AI generated images

AI generated art is really neat and I like to showcase some examples of what it's like. I'm using a Google Colab program called "Looking Glass" which used the Russian AI, ruDALL-E (https://colab.research.google.com/drive/15MQTRMqc7yn626PiC3KPb-DvVx_FrBAr). Using what I can only assume is black magic the AI takes and input image and tries to understand what it is and dumps out an iteration of that. The program is a little finicky and it has it's limitations but the results are really fascinating. If you want to use the program for yourself be aware it'll take a bit of time to generate the images and I recommend you watch this tutorial.

For this first example I'm using a picture from the illustrated children's book, A Bad Case of the Stripes by David Shannon, it is an already surreal book so I figured it would be perfect for this program. Here is the sample image:


And here are the results:

Spoiler: Results

Every one of these would make a great cover for your next indie progrock album

 

Next I put in the American Gothic painting, "Death on Ridge Road" by Grant Wood another striking and surreal painting:

Spoiler: Large picture

The results:

Spoiler: Results

These are perfect for posting on conspiracy image boards claiming to be photos of other dimentions

 

This thread may need the "wat" and "heresy" tags.

 

Have you tried any of Chris's drawings with this? Or would that break the AI or unleash an unholy demon on the world.

 

Great idea!

First up our old pal sonichu!

Spoiler: Results

The first one looks like a Paul Mccartney solo album cover from a hell dimension, the second is my favorite, it looks like faded illustration on Japanese cigarette carton. The rest amazingly look like actual drawings and if you squint hard enough you can see sonichu in them.

 

Next up is the classic cover to the first issue of sonichu

Spoiler: Results

Half of these look like impressionist art pieces and the other half look like children's drawings if the children were on heroin.

 

Next is the classic picture of Chris in his room:

Spoiler: Results

Every picture here is a masterpiece

 

heresy that must be driven back into the warp

(also thanks for adding the wat tag)

 

Holy shit, these are incredible.

It's like Being John Malkovich. Is this how Chris sees the world?

 

this is amazing, I love this
I added heresy and not heresy, as a paradox

 

Looks like the holder of rainbow text can add tags in threads that can normally have tags only modified by OP or mods.

(also thanks)

 

Decided to try some charactures, first we have @Fauna Fox 's favorite fury bait, Krystal from Starfox

Spoiler: Results

The 3d void ate this one, jeez, the first one is pretty great though

 

Next is Misato from Evangelion, a test of the AI's ability to replicate anime:

Spoiler: Results

The AI must get a lot of anime as training data because these didn't turn out that bad, I love that stanky arm on the second to last one

 

I decided to put in my own art and see what happens, first up is the Sonichu main cast I drew for the Sonichu comic thread:

Spoiler: Results

While the AI didn't really output something that resembled what I drew it did make some cool and creepy heads. I guess having multiple faces in the picture makes the AI freak out a bit.

 

Next up is something I drew just for this thread:

Spoiler: Results

 

Is this AI man enough to take on the Throbbing King of Desire himself?

Spoiler: Rated M for Mara

 

This time I generated three batches of images, high, medium, and low universal similarity, for all of the past generations I used medium so I wanted to see how this setting would effect the results.

First up is high similarity, which means AI attempts to make the image as close as it can to the original:

Spoiler: High universal similarity

 

Spoiler: Medium universal similarity

 

Spoiler: Low universal similarity

This batched turned out so good I did two