MidJourney - The AI generated images - robots painting dreams of electric sheep

[Saint Agustin]

Midjourney is an AI generated app running currently on Discord that takes your prompt and creates rendered artwork. In comparision with Dalle and Dalle-2, this AI is more focused on creating abstract renditions of the prompt than to bring to life accurate representations of it.

Right now they went into Open Beta, so their Discord is open for everyone to try to experiment with the software.

Discord here

Full site here

Some notable images I found/did on the fly:

Neon Tiger

Empty Office at night scary photorealistic

neon clouds in a sunset valley

Spoiler: more random ai images

red angel in the night

Jesus cyberpunk style

robot visits 1970s Ohio mall photorealistic

textured Cthulu Ghost in dark moonlight

Abstract invasion of evil angels

Spoiler: more random ai images

galactic samurai jedi fighting

an unfathomably violent alien race experiments on humans

robots dreaming of electric sheep (this is going as my screensaver)

Also the Discord has like 7 free trials (didn't confirmed but something like that). So if you just want to keep generating images just create another discord. Discord is gay, so is easy, it takes you like 2 minutes at most.

The pay version has a lot more channels and more features I believe, but because I'm a poorfag can't access it. Nevertheless, this is freaking fucking cool, and lots of these images can still be improved in photoshop to create something like album covers or stuff like that in my opinion.

 

[Just A Butt]

how is this different to the other AI art things that i don't need discord or money for?

 

[Celebrate Nite]

>paying for AI to draw images
>using Discord to have Ai draw images
Craiyon literally does the same thing and it's free

Craiyon, AI Image Generator

Craiyon is an AI model that can draw images from any text prompt!

www.craiyon.com

 

[Toolbox]

how is this different to the other AI art things that i don't need discord for?

slightly higher resolution than dall-e mini and allows you to choose iterative changes? That might be another project that also forces you to use discord though.

This video compares the major ones out right now.

Their shortcomings, their similarities, what they are best at. I am particularly suspicious that these AI are not nearly as inventive as their developers put forward, and there's a part in this video showing how some of these ai can be seen almost copying directly from existing photos. It all comes down to how they are trained.

 

[The Mass Shooter Ron Soye]

this AI is more focused on creating abstract renditions of the prompt than to bring to life accurate representations of it.

DALL-E 2 was already pretty good at that, I think. As long as you give it the right prompts:


I'm going to look at Disco Diffusion later:

https://analyticsindiamag.com/in-the-world-of-dall-e-2-and-midjourey-enters-open-source-disco-diffusion/

 

[Saint Agustin]

>paying for AI to draw images
>using Discord to have Ai draw images
Craiyon literally does the same thing and it's free

Craiyon, AI Image Generator

Craiyon is an AI model that can draw images from any text prompt!

www.craiyon.com

Craiyon is shit.
First Crayon or dall-e takes longer and the results are incredibly, superbly shitty

Case in Point

Gordon Ramsey as an Avenger (by Craiyon)

Gordon Ramsey as an Avenger (by Midjourney)

like water and oil.

 

[Mapface]

Something caught my eye today. all this A.I. image generation stuff.

NFT projects are usually about 10,000 images each.
10k images people are willing to pay more than $100 for. Scrap those dollar figures for now, but with that said, we know there is demand for these images.

Combine this with a DAO (*cough cough*) that approves images. And does so for the sake of bandersnatch style story telling.
Yes, a crowdsourced bandersnatch game.

The only thing left to debate is "how can you make a fair DAO" and I can cover that for you.

The post below. 250k images, and text to go with them.
250k is nothing.

I was thinking about this the other day "how many images need to be produced to be valuable to a learning algorithm"
250k is nothing.

I'm not in A.I. I'm in blockchain, so I couldn't have known. But now I do.

The value of the bandersnatch data is arguably more valuable than what our boy John is talking about. Images with text that describe events happening from page to page, choice to choice. That's rich training data. And people are incentivized thanks to innovative blockchain tech to keep pumping images into this thing

 

[FatalTater]

MidJourney does not know what corn is. Or at least, it doesn't know where it comes from.

Try getting it to produce anything like a corn field, corn stalk, foddershock, anything like that. Strangeness occurs.

 

[Drain Todger]

This one is very cool. I asked it to do "two giant mechs fighting in the countryside renaissance tapestry style":



"The one sinister-looking man who secretly holds the fate of the world in his hands, lurking in the shadows":



It's weird, but these neural networks tend to do better with descriptive, emotional prose than with short and literal prompts.

Let's give it another go. "The chief representative of clown world, honking his nose at poor, silly, obsolete humanity":



Aaaaah, god! What the fuck?!

Let's hit U4 and V4, just for shits and grins.



Okay, okay. Let's try something else.

"generic anime avatar hero with a determined look on his face":



That's pretty generic. Or, wait a minute, let's try that prompt I used with Craiyon the other day.

"Donald Rumsfeld eating a hot chicken chili and burning his lip because it's too spicy".

Wait. They banned the prompt because the string contained "hot chick". Okay, let's swap the words around.

"Donald Rumsfeld eating a spicy chicken chili and burning his lip because it's too hot":



His head became the bowl of chili. Magnificent.

 

[Drain Todger]

This is the coolest goddamn thing I have ever fucking used, to be quite honest. I went ahead and got a $10 monthly subscription because why the hell not? Deep Learning tools are very costly to use, processor-time wise. You're basically making API calls to a fucking supercomputer every single time you use one of these. Nothing wrong with paying money for literal supercomputer time.

I had it do everything from art deco portraits of tyrants, to Beksiński paintings, to Lovecraftian hellscapes, to images of pollution and environmental devastation, to character designs, to art deco concrete bas reliefs in the style of H.R. Giger, to Blade Runner-esque brutalist architectural photos. If nothing else, this is probably the best goddamn avatar generator I have ever seen. It's amazing how far this shit has come in such a short period of time. This thing could be used as a serious tool and architectural design aid. Collaborative human-AI reciprocal generation of art. Prompt it to design a relief, and then actually do the relief and make a mold for it by hand and cast it in actual concrete. These designs and textures are incredible. Inspired, even. And they're coming from a machine with no self-awareness at all. Bizarre.

Spoiler: More Stuff

 

[SuperShibuyaFighter]

I hate how Midjourney is used via tranny cord

 

[Drain Todger]

I hate how Midjourney is used via tranny cord

Making it into a Discord bot was a stroke of genius on their part. It means that they get to save shitloads of money by offloading all their traffic onto Discord's servers.

I used it to generate a new avatar and banner. It took very, very few reiterations to get results:





I had it do another Rumsfeld.



Then, someone else showed me how, with another pass through a different "face-fixer" algorithm, the glitches produced by the AI can be repaired and the image denoised:



This shit is crazy.