`Stable Diffusion Models`

Quick & Easy Torrent Downloading
Models
Dreambooth
Upscalers
Face Restorers
1. GFPGAN
Textual Inversion
Model Comparison
1. Anime Model Comparison
2. Waifu Diffusion EMA vs Pruned

Quick & Easy Torrent Downloading

apt update
apt install -y aria2
aria2c "<magnet URL here>"

A model designed specifically for inpainting, based off sd-v1-5.ckpt. For inpainting, the UNet has 5 additional input channels (4 for the encoded masked-image and 1 for the mask itself) whose weights were zero-initialized after restoring the non-inpainting checkpoint. During training, synthetic masks were generated and 25% of the image was masked.

https://github.com/runwayml/stable-diffusion

Torrent

magnet:?xt=urn:btih:b523a9e71ae02e27b28007eca190f41999c2add1&dn=sd-v1-5-inpainting.ckpt&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce&tr=udp%3a%2f%2f9.rarbg.com%3a2810%2fannounce&tr=udp%3a%2f%2ftracker.openbittorrent.com%3a6969%2fannounce&tr=http%3a%2f%2ftracker.openbittorrent.com%3a80%2fannounce&tr=udp%3a%2f%2fopentracker.i2p.rocks%3a6969%2fannounce&tr=https%3a%2f%2fopentracker.i2p.rocks%3a443%2fannounce&tr=udp%3a%2f%2ftracker.torrent.eu.org%3a451%2fannounce&tr=udp%3a%2f%2fopen.stealth.si%3a80%2fannounce&tr=udp%3a%2f%2fvibe.sleepyinternetfun.xyz%3a1738%2fannounce&tr=udp%3a%2f%2ftracker2.dler.org%3a80%2fannounce&tr=udp%3a%2f%2ftracker1.bt.moack.co.kr%3a80%2fannounce&tr=udp%3a%2f%2ftracker.zemoj.com%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.tiny-vps.com%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.theoks.net%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.swateam.org.uk%3a2710%2fannounce&tr=udp%3a%2f%2ftracker.publictracker.xyz%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.pomf.se%3a80%2fannounce&tr=udp%3a%2f%2ftracker.monitorit4.me%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.moeking.me%3a6969%2fannounce&tr=udp%3a%2f%2ftracker.lelux.fi%3a6969%2fannounce

HuggingFace

https://huggingface.co/runwayml/stable-diffusion-inpainting/resolve/main/sd-v1-5-inpainting.ckpt

v1.4

Torrent

magnet:?xt=urn:btih:3A4A612D75ED088EA542ACAC52F9F45987488D1C&tr=udp://tracker.opentrackr.org:1337

Web

Download from HuggingFace. Fast, global CDN but you need to login and share your contact information with the repository.

https://huggingface.co/CompVis/stable-diffusion-v-1-4-original/resolve/main/sd-v1-4.ckpt

Alternatively, you could use this Google Drive link that the author of the WebUI shared:

Google Drive

https://drive.google.com/file/d/1wHFgl0ivCmIZv88hVZXkb8oy9qCuaBGA/view

VAE

StabilityAI has relased two autoencoders for Stable Diffusion.

https://huggingface.co/stabilityai/sd-vae-ft-mse-original#improved-autoencoders

ft-EMA

Resumed from the original kl-f8 VAE checkpoint, trained for 313198 steps and uses EMA weights.

https://huggingface.co/stabilityai/sd-vae-ft-ema-original/resolve/main/vae-ft-ema-560000-ema-pruned.ckpt

ft-MSE

Resumed from ft-EMA and uses EMA weight. Trained for another 280k steps using a re-weighted loss, with more emphasis on MSE reconstruction (producing somewhat "smoother" outputs).

https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.ckpt

Waifu Diffusion

Trained on Danbooru, slightly NSFW.

v1.3

Trained on 600,000 high-resolution Danbooru images for 10 Epochs.

The float16 version is smaller than the float32 (2GB vs 4GB). Always use float16 (unless your GPU doesn't support it) since it uses less disk space and RAM.

float16

Torrent

magnet:?xt=urn:btih:AWJJJZNFOOK7R2XXXGZ4GFNKUEU2TSFP&dn=wd-v1-3-float16.ckpt&xl=2132889245&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

HuggingFace

https://huggingface.co/hakurei/waifu-diffusion-v1-3/resolve/main/wd-v1-3-float16.ckpt

float32

Torrent

magnet:?xt=urn:btih:PK4UPXMW5TC7O2MKF7YJMIDX4NRBYO5R&dn=wd-v1-3-float32.ckpt&xl=4265384157&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

HuggingFace

https://huggingface.co/hakurei/waifu-diffusion-v1-3/resolve/main/wd-v1-3-float32.ckpt

EMA weights here.

v1.2

Trained 56,000 Danbooru images for 5 epochs.

Pruned Model

The full-size model was pruned to produce this model. Basically, all the unnessesary data was removed.

magnet:?xt=urn:btih:153590FD7E93EE11D8DB951451056C362E3A9150&dn=wd-v1-2-full-ema-pruned.ckpt&tr=udp://tracker.opentrackr.org:1337

EMA Model

There is no difference between the EMA and Pruned models.

magnet:?xt=urn:btih:F45CECF4E9DE86DA83A78DD2CCCD7F27D5557A52&tr=udp://nyquist.localghost.org:6969

The creators have provided a web download, but it's very slow.

https://thisanimedoesnotexist.ai/downloads/wd-v1-2-full-ema.ckpt

trinart_stable_diffusion_v2

Another anime finetune. Designed to nudge SD to an anime/manga style. Seems to be more "stylized" and "artistic" than Waifu Diffusion, if that makes any sense.

The 60,000 steps version is the original, the 115,000 and 95,000 versions is the 60,000 with additional training. Use the 60,000 step version if the style nudging is too much.

See the comparison below.

60,000 Steps

https://huggingface.co/naclbit/trinart_stable_diffusion_v2/resolve/main/trinart2_step60000.ckpt

95,000 Steps

https://huggingface.co/naclbit/trinart_stable_diffusion_v2/resolve/main/trinart2_step95000.ckpt

115,000 Steps

https://huggingface.co/naclbit/trinart_stable_diffusion_v2/resolve/main/trinart2_step115000.ckpt

trinart_characters_19-2m_stable_diffusion_v1

Trained on 19.2M anime/manga-style images. This model seeks for a sweet spot between artistic style and anatomical quality.

https://huggingface.co/naclbit/trinart_characters_19.2m_stable_diffusion_v1/resolve/main/trinart_characters_it4_v1.ckpt

VAE

https://huggingface.co/naclbit/trinart_characters_19.2m_stable_diffusion_v1/resolve/main/autoencoder_kl-f8-trinart_characters.ckpt

Hiten

Based off Waifu Diffusion v1.2, trained on the artist Hiten. More Info. I've heard people say this model is best when merged with Waifu Diffusion or trinart2 as it improves colors.

To use the model, insert Hiten into your prompt. For stronger results, append girl_anime_8k_wallpaper (the class token) after Hiten (example: 1girl by Hiten girl_anime_8k_wallpaper).

https://huggingface.co/BumblingOrange/Hiten/resolve/main/Hiten%20girl_anime_8k_wallpaper_4k.ckpt

WD v1.2 and SD v1.4 Merged

WD merged back into SD, I don't know what the ratio is. This was one of the first merges out there.

magnet:?xt=urn:btih:UFIV4BI4MGWFLZSKPFQ5VFLNYL24ADUQ&dn=wd1-2_sd1-4_merged.ckpt&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

NovelAI Leaked Models

On October 7th, NovelAI's server were hacked by some guy from 4chan. He stole their Stable Diffusion and GPT models. No user information/PII.

Part 1

55GB and contains the main models used by NovelAI, located in the stableckpt folder. I suggest using your torrent client to download exactly what you want or using this script. You can use this guide to get set up.

magnet:?xt=urn:btih:5bde442da86265b670a3e5ea3163afad2c6f8ecc&dn=novelaileak&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce&tr=udp%3A%2F%2F9.rarbg.com%3A2810%2Fannounce&tr=udp%3A%2F%2Ftracker.openbittorrent.com%3A6969%2Fannounce&tr=http%3A%2F%2Ftracker.openbittorrent.com%3A80%2Fannounce&tr=udp%3A%2F%2Fopentracker.i2p.rocks%3A6969%2Fannounce

Part 2

103GB and contains more GPT models and in-development Stable Diffusion models. You don't really need anything in here.

magnet:?xt=urn:btih:a20087e7807f28476dd7b0b2e0174981709d89cd&dn=novelaileakpt2&tr=udp%3A%2F%2Ftracker.openbittorrent.com%3A6969%2Fannounce&tr=http%3A%2F%2Ftracker.openbittorrent.com%3A80%2Fannounce&tr=https%3A%2F%2Ftracker.nanoha.org%3A443%2Fannounce

Unlisted Models

There's quite a lot of unlisted .ckpt torrents out there. Use a DHT search engine to search for ".ckpt"

https://btdig.com/search?order=2&q=ckpt
https://bt4g.org/search/.ckpt

Dreambooth

HuggingFace publishes community-submitted Dreambooth models. View them here.

https://cyberes.github.io/stable-diffusion-dreambooth-library/

It updates automatically twice a day.

Upscalers

Lollypop

For anthropomorphic figures.

magnet:?xt=urn:btih:JDZHD4PQVPHJU35C32GRJMSLYSR7YRHS&dn=lollypop.pth&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

Remacri Upscaler

For landscapes.

magnet:?xt=urn:btih:TNSKQM7JWWWOTIVEYS4GPSG2L2HFKSI7&dn=4x_foolhardy_Remacri.pth&tr=udp%3A%2F%2Ftracker.opentrackr.org%3A1337%2Fannounce

SwinIR

According to the developers, SwinIR achieves:

State-of-the-art performance in bicubic/lighweight/real-world image SR, grayscale/color image denoising, grayscale/color JPEG compression artifact reduction.

Get it on GitHub >