Home Artists Posts Import Register

Content

I've been trying something new this week! Artificial Intelligence (or Machine Learning) programs that can create an image based on text input. This is *very* different from my usual process and I wanted to test ideas and techniques to see what I could do with it.

This is a gallery of "My best so far"

The first Three images are taken from five images have been enhanced to bring out better facial features and to correct some minor problems. The last four are entirely unaltered, so you can see that I am already getting better at telling the program what I want to see.

Something that I was hoping for was to have it create a reasonable image showing super-strength or a giant rampaging through a battlefield or something. I have not got the hang of that yet. The A.I. seems to have a lot of trouble with relative scale - and it will do things like put a guy *next* to a car, instead of lifting if off the ground.

Back to normal, but I will continue to explore and see what I can improve by using this as another tool in my arsenal. 

Files

Comments

Walter37

A.I. art, when the dark side of tech can't draw hands. ^^

pratev

Yes, or putting hands where they are not supposed to be - like a huge fist on the end of a muscular neck - argh! I will continue to play with it. :-)

S Munson

But you know what? These are the very same utterings i heard when I first met you, and you were experimenting with and learning another program. It's amazing whet this AI program is. Even the mistakes look good!!!!!!!

Eddy

Your AI-enhanced pictures are much better than the ones I've seen so far. Faces are at best blurry but they usually look "melted". And let's not talk about the hands and feet... Maybe they first should teach the AI algorithms what a human being looks like 😅🤣

pratev

Well, thank you! But, I didn't show the *worst* of the mistakes - and believe me, some of them were very disturbing! (Faces poking out from the sides of bodies, reversed legs, so the buttocks were in front merging smoothly into the abs, two heads, two torsos, and three legs, all somehow joined at the pelvis - real nightmare fuel. I just need to figure out where this fits with the rest of my artwork.

pratev

Thanks! Yeah, I had to work way harder than I wanted to to get results that were acceptable for public posting. I used (in some cases) three or four different similar images and tried to mash them all together - but there were subtle differences in positioning and coloration. The thing about machine learning is that there is no cognition behind it. You just set thousands of different parameters and tell the program to "try this" and it 'learns' as you accept or reject the output it generates, but it does not *understand* what it is doing. Neither do the programmers and experts. Like the flap about some app I heard about that was supposed to take the picture of a face and make it more beautiful or handsome - and it tended to turn darker skinned people lighter. Well, people were understandably upset about that - but it wasn't the program's fault or the programmers' fault. Unconscious biases introduced in the learning process or by users did that - there was no malicious intent, I don't think. Anyway, back to work on this month's final scene!

S Munson

I love how you describe the details of the mistakes!!! Good Halloween fodder. As far as displaying...I wouldn't worry about what we think. Because this is new and exciting for you I say just display what you want. A few turrets here and there mixed with your digital AI mens. What more could I ask for???

pratev

Maybe I'll post a rogue's gallery of some of the more spectacular failures!

S Munson

I agree! You must CELEBRATE your mistakes SPECTACULARLY!!! 😀😄😁