Feet / Full Body b4_t16_noadd had a huge amount of catastrophic oblivion . Also, because it has become too specialized, it has lost a lot of versatility and stylization.'

◆Lewd Diffusion
'Lewd' means 'obscene'. Waifu-Diffusion uses the same Danbooru21 as Waifu-Diffusion for the dataset, but the one used by Lewd Diffusion contains NSFW images. In other words, it is a model specialized for generating two-dimensional NSFW images.

There are three types of models, Lewd Diffusion v0 is 20,000 images selected from the dataset, and Lewd Diffusion 70k 1e and Lewd Diffusion 70k 2e are learning with 70,000 images. '1e' and '2e' indicate the number of epochs respectively.
・(BitTorrent required) Lewd Diffusion v0
・(BitTorrent required) Lewd Diffusion 70k 1e
・(BitTorrent required) Lewd Diffusion 70k 2e

◆Yiffy
Yiffy is a model developed around the Discord channel of the overseas chemer community. We are learning with a dataset of up to 70,000 images posted on the chemer image posting site e621 . Yiffy has three types of models depending on the number of epochs.
・yiffy-e13.ckpt
・yiffy-e15.ckpt
・yiffy-e18.ckpt

Images actually generated by Yiffy are posted on Twitter.




◆Furry
Like Yiffy, Furry is also a chemer-specific image generation model, and has been trained with a dataset of 300,000 images posted on e621. Two types of models are available according to the number of epochs.
・Furry_epoch1.ckpt
・Furry_epoch4.ckpt

The image generated by Furry looks something like this.



◆Zack3D_Kinky-v1.ckpt
Zack3D_Kinky-v1.ckpt is also a model trained using over 100,000 images uploaded to e621. The image actually generated by Zack3D_Kinky-v1.ckpt is below.



However, since the Zack3D_Kinky-v1.ckpt dataset includes NSFW images, it is possible to generate NSFW images for chemoners. It is possible to generate beast images corresponding to a wide variety of propensities such as 'Transformation', 'latex', 'tentacles', 'ferals', and 'bondage'. That's what I'm talking about. The model can be downloaded from:

Zack3D_Kinky-v1.ckpt ~ pixeldrain

https://pixeldrain.com/u/DEocAHsx

◆r34_150k_epoch0.ckpt
A model trained on 150,000 NFSW images uploaded to Rule34.
・r34_150k_epoch0.ckpt

◆pony-diffusion
A model trained on a dataset composed of SFW images from the popular anime for girls overseas, My Little Pony . Therefore, we specialize in image generation of characters that appear in My Little Pony.

Below is an example of an image generated by pony-diffusion.



◆mio-wd-v1.2-e24-ex-ad
A fork model of Waifu-Diffusion learned with about 500 images and 24 epochs of

Mio Naganohara , a character that appears in the manga ' Nichijou ' which was also made into a TV animation. A fairly small model, specialized in generating images of Mio Naganohara.



mio-wd-v1.2-e24-ex-ad can be downloaded as 'epoch=000023-pruned.ckpt' from the following.

chavinlo/mio-naganohara-waifu-diffusion at main

https://huggingface.co/chavinlo/mio-naganohara-waifu-diffusion/tree/main

◆ fubuki-ld-v1-e13-ex-ad
About 5000 images of Fubuki Shirakami from the popular VTuber group Hololive , a medium-sized model learned with 24 epochs.
・(BitTorrent required) fubuki-ld-v1-e13-ex-ad

◆asuka-ld-v1-e4-ex-ad
A pruned model trained with 17,000 images and 4 epochs of Asuka, the character of the popular anime Neon Genesis Evangelion .
・(BitTorrent required) asuka-ld-v1-e4-ex-ad

◆tomoko-kuroki-ld-v1-e20-ex-ad
An image of Tomoko Kuroki, the main character of the popular manga ' I'm not popular, it's your fault! '
・(BitTorrent required) tomoko-kuroki-ld-v1-e20-ex-ad

◆70gg30LD70k.ckpt
A model made by combining gg1342_testrun1_pruned.ckpt and Lewd Diffusion 70k 1e at a ratio of 70:30. This does not have a specific model, but is made by combining it yourself, but it is also distributed in Torrent.
・(BitTorrent required) 70gg30LD70k.ckpt

◆wd1-2_sd1-4_merged.ckpt
The model is a combination of Waifu-Diffusion v1.2 and Stable Diffusion v1.4, but the ratio is unknown.
・(BitTorrent required) wd1-2_sd1-4_merged.ckpt

◆Hiten girl_anime_8k_wallpaper_4k.ckpt
Hiten girl_anime_8k_wallpaper_4k.ckpt is an image generation model that specializes in reproducing Hiten's designs with a high degree of accuracy by learning in 4000 steps from a total of 40 images of Mr. Hiten , an illustrator from Taiwan. It is possible to download the model from Hugging Face.

Bumbling Orange/Hiten · Hugging Face
https://huggingface.co/BumblingOrange/Hiten

Below is an image generated by Hiten girl_anime_8k_wallpaper_4k.ckpt.



In addition, when using multiple models properly, it is recommended to use 'AUTOMATIC 1111 version Stable Diffusion web UI' which can be easily operated with a browser and has many functions and can be said to be the definitive version of the web UI. You can understand how to install by reading the following article.

Image generation AI ``Stable Diffusion'' works even with 4 GB GPU & various functions such as learning your own pattern can be easily operated on Google Colabo or Windows Definitive edition ``Stable Diffusion web UI (AUTOMATIC 1111 version)'' installation method summary - GIGAZINE



Also, for how to change the model in the AUTOMATIC 1111 version of Stable Diffusion web UI, please refer to 'Change model data' in the article below.

Summary of how to use model data 'Waifu-Diffusion' specialized for drawing illustrations with image generation AI 'Stable Diffusion' - GIGAZINE

Related Posts:

in Review,   Software,   Web Service,   Web Application,   Manga,   Anime,   Art, Posted by log1i_yk