repAInt, iconic images recreated by AI

[+Philsan]

@pseudografx presented an awesome video at Lost Party 2023 wild competition.

 

It shows some iconic A8 images recreated by AI (perhaps two images are NSFW).

 

 

[+gnusto]

@pseudografx I'd love to see your prompts (and models used) here if you don't mind supplying. The whole thing is quite entertaining, thank you.

 

Also I couldn't resist the idea so I have a RastaConverter running now on taking your Prince of Persia image and presenting it on the 8bit. So we can go full circle =). It has a ways to go yet but I'll add a post here once it's done.

[Beeblebrox]

29 minutes ago, gnusto said:

@pseudografx I'd love to see your prompts (and models used) here if you don't mind supplying. The whole thing is quite entertaining, thank you.

 

Also I couldn't resist the idea so I have a RastaConverter running now on taking your Prince of Persia image and presenting it on the 8bit. So we can go full circle =). It has a ways to go yet but I'll add a post here once it's done.

Heh heh, you beat me to it. I was planning the same.

[+Stephen]

What a neat idea - that turned out really well.

[pseudografx]

Thanks for the post. I've uploaded my video to YouTube:

 

[+MrFish]

It wasn't smart enough to pick up on the black cat perched on the dark-alley wall. Yet, hardly a human would miss that, given a short time to examine the original. Personally, I saw the cat immediately, and was guessing the results would be as such (AI not picking up on it).

 

[pseudografx]

Yes, I might have preserved the cat if I focused on it but there were so many things to keep in the image and achieving a perfect result would require a lot more time. Some of the generated images were done in a couple of minutes, others took half a day.

Edited July 10, 2023 by pseudografx

[+MrFish]

Oh, ok. I'm not aware of how much input is required from the user side. I've never monkeyed with such toys yet.

 

[MrTrust]

The way it draws the female figures makes you think it's training data was nothing but hentai.

 

What were the prompts that you used?

[pseudografx]

There were too many. I can post them if you ask for a particular one 🙂

[vitoco]

So it really was a recreation... I'd love to see the propts from any of those which have texts.

 

(I also missed the black cat in the alley)

 

[pseudografx]

I had to touch up many of the images that contain text using "inpaint" feature as otherwise AI likes to ruin the texts with nonsense. I even did not bother to fix all of them and instead focused on the imagery and the most prominent texts. In some cases I removed the texts from the original images.

All text could be preserved but that would really be a tedious task. Not to mention that I'm still only learning what can be done.

[+Philsan]

Now you have to recreate all Strip Poker images, including data disks 😂

[twh/f2]

Whoa that had an impact. Great work PG and Poison.

 

It's bit like browsing in a photo album from times long long ago. It felt a bit like a school reunion and seeing all those people again.

What really got me was this one:

 

 

I guess it's the combination of feelings: me as little boy playing Boulderdash and now my little boys who also play Boulderdash.

 

and of course the girls look amazing... but that's another emotional part *g

 

grüße,

\thomas

 

Edited July 10, 2023 by twh/f2

[pseudografx]

5 hours ago, twh/f2 said:

Whoa that had an impact. Great work PG and Poison.

 

Thanks! What does Poison have in common with that, though?

[+Allan]

That was really cool. I can think of some game screens that might look great with this. All four Lucasfilm games come to mind. Star Raiders, Seven Cities of Gold, screens from the Adventure International SAGA games.

[MrTrust]

14 hours ago, pseudografx said:

There were too many. I can post them if you ask for a particular one 🙂

 

Well, just a couple of examples, maybe.  I guess getting useable results out of these AIs requires good prompts, but I can't seem to make one myself.

[Beeblebrox]

Blinky's scary school would be an interesting one.

 

 

Or spacce lobsters:

 

 

 

[Faicuai]

"Lovely" fakeness.

 

That's the world coming, where the joy of a "livable" fiction subdues the power of reason and critical thinking.

 

Just imagine the kind of world inhabited by those born after all of this technology is mainstream... sort of like the same reality of those born with Facebook, Instagram, Twitter mainstream vs. those that never had that when growing up...

 

Pathethic...

 

[+gnusto]

It's getting there! Still has a couple of billion passes until it's done 😃 (This is PAL image/palette)

 

Now I'm wondering if we should re-do Artworx Strip Poker with RastaConverter images...for science.

output.xex

[+CharlieChaplin]

3 hours ago, gnusto said:

It's getting there! Still has a couple of billion passes until it's done 😃 (This is PAL image/palette)

 

Now I'm wondering if we should re-do Artworx Strip Poker with RastaConverter images...for science.

output.xex 19.83 kB · 4 downloads

 

And if the Rasta pictures look similar to this, one should name it Stripe Poker... 

[+Stephen]

22 minutes ago, CharlieChaplin said:

 

And if the Rasta pictures look similar to this, one should name it Stripe Poker... 

Yeah - sometimes not even a trillion iterations are enough unfortunately.

[Beeblebrox]

How about this for full circle.

 

This is the original Atari artwork (G2F presumably(?)) from which the AI image in the YT video was generated, (hereafter known as Taxilady):

 

I've taken a snapshot of the AI generated image taken the YT vid here and preprocessed it to make a source image for Rasta converter:

......

 

...then I've been running the PAL RastaC conversion.

 

Still running it but it's coming out really well

 

Here is an output XEX, (PAL pallette) where so far definition and colours are looking good, (where I am particularly pleased with the details on the taxi and those on her face):

(snapshot taken from Altirra with PAL high artifacting setting in place):

Taxiladysm1_05.xex

 

This may or may not be the case but given the original Atari G2F image was of course laid out in such a way for the Atari to display, because the AI image is based on it, it stands to reason this potentially provides an ideal source image for Rastaconverter to create a decent conversion.  Proper full circle stuff here.

Edited July 11, 2023 by Beeblebrox

[twh/f2]

21 hours ago, pseudografx said:

Thanks! What does Poison have in common with that, though?

Haha .. I'm sorry. I misread the credits. So this time NO credits to Poision .. *g

[pseudografx]

As for the prompts, here is a couple of examples. Note that these prompts may not have been applied to the original images. In many cases I have generated an output image and then used that image to generate further iterations.

Positive prompt: Beautiful girl lying on her side, blonde, long hair, lilac dress, shorts, european, green eyes, RAW photo, cleavage, busty
Negative prompt: (easynegative), (badhandv4), bad quality, worst quality, lowres, illustration, painting sketch, ((asian))
Additional parameters: Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 3872395887, Size: 1280x576, Model: laura_v11, Denoising strength: 0.24, Clip skip: 2

 

Positive prompt: RAW photo, beautiful girl, long hair, black hair, ((green bra)), busty, cleavage, muscular man, turban, white shirt, brown vest, holding sabre, blue silhouette of a Persian palace, Prince of ((Persia)), best quality, high resolution, masterpiece
Negative prompt: (easynegative), (badhandv4), bad quality, worst quality, lowres, illustration, painting sketch
Additional parameters: Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 348494506, Size: 1280x864, Model: perfectWorld_v3Baked, Denoising strength: 0.24, Clip skip: 2

Edited July 11, 2023 by pseudografx