Steam Community :: Guide :: AI Art Generator Bites

Steam

All Discussions Artwork Videos News Guides

Steam > Guides > 🦈Da Great Magician Aike🐙's Guides

Not enough ratings

AI Art Generator Bites - Dive Into AI Art Generators

By 🦈Da Great Magician Aike🐙

This guide is about a nutshell guide about ai art generators.

Warning:

Under construction, since I'm still learning.

Award

Created by

🦈Da Great Magician Aike🐙
Offline

Category: Gameplay Basics, Walkthroughs

Languages: English

Posted

Updated

May 3, 2023 @ 8:06pm

Jan 19 @ 10:53pm

135	Unique Visitors
2	Current Favorites

Guide Index

Overview

Intro - The Future Is Now

Glossary

Wonkavision - How It Works

Things To Keep In Mind

Map Of The Art - Seeds And Stuff

Appendix

Tool Box And Credits

Other Guides

Other Guides (Cont.)

Other Guides (Cont.)(Part 2)

Other Guides (Cont.)(Part 3)

Comments

Intro - The Future Is Now

All the ai art that ya see in this guide is raw,
they don't have fixes, except by Fire Alpaca SE.

Warning:

There's some legality involving ai art generators, so, before ya go REALLY gunho with them (like other monetary activities). I would advise to just wait for things to settle down. - I won't go into deets (Because, this is a "happy, happy, joy, joy" guide).

This guide is mainly a tiny intro of what are ai art generators and a very basic outlook about these tools. This won't talk about the murky grey ethics and those other stuff concerning it. This is a merely a "happy happy joy joy" guide.

Note: I'm on survival mode (being a bit worried about being replace). So, if ya wanna become an ai art artist of sorts. I won't be able to help ya much. All the ai pieces that ya see are either made just for this guide alone.

Second Note: I only have experience with some version of Stable Diffusion (like DreamUp, Craiyon, and Dezgo). This particular art generator is "easy" to pick up, due to it's open source-ness. The others either need a subscription, like Midjourney, or are in a too close of a wall garden, like Firefly, only accessible to very, very few people,* or I don't have any knowledge of the other ai art generators (like Deep Floyd or Leonardo ai).

Third Note: Help me correct the grammar, misinformation and misspellings. I am not the best prompter/img2img-er, I'm just winging it, and Dezgo allows me to do that (It has an image-to-Image feature that I like.). Thanks in advance.

Don't forget to:

* I won't go through my usual "Buyer beware" warnings tirade about Adobe, since this is suppose to be a "happy, happy, joy, joy" guide. Plus, you're here for the generators.

Glossary

AIIA - 1) An acronym for "Artificial Intelligence International Arts" festival. 2) An acronym for "Artificial Intelligence Infused Art."

Automatic1111 - Automatic1111 is a Stable Diffusion webui. Advance users love this GUI (aka "A1111").

Batch - Batch is a feature that lets one perform image-to-image for tons of images or does inpainting (Fixing mistakes in a geneated art).

CFG Scale - Short for "Classifier Free Guidance scale". It's mainly a parameter of control of how much the model should diviate or stay within your prompt. It ranges from 1 to 30:

1 - Ignores your prompt mostly.
3 -Be more creative
7 - A balance between running wild and following your prompt.
15 - Adhering to the prompt.
30 - Strictly follows instructions.

Checkpoint Merger - This is one of A1111 features were one combines two or more models. Checkpoint Merger is usually use for mixing the styles of two or more models.

CodeFormer - CodeFormer is a face image restoration ai program.

ChatGPT - A very versatile and ai generator that can be program and use for almost anything. It's a jack-of-all trades tool, it can be used as a plug in or a even use to train an ai art generator.

Computer Vision - It's the process that a computer that "learns" to identify stuff, like items or people. (Aka Machine Vision)

ControlNet - ControlNet is a neural structure that adds control and extra conditions to diffusion models, this is very helpful towards ai generators.

Denoising Strength - The "Denoising Strength" controls the noise that's been added to the latent images. This happens before doing the "Hires Sampling Steps."

Fine-Tuning: - A process that's training a base model with an additional dataset.

Hires Steps - This process is only for latent upscalers, the Hires Steps samples the steps after an image gets upscaled.

Image-To-Image - A process were the user uploads an image into the art generator to make an ai art.

Img2Img - A slang term meaning "Image To Image."

Inpainting - Inpainting is the act of fixing their text2images that have imperfections. There's three kinds:

Inpainting - The tool that's in ai art generators that artists use to fix their artworks.
Inpaint Sketch - This tool allows one to paint/draw like when sketching, but it only regenerates the painted area. The rest is unchanged.
Inpaint Upload - This type of inpaint is when one uploads a separate mask file.

LAION - A large non-profit organization that makes humongous AI learning models, datasets and other code stuff that's ready made available to the general public. LAION also does machine learning research.

LoRA - It's an acronym for "Low-Rank Adaptation of Large Language Models". LoRA is mainly an editing program developed by Hugging Face to help people fix up the "text-to-image" generated art. It's the cousin to Adobe's Dreambooth.

ML - An acronym for "Machine Learning."

Pickle - Pickle software are malware scanners to make sure that models don't come with nasties, for example, "Python."

Pickling - The act of "pickling" is taking measures to protect the computer of getting malware, like running a scan or installing a pickle software.

SadTalker - An addon that captures facial expressions.

Seed - The "Seed" is a value of numbers that tells the ai generator of how to generate certain image. Each generated art work has their own special seed value. (The "Seed" stuff is not exclusive to the ai art generator scene).

Steps - It's telling the ai art generator how many times that they should work on the prompt.

Synthography - It's the method of producing art works by using an artificial intelligence robots to create music, songs, art, written works or videos. Aka "AI art."

Text-To-Text - A processes were the user types instructions into the art generator to make an ai art.

Tiling - The "Tiling" is an option on in Auto1111 that can make an image that can be tiled.

TTI - An acronym for "Text-To-Image."

Wav2Lip - This is an improved version of LipGan, a lip sync addon.

Weight - It's a term of how much intensity that ya want your LoRA to be.

Wonkavision - How It Works

How these art generators work is that it's similar how Wonka Bars are made. This chapter is a nutshell version of the process, the references that I have go deeper than that.

I'm insanely simplifying things here. Just think of the "art bar" representing billions upon billions of artworks that's being trained for just one thing.

The large bar is the art piece and that camera laser is the ai art generator.

The lazer is the art generator that dissolves it, mainly creates it into noise/static. (The original art work is thrown in the trash, while all of the dissolving and noise is going on).

The final end of the journey is that the style that's the only thing that's left of the original piece.

Things To Keep In Mind

Town Mouse And Country Mouse

The copyright issues and concerns are mainly a North American cultural thingy AFAIK*, there's other counties that don't really care or outright welcome the ai art technology.

* I am a stupid American, so, I dunno which countries are pro, neutral or anti-AI Art and how their copyright or fair uses work.

Some Industry Vets

There's some veteran artists, like

Arron Blaise

Vid Link in case vid ain't working.

James Gurney

Link in case vid ain't working.

That see this ai art generators as a good thing or will eventually become a good thing.

Art Generators

There's different types of ai art generators. I categorize into three simplistic categories:

The image-to-image - The user to uploads an image into the generator to create a new piece.

The text-to-image - The user types in the prompt into the generator to create a piece.

The Hybrid - The user can use use either the prompt, the image or all of it's features to create a piece.

Payment

The Free/Freenium - This type of ai art generator is free to use, just download or go to a site and start creating. The payment methods are optional.

Subscription - These ai art generators are subscription base. They want money, after enduring the free trial/finishing the free trial period.

Sky's The Limit

People have been using the ai art generators to make "green screen" effects for their cartoons. While it's true that some jobs will get phase out/semi-phase out or have to do an insane restructuring/evolve, new jobs will come out of ai art generators. Some of these mean that ya have to play your cards right.

Here's some examples:

Art and design background
Fine artist
Art career - If provided that one knows how to navigate and understands the different art markets, how to do strategic pricing and financial aspects of this art world.
illustrator
Machine learning scientist
A Vtuber, like Preston Ch.
Data Visualization - Being able to do AI art will come in handy when dealing with tools, like ggplot2 or matplotlib.
Associate Art Director
Brand Designer
Programmer jobs (like for video games)
Concept Artist
Social Content Designer

Tips: Don't work for:

Companies that ya know will really replace ya for the bot 100% - They don't even want ya as a prompter.
Black Companies or companies that ya know that have the IQ of a Black Company, they're mainly sweatshops.

Note: This is a basic guide, I'm teaching ya about your options. Not to make an intro for a YT vid intro or some high caliber movie.

Map Of The Art - Seeds And Stuff

Dumb cat is a "fake" gamer girl, she's holding
the PSP or Switch incorrectly.

Depending on the art community surrounding ai generators, some are very big on sharing the workflow, it's an extreme form of "sharing is caring". While others communities aren't/neutral/ "It's all up to ya."/don't care. So, in order to be polite or if ya really wanna join them, the strict communities, just list your ai art workflow in the art piece description. It's frown upon, if ya don't do that.*

So, this is a wonky demonstration on how to share.

Example

Image Prompt/Prompt: (like video gamer girl, wearing cat ear hairband, messy room, chibi)

Generator(s): [Insert ai art generator here] (like Denzgo)

Model(s): The ai that's used to generate the image (like Anything 3.0 (anime))

Seed: Random number (like 2434878567) - The seed is one of the important ones, it's so ya (and your friends) can re-generate the same particular piece.

Workflow Tips:

The workflow usually contains: the prompt/image, negative prompt (optional), CFG scale, resolution, seed, steps, checkpoint and other nitty gritty stuff.

It also includes stuff like "Several img2img steps", inpainting on troublesome areas, strengths, touch ups and other things.

* The reason why some ai art communities, like Stable Diffusion, are strict and encouraging to post prompts, workflows and other nitty gritty stuff. It's to get rid of some of the negative stereotypes, like the "prompt and post the final results."

Appendix

Tool Box And Credits

Fire Alpaca SE
Intuos Comics (Wacom tablet)
Official Steam mobile app
Medibang Paint
Dream Up (ai art generator)
Dezgo (ai art generator)

Screenshot Credits

Willy Wonka & The Chocolate Factory - It's WonkaVision Scene
Among Us (my gameplay footage)
Bleach Episode 264

AI Art Works

Be aware that, I don't think that ya can generate the same pieces, even though I have listed my backtracks here. Most of the stuff here is vague when I made these (and I don't know what I'm doing). Also, they're raw, I cleaned them up via FireAlpaca SE. - I didn't use any of Dezgo's inpainting features and other tools to fix them (except for one piece).

Ya can try though.

With the exception of the Dream Up Creation, I used Fire Alpaca SE to fixed the stupid.

Dream Up Creation

Prompt: chibi cute llama in a meadow

Generator(s): Dream Up

Gamers - The Red Duo

Wokrflow

Prompt: science fiction landscape, a red haired cat girl is typing a video game guide on her PC

Generator(s): Denzgo

Model: Vintedois Diffusion (simple, general)

Sampler: Euler Ancestral

Guidance: 7

Seed: 1178965665

Fixed Up Version)

Mask Prompt: Cat's iris are diamond shape.

Generator(s): Denzgo

Prompt: Woman types while cat looks on

Model: Stable Diffusion Inplant 2 (inpaint, general)

Generator: Denzgo

Seed: 1178965665

Tools: Fire Alpaca SE

BTS Makings:

https://steamcommunity.com/sharedfiles/filedetails/?id=2971005953

https://steamcommunity.com/sharedfiles/filedetails/?id=2971026184

Moonlit Howl

Workflow

Prompt: it's a full moon night in the forest and there's a wolf howling

Generator: Dezgo

Model: Epic Diffusion 1.1 (general)

Sampler: Euler Ancestral

Guidance: 7

Seed: 3156299997

BTS Makings:

https://steamcommunity.com/sharedfiles/filedetails/?id=2970998375

https://steamcommunity.com/sharedfiles/filedetails/?id=2971007557

Gamer Cat Girl

Workflow

Prompt: video gamer girl, wearing cat ear hairband, messy room, chibi

Generator: Dezgo

Model: Anything 3.0 (anime)

Guidance: 7

Sampler: Euler Ancestral

Seed: 2434878567

BTS Makings:

https://steamcommunity.com/sharedfiles/filedetails/?id=2971022425

https://steamcommunity.com/sharedfiles/filedetails/?id=2971019664

Non-AI Art Works:

My Fanarts