AI generated art

Mon0Zer0

Well-Known Member
Supporter
Joined
May 24, 2021
Messages
1,004
Generated using a deep dreaming algorithm in Google Collab and trained on Artstation.

TimeToDisco(0)_0.png


Below was generated, then edited afterwards in Photoshop.

CYBERPUNK-DISCO-EDIT.png

spaceship(1)_4.png
 
Okay, you have got me excited. Any supporting info appreciated.
I found this
And vids of some guy doing TLDNW interactive sessions.

I'll PM you a script you can run on google collab. You'll need to connect a google drive account.
 
I'll PM you a script you can run on google collab. You'll need to connect a google drive account.
Got it, thanks.
I don't have a google drive or anything beyond gmail but I can probably find someone I know.
 
Got it, thanks.
I don't have a google drive or anything beyond gmail but I can probably find someone I know.

If you already have gmail then you should already have a drive and collab account.
 
"The Censor" - this is a photoshopped AI image, using iterations 50,100,150 and 250, with added haze and a light beam.

CENSORSGAZE2.png
 
Last edited:
I'm sorry, is this "AI generated", or "AI collaged"? This stuff looks like other art that has been chopped up and stuck back together using some some algorithm that reads for common shapes and depth.
 
I'm sorry, is this "AI generated", or "AI collaged"? This stuff looks like other art that has been chopped up and stuck back together using some some algorithm that reads for common shapes and depth.

Generated. As I understand it, it starts with randomly generated noise as a seed (see below), and uses weighted probabilities to generate shapes based on a training set and shape recognition and in accordance with a prompt.

noise.png


50 iterations on from the noise - generated shapes (prompt was: "A spaceship by John Berkey")

noise+50.png


100 iterations from base

noise+100.png




The "other art chopped up" is also how the original art on which this is made (photobashing). It's a stylistic choice. For the warrior monk / censor images I'm using Johnson Ting's art as the intended look:


 
Last edited:
Generated. As I understand it, it starts with randomly generated noise as a seed (see below), and uses weighted probabilities to generate shapes based on a training set and shape recognition and in accordance with a prompt.

View attachment 89912

50 iterations on from the noise - generated shapes (prompt was: "A spaceship by John Berkey")

View attachment 89913

100 iterations from base

View attachment 89914



The "other art chopped up" is also how the original art on which this is made (photobashing). It's a stylistic choice. For the warrior monk / censor images I'm using Johnson Ting's art as the intended look:


Your description still strikes me that the AI has a catalog of "John Berkley spaceships" to work from, otherwise it wouldn't be able to produce spaceships.
 
Your description still strikes me that the AI has a catalog of "John Berkley spaceships" to work from, otherwise it wouldn't be able to produce spaceships.
This is true - but isnt this conceptually how human creativity works?

We just draw from our own organic memories in the same way.
 
Your description still strikes me that the AI has a catalog of "John Berkley spaceships" to work from, otherwise it wouldn't be able to produce spaceships.

The AI is trained on a body of his work (and hundreds of other painters, images and objects). There's no catalogue of images but a relational model, or a series of "weights". You can instruct it to paint something Berkey never once painted in his life and it'll use his sense of composition, colour and brush strokes to create something.
 
This is true - but isnt this conceptually how human creativity works?

We just draw from our own organic memories in the same way.
I don't think so. People don't generally read a lot of Balzac and then write a series of Balzac-like things.

The AI is trained on a body of his work (and hundreds of other painters, images and objects). There's no catalogue of images but a relational model, or a series of "weights". You can instruct it to paint something Berkey never once painted in his life and it'll use his sense of composition, colour and brush strokes to create something.
The AI isn't trained to think like Berkley or to have a rationalization of technology like Berkley. It is being trained to reassemble visual cues around certain "spaceship-like rules", which have more to do with some rules about up and down, symmetricality and perspective combined with "pointy to the left, holes to the right" guidance. It just doesn't seem much different than a text generator.
 

Similar threads


Back
Top