Using A.I to recreate ZX Spectrum loading screens.

[MustardTiger]

A new feature was added to Bing image search recently that lets you use A.I. to create images from a text prompt.

https://www.bing.com/images/create

I asked it to create the Atic Atac screen and this was one of the best results. It took a few changes to the prompt to get it right but I think it did a great job.



The prompt was "a thick iron grate in a cobbled dark dungeon floor is lifted up, a green goblin creature with glowing eyes peering out of the darkness claws digging into the stone floor. in the background a wizard in a purple hooded robe, hidden face but yellow eyes looks at the camera and casts a spell. A sign above reads "ATIC ATAC" draped with chains. dark and gloomy classic style"

It'd be interesting to see other peoples attempts at classic loading screens.

[SteveSmith]

I seem to remember someone doing a few of these and posting them on Twitter recently. Might have been Gary Arnott? (Although, to clarify, Gary does a lot of amazing images by hand).

[bluespikey]

Theres probably a new forum game in here somewhere.

[ParadigmShifter]

Does it take forever to come up with an image or is my internet really slow?

I guess using SJOE as one of the words is confusing it a bit

EDIT: I guess it's really slow... 15 minutes and counting. I can use a boost if I spend some tokens, I can wait

Curious to see what it comes up with though

I typed:

zx spectrum loading screen for a word based columns style game called SJOE

It said

Please wait. Your images are currently in progress.
Check back later to see if your images are ready.

[MustardTiger]

Does it take forever to come up with an image or is my internet really slow?

I guess using SJOE as one of the words is confusing it a bit

It should take around a minute. Just to the right of the input prompt there's a yellow icon. These are boosts, basically a number of goes that will generate quickly, after you've used those up it can take longer. You should have 100 per day. If this is greyed out then an image will take longer to generate. I think it's a throttling mechanism to stop the servers getting overloaded.

However, it does get quite busy once the USA gets out of bed, so bear that in mind. Also, some images can get stuck and never finish, so clear the prompt and you should see the yellow boost icon and enter a new search.

[ParadigmShifter]

Yeah I think it hung cos I had to do a 2 step authentificationicationmajig

Not very impressed

[/url]

Bottom left has the right idea but it's not what I am looking for. I do like that some of the bystanders have bicycles though lol

EDIT: I see it has made the google mistake of "Do you mean SHOE?" though

EDIT: This was even worse

zx spectrum loading screen for a psychic esp woo woo game involving zener cards, like columns

Not even going to show what it came up with lol. It did know about "cards" and psychic woo woo stuff like fortune tellers though.

I guess I'm going to have to draw my own loading screens then Programmer art

[Mpk]

prompt :
A fat green worm reclining on a DIMM chip raising it's smug pink-faced head to look directly at the viewer. A white glowing orb appearing from its tail. The whole scene is presented through the lens of a large magnifying glass. Cartoon style

[MustardTiger]

Yeah I think it hung cos I had to do a 2 step authentificationicationmajig

Not very impressed

[/url]

Bottom left has the right idea but it's not what I am looking for. I do like that some of the bystanders have bicycles though lol

EDIT: I see it has made the google mistake of "Do you mean SHOE?" though

EDIT: This was even worse

zx spectrum loading screen for a psychic esp woo woo game involving zener cards, like columns

Not even going to show what it came up with lol. It did know about "cards" and psychic woo woo stuff like fortune tellers though.

I guess I'm going to have to draw my own loading screens then Programmer art

I'm not sure it could create an actual spectrum loading screen, it probably hasn't been trained for that. You can be very descriptive about what you want and tell it what style to use, eg pencil drawing, or 8-bit style and monochrome.

[stupidget]

Anyone care to guess what loading screen this is a recreation of?:

[ParadigmShifter]

Well it's Knight Lore innit.

It's definitely not Stonkers anyway.

EDIT: I wasn't expecting it to come up with a loading screen I was just looking to see if it would do something in the zx spectrum graphics style (not really) and understand the type of game it is (1 out of 4 ain't bad I suppose).

It's quite impressive anyway but as you said it needs training on more data (it currently trains on the internet so if nothing matches a search you're going to be out of luck most of the time I think).

I think what it does is:

Look up on the internet for images that match description or has keywords etc.

Bang that into a complicated neural net that tries to draw something similar to what it found (that's the difficult part).

I only really know how chatGPT works tbh

[stupidget]

LOTR:



This is what I used to create the image:

face on a faceless hooded figure in black rides a horse holding a sword aloft , it's cape billowing in the wind in the background there are rolling green hills and fields

[AndyC]

Not bad for a quick attempt I think

Prompt:
egg wearing boxing gloves standing on a mountain made of crystals, facing away from the camera, overlooking a distant castle in cartoon style

[ParadigmShifter]

Close but no cigar

[bluespikey]

Can't get it to draw a tank with no turret.

[Mpk]

First attempt, from memory




A tank in green camo pattern advances across a map of europe, travelling west to east. Camera is low to the map, fish eye lens

[ParadigmShifter]

It should say "AI en route with pictures" if you type Stonkers in.

[MustardTiger]

I recognise it but can't place it. Thought it might be the Bug Byte logo.

[Lee Bee]

Since AI can now emulate the styles of various artists and media, I'd love to see AI's attempt to create images in the general style of Spectrum games. Or whether AI could create actual Speccy graphics that are any good.

[R-Tape]

I recognise it but can't place it. Thought it might be the Bug Byte logo.

I thought Fatworm blows a Sparky, but your guess fits the bill better.

[ParadigmShifter]

It's definitely Fat Worm Blows A Sparky cos he seems to be lying on a silicon chip And no electricians in sight

[Lee Bee]

Not bad for a quick attempt I think

This stuff is insane. Last year, an image like this would be the result of some top artist spending months of hard work. Now it's a "quick attempt" on an AI prompt. Awesome and terrifying how quickly the world of art and illustration has been shattered by this technology.

[Joefish]

Can people post the prompts they used, too? It's interesting with these things to feed it the same prompt from another machine on a different day and see what comes out.

[MustardTiger]

I thought Fatworm blows a Sparky, but your guess fits the bill better.

Yeah, definitely Fat Worm. Good spot

[ParadigmShifter]

It's basically just plagiarising stuff which is already out there though.

It's very interesting how they work - search for "Generated Adversarial Neural Networks" to get started.

(VERY) Basically AI #1 draws a picture and AI #2 (trained on pics of what it was trying to draw) tells it how well it did. Repeat until it did quite well!

EDIT: Plus you know where it gets the training for what stuff looks like from...

"Click on all pictures containing a traffic light"

[AndyC]

Can people post the prompts they used, too? It's interesting with these things to feed it the same prompt from another machine on a different day and see what comes out.

Good point, I've edited mine to include the prompt. Also gives people a chance to refine them.