AI Image Generation

Captain Suave · Sep 30, 2022

Most of you have seen the text-to-image products of Dall-E, Midjourney, etc. Stable Diffusion is a comparable approach which you can run on your own hardware.

Stable Diffusion Public Release — Stability AI

We are delighted to announce the public release of Stable Diffusion and the launch of DreamStudio Lite.

stability.ai

How to Run Stable Diffusion on Your PC to Generate AI Images

You can generate AI art on your very own PC, right now. Here's how to use Stable Diffusion.

www.howtogeek.com

GitHub - CompVis/stable-diffusion: A latent text-to-image diffusion model

A latent text-to-image diffusion model. Contribute to CompVis/stable-diffusion development by creating an account on GitHub.

github.com

I have an 8GB 2070 Super and kept running out of VRAM with the main version, so I ended up running an optimized fork which exchanges speed for RAM allocation. It still produces images in well under a minute.

GitHub - basujindal/stable-diffusion: Optimized Stable Diffusion modified to run on lower GPU VRAM

Optimized Stable Diffusion modified to run on lower GPU VRAM - basujindal/stable-diffusion

github.com

There's also a fork which enables HD upscaling without requiring multiple GPUs to hold the problem.

Want Stable Diffusion in HD? This AI Art Generator Delivers

Enhance!

www.howtogeek.com

GitHub - jquesnelle/txt2imghd: A port of GOBIG for Stable Diffusion

A port of GOBIG for Stable Diffusion. Contribute to jquesnelle/txt2imghd development by creating an account on GitHub.

github.com

It took a bit of fiddling for me to get everything working. The process requires some comfort with Git repositories and Python, and the model is run via CLI.

There's definitely some art in sculpting the prompts to get what you want. I'm not good at it yet, but there are lots of examples online.

Tuco · Sep 30, 2022

huge fan of AI art. I've got a large collection cycling on a TV in my office, it's great.

AI images

This thing generates some really cool images! https://www.midjourney.com/ Prompt (not mine) 'bear warrior realistic'

www.firesofheaven.org

Dall-E AI image generator

Dall-E is a new AI image generator that creates completely novel images from a text prompt. Sometime the results are junk but some are amazingly accurate to the description. Sometimes it creates strangly compelling surrealistic art. Here are a few I've had it create. Create your own at DALL·E...

www.firesofheaven.org

Captain Suave · Sep 30, 2022

First lame attempt:

"the fires of heaven"
seed 293258, upscaled

In all my other samples it keeps trying to put text in the picture. I think this prompt phrase is heavily biased towards the Robert Jordan book covers. Looks like it's about to take a flaming dump.

Flobee · Oct 1, 2022

Figure this is a good spot to drop this. Video of a dude generating game art with AI from crappy drawings. He can make really serviceable inventory icons in a few minutes, pretty wild.

Kharzette · Oct 1, 2022

There was a debate if this stuff should be allowed in Ludum Dare. I never went back to see if it was ok'd or not. Pretty neat.

Tuco · Oct 2, 2022

Kharzette said:
There was a debate if this stuff should be allowed in Ludum Dare. I never went back to see if it was ok'd or not. Pretty neat.

I have to imagine that game art sill embrace ai art waaaaay more than any other art uses given how many game devs would love to mess around with ai and the barrier for entry is so low.

I think for other competitions, ai will just be its own category.

Big Phoenix · Oct 19, 2022

Got stable diffusion up and running on my computer.

Sometimes things go alright;

Sometimes things go horribly wrong;

Also low/mid tier NSFW artists are gonna be out of a job soon enough;

pharmakos · Oct 20, 2022

Stable Diffusion (via NightCafe) is very good at making abstract/surrealist Salvador Dali style paintings

2286501071__The_Birth_of_Yaldabaoth__by_Salvador_Dali__1931__surreal.png

Just a very small sampling from this very deep rabbit hole I'm going down

Big Phoenix · Oct 21, 2022

The novelai is 100x better at creating characters than either the stable diffusion or waifu diffusion final models.

Spent a few hours with that and ended up with this;

Pasteton · Oct 22, 2022

If someone spends hundreds of hours fine tuning an ai output to create a masterpiece, are they now an artist

pharmakos · Oct 22, 2022

I bet back 700 years ago, people bitched that the famous rich artists of the world had access to better pigments and brushes and therefore their art required less effort, and therefore wasn't real "art"

Big Phoenix · Oct 23, 2022

To get good results you cant rely on AI alone(at least now, god only knows in the future), you have to edit and touch it up manually in your editing software of choice.

For example in the above picture I had to go into paint and draw a very crude outline of the face I wanted as i couldnt get it to randomly generate anything close to what I wanted. But given it the outline it was able to get going in the direction i wanted;

Also use the power of editing to give her a better stomach;

If youre an already skilled artist, this has to be an amazing tool to speed up your workflow.

Also SD is weird. You change one setting and it gives you something interesting though not what you wanted;

Kharzette · Oct 26, 2022

I have installerized it. It is neato

Kharzette · Oct 26, 2022

This is fun

Kharzette · Oct 26, 2022

I was trying to do a nice angel. Best I got was:

But then I switched over to img2img and put in this input image:

Screen_Shot_2021-11-03_at_12.03.40_PM.jpg

and here's the result:

Big Phoenix · Oct 26, 2022

Gotta lower down the denoise strength. Anything above .6 you start getting significantly altered results.

Kharzette · Oct 26, 2022

Your denoising tip helped alot with inpainting. The cross on the choker was coming out as a blurred fuzzy blob till I cranked up the denoising to 0.69

It looks kind of crooked but I like it that way.

Kharzette · Oct 27, 2022

This is fun, I've been adjusting screenshots of various games. Merging in the AI-ized bit is always tricksy though. I've been using lazy blurs and such, but I realized you can load the whole thing in, go to inpaint and draw over the thing you want vanishized, and do masked content fill. I have to go low res because I only have 4 jigglybytes:

The AI-ized:

And then splice in the magic:

Kharzette · Oct 27, 2022

Some earlier ones I did the lazy way:

Kaines · Oct 27, 2022

Kharzette said:
This is fun, I've been adjusting screenshots of various games. Merging in the AI-ized bit is always tricksy though. I've been using lazy blurs and such, but I realized you can load the whole thing in, go to inpaint and draw over the thing you want vanishized, and do masked content fill. I have to go low res because I only have 4 jigglybytes:
View attachment 440023

The AI-ized:
View attachment 440024

And then splice in the magic:
View attachment 440025

What Luclin Models NEEDED to have looked like....

AI Image Generation

Caesar si viveret, ad remum dareris.

I got Tuco'd!

Caesar si viveret, ad remum dareris.

Vyemm Raider

Watcher of Overs

I got Tuco'd!

Pronouns: zie/zhem/zer

soʞɐɯɹɐɥd

Pronouns: zie/zhem/zer

Blackwing Lair Raider

soʞɐɯɹɐɥd

Pronouns: zie/zhem/zer

Watcher of Overs

Watcher of Overs

Watcher of Overs

Pronouns: zie/zhem/zer

Watcher of Overs

Watcher of Overs

Watcher of Overs

Potato Supreme