Stable Diffusion

From Soyjak Wiki, The Free Soycyclopedia
Jump to navigationJump to search
This page or section contains some outdated stuff!

(You) WILL help by adding relevant information.
A bedroom generated with Stable Diffusion. It's just a trans folx chilling in her room chud, stop noticing things.

Stable Diffusion is an open-source text-to-image AI image generator made by the tech startup Stability AI, released in 2022. Stable Diffusion can be ran at home, used with minimal programming skills, and can make almost any conceivable image. As such, Stable Diffusion has been used by soyjak.party users to make gemeralds. By utilizing its img2img feature, ‘teens can turn a pic of anything, from a hanging troonjak to a Gapejak to a bottle of soylent, into another image that depicts something entirely different, but is still eerily similar to the original.

Some of the soyjak stable diffusion images can be found on SoyBooru under stable_diffusion tag

DALL·E

Dall·E 2 is a text-to-image AI model by OpenAI released before Stable Diffusion was available to the public, thus making it a precursor to Stable Diffusion. It also watermarks all images, uses a freemium credits system, and has a TOS that prohibits even the slightest of wrongthink.[1] On July 31, 2022, a soyteen got early access to DALL·E, and proceeded to flood the soyosphere with dozens of AI-generated variants of cobson, schizojak, and more.[2] This ultimately led to AI-generated images being banned from the booru for a brief period, until the proliferation of Stable Diffusion soyjaks effectively reverted this rule.[3]

As of now, DALL·E is rarely used to make soyjak content. As the original DALL·E teen said, "Their content policy is incredibly autistic and watched closely, just the little hammer & sickle (referencing an image of Kuz in his iconic hat) might be in violation."

Bing Image Generator

You WILL fill this section with relevant information NOW

AKA Dall·E 3, way better than Dall·E 2 .

This one restricts generations be aware, nobody knows where the information come from but probably 4cuck did it

Running Stable Diffusion Locally

As long as you have a computer with a compatible graphics card, you can run Stable Diffusion on your own hardware. Doing this can not only save the hassle of many cloud instance's NSFW filters, but can also provide faster speeds if your computer is good enough.

AUTOMATIC1111 WebUI - Has more features than any other Stable Diffusion fork, including plugin support. WILL have to understand a bit of Python and Git to install it. A one-click installer exists for this WebUI, though the WebUI interface itself is more technical than other forks and websites.

>Do not attempt to create, upload, or share images that are not G-rated or that could cause harm.

cmdr2 WebUI - Insanely easy to download and use, but is lacking some of the features from AUTOMATIC1111's WebUI. Will work for the majority of 'jakking cases.

CompVis Stable Diffusion - The official Stable Diffusion release. Command line interface, bare minimum of features. Not reccomended.

If your graphics card is not supported, tough shit. You can either wait for it to be supported, use a cloud service, or try using CPU mode to get an image generated in a cool, crisp 30 minutes.

Running Stable Diffusion Online

If you are unable to generate a gem at home, there are many cloud services you can use to run Stable Diffusion.

Dezgo - Kinoful site with fast enough generations, img2img, no credits system, and no NSFW filter. Your best option if you don't have a good graphics card.

Replicate AI - Very fast generations, with no set credit system and good control over output, but has a forced NSFW filter on all models. The site itself has a good variety of models avaliable, including ones for ControlNet. Possibly paywalls you after some time, unconfirmed and untested.

Dreamstudio - Stability AI official web service. Freemium credits system, you start out with 200 credits. Hyperactive NSFW filter is present, which is impossible to disable.

Hugging Face - Can be slow, especially during peak hours.

List of other Stable Diffusion sites - Note that none of these have been tested, and likely have no img2img support, aggressive NSFW filter, a credits system, or all 3 at the same time.

An example of a GIF made with Stable Diffusion. Note the differences in low and high prompt strengths

Generating Images

Image to Image (img2img) creates a new image from a prompt, retaining some of the colors and shapes from the original. This setting is used to make the vast majority of AIjaks, being both versatile and recognizable. A good prompt strength to use for this is in-between 6 and 8, with 7.3 giving good results. Too little, and you have a slightly distorted messy-looking soyjak. Too much, and the image is completely unrecognizable as anything soy-related. A good image will be suspiciously soyjak-shaped, and yet ordinary at first glance.

img2img can also work on GIFs, though with some effort required. You'll need to run img2img on every frame, using a consistent prompt and/or seed, though you could just use gif2gif to automate this process. You could also keep generating slightly-changed frames from the same base image until you have enough to make an animation.

Most models based off the original Stable Diffusion model are not good for creating soyjaks from text-to-image, as there is no training data for "soyjak" in Stable Diffusion.[4] Using that as a prompt will always generate nonsensical combinations of actual soybeans, polish text, and products vaguely resembling vegan food. Wojak is represented in the database in the form of soyjak/chad comics, but rarely generates good results. Because of this, it's best to use a model made specifically for soyjaks when generating, of which the most IAS one is this.

Models

On December 15, 2022, a hypernetwork to generate chudjaks was released, to mixed reception.[5] Of note, the OP also misspelled soyjak as "soyjack" multiple times, including as "chudjack" in the generation keyword. You can download it here on Google Drive. (link broken, if anyone has a copy of the chudjak model, please reupload it.)

Models based off trainings of the SoyBooru exist for Stable Diffusion.

Soyjaks (LoRA) - The definitive model for generating soyboys

Two Soyjaks Pointing - For generating NAS images in the Two Soyjaks Pointing pose

See Also

https://wiki.installgentoo.com/wiki/Stable_Diffusion - /g/ wiki article on this

Stable Diffusion is part of a series on
Soyjak variants
Main variants [-+]

Variants with over 1,000 Booru posts

Classic SoyjakA24 Slow-Burn SoyjakGapejakMarkiplier SoyjakBerndFeraljakCobsonImpish Soyak EarsChudjak


Other variants [-+]
Subvariants [-+]
Soyjak trends [-+]
By artform [-+]
NAS [-+]

Main article: NAS

WojakNon-Wojak SoysWikipe-tanSoy-tanNPCGigachadPepeSidsonSoySubaChadjakPissluffarePoopyPoopson

Citations