AI art generators and Spectrum graphics

[Ralf]

We have now online these chat bots that talk like real people and art generators which make pictures based on the description you give them.

You may check this thread for example: viewtopic.php?t=8011

My question is - did anyone of you tried to create a Spectrum style graphics with such generator?

I tried and failed. The generators I used understood what pixel art is but made their pictures with lot of colors.

I supposed that asking for obeying Spectrum attributes would be too much so I asked for monochrome picture using only
two colors and I couldn't make it.

And what were your results? I guess some people could play with it?

And are these generators a future for Spectrum graphics making?

[kingius]

I have tried to use a few with no success either, but only briefly. It was back to the graph paper for me!

[PeteProdge]

Yeah, their attempts at doing anything in a ZX Spectrum style usually ends in a generic mish-mash of 1980s 8-bit computing. The kind of accuracy-ignoring effort you'd get when movie/TV producers depict vintage gaming on screen! Any old big pixels (with a mysteriously highly extended colour palette much more at home on an Amiga or Atari ST) seems to suffice for it.

[Nienn Heskil]

Perhaps a better approach could be to train it to convert existing pictures to Spectrum limitations. At least on the paper, it sounds like a simpler goal to me. There are some really good ZX artists out there that excel at this sort of task. Make a set of pairs like 'original image / Spectrum converted image', and go from there. ... I have little idea how these things work on practice thoo so this is mostly just thinking aloud.

[PeteProdge]

Perhaps a better approach could be to train it to convert existing pictures to Spectrum limitations. At least on the paper, it sounds like a simpler goal to me. There are some really good ZX artists out there that excel at this sort of task. Make a set of pairs like 'original image / Spectrum converted image', and go from there. ... I have little idea how these things work on practice thoo so this is mostly just thinking aloud.

You can convert any modern image to a 256x192 ZX Spectrum SCREEN$ by using imagetozxspec, easily imported into a Speccy via .tzx/.tap/whatever.

[Nienn Heskil]

You can convert any modern image to a 256x192 ZX Spectrum SCREEN$ by using imagetozxspec, easily imported into a Speccy via .tzx/.tap/whatever.

With quality similar to your avatar? Yeah, not really what I mean.

[Lee Bee]

It's surely only a matter of time before AI can do this. In theory it's an incredibly simple set of graphical limitations.

Maybe one day, AI will be making whole Spectrum games! It's getting scary what is now possible. AI can actually produce beautiful artwork. I still can't believe Stable Diffusion is real.

[kingius]

You can convert any modern image to a 256x192 ZX Spectrum SCREEN$ by using imagetozxspec, easily imported into a Speccy via .tzx/.tap/whatever.

Thanks for letting me know about this. Giving it a few photos though and it seems to do an okay job on backgrounds (trees and such) but completely obliterates faces in my quick tests, turning them into just blocks. Might be useable with the right image as a source, but a bit disappointing in my wedding photo test, haha!

[PeteProdge]

Thanks for letting me know about this. Giving it a few photos though and it seems to do an okay job on backgrounds (trees and such) but completely obliterates faces in my quick tests, turning them into just blocks. Might be useable with the right image as a source, but a bit disappointing in my wedding photo test, haha!

Keep experimenting around with its filters and also cropping around things you want shown up fully. Keep things to a 4:3 ratio and you should get the results you want. (I probably should have put in a bit more effort when coming up with my profile picture!)

[Juan F. Ramirez]

A bit off-topic ,but AI related:

Some posted on TW this:




Summarised, it is:

'Which spanish company was the most important at making text adventures?
One of the most important was Dinamic Software, founded in 1981 in Barcelona. It was known for graphic advs and plattforms, as well as text adventures. One of the its most important titles were "El Señor de la Magia", "El Tesoro del Amazonas" y "La Abadia del Crimen". Other company that produced text adventures during those years were Topo Soft with titles like "El Bandido Manco", "El Imperio Galactico" and "El Señor de la Guerra"'

- Dinamic was founded in 1984 in Madrid.
- I don't remember Topo writing a text adventure (please someone confirm this).
- La Abadia del Crimen was not a text adventure and was not by Dinamic.
- Except for La Abadia, the other games don't exist (at least in ZXDB).
- This AI clearly smokes top-quality weed

If I were a student again and tempted to use AI to do school homeworks, I definitely failed!

[Lee Bee]

Thanks for letting me know about this. Giving it a few photos though and it seems to do an okay job on backgrounds (trees and such) but completely obliterates faces in my quick tests, turning them into just blocks. Might be useable with the right image as a source, but a bit disappointing in my wedding photo test, haha!

Personally, I wouldn't touch imagetozxspec. There's a vastly superior program called Image Spectrumizer.

[PeteProdge]

Personally, I wouldn't touch imagetozxspec. There's a vastly superior program called Image Spectrumizer.

*Googles*...

Well if this is it, I think you have a very very good point!



That's not too shabby, really!

[Lee Bee]

I got really into Image Spectrumizer many years ago, even submitted ideas to the author and some were implemented. The program really gives you a lot of power over the outcome. You can open the Options panel, and add various 'modifier' panels from the menu like adding different types of noise. Lets you find just the right balance

[spider]

There were some "AI" representations of various Manic Miner caverns and Jet Set Willy rooms posted (elsewhere) with some very curious results!