AI frames with Red Dwarf auras

[gerrydelasel]

Sirs, I have a little ongoing project, which is to create high-res images of Red Dwarf using a combination of Ai and photoshop. So I’ve created this thread to avoid tarnishing the other threads. Please, don’t hit me.

Topic

[gerrydelasel]

Here, your majesties

[Frank Smeghammer]

I’m going to be honest with you. I think people are initially critical of this because they don’t see the value of the project in the first place. Me included. If you want to upscale and polish up the original shots it needs to be almost flawless down to the very detail, otherwise it doesn’t beat the original shots which are realistic because they are…well real shots.

But I also think you aren’t doing yourself any favours suggesting that you are doing anything more than feeding an AI a prompt and maybe making some really tiny alterations (though I can’t see them). There is also not a lot of effort put into fixing the obvious AI errors.

If you were putting any sort of time and effort into this, you wouldn’t be churning out 3 or 4 of them a day, every day.

These are obviously AI generated images with your name slapped on them. It’s as plain as a Bulgarian pin up.

[gerrydelasel]

They’re not posted same day, I have a whole archive already.

why is Officer Crabtree having a conversation with Vin Diesel?

Why indeed, you’ll have to ask the producers.

[Frank Smeghammer]

I mean you fixed the sign above the bed but anywhere there is text is just rubbish. I know your point is that if it’s too far away then it is unreadable anyway.

Fine. But it doesn’t resemble any real language or alphanumeric characters. The closest any of it looks to real text is Rimmer’s certificates that look like they are in Greek.

This is observations made after a couple of minutes inspection. It would take hours and maybe days of work to fine tune it, but that is the price of doing something to a standard you can be proud of.

If you aren’t prepared to do that sort of work, fine. Me neither. Sounds like a ballache and not something I have the time, skill or inclination to do. But then call it what it is, which is AI slop.

Sorry to be mean, but I know lots of others are thinking this and not saying it. I don’t know anyone on this site personally, or even well virtually. That is to say, I’m not one of the “in gang” on this site. But even I could sense that everyone is thinking it and nobody saying it

[FileTrekker]

At the resolution you upscale to, all the text I pointed out should be readable, and the fact it just looks like jibberish, because it is jibberish, snaps you out of whatever it is you’re trying to achieve.

Also, that looks like Chris & Bobby because, it is Chris & Bobby. It’s lower resolution and the lighting washes out the Kryten mask a little so, at a glance, he looks like he has a smoother head than usual but there’s enough real detail there that they look like the characters.

Your AI version looks nothing like them.

It’s a bit weird you don’t see that.

[Warbodog]

everyone is thinking it and nobody saying it

We’re all Red Dwarf fans employing our time in profitless and non-practical ways, let them play around.

[gerrydelasel]

I’m so tired of boomers who have just learned the phrase ‘Ai slop’ and don’t understand what it means. Humans produce slop all the time and it no one bats an eye. AI can never be a true photograph obviously, but it’s improving every day and when all you have is poor quality media and no superior ‘master tapes’, AI does a damn fine job.

BTW, if anyone can tell me what Rimmer’s certificates are supposed to say, I’ll fix them.

[Frank Smeghammer]

I’m so tired of boomers who have just learned the phrase ‘Ai slop’ and don’t understand what it means. Humans produce slop all the time and it no one bats an eye. AI can never be a true photograph obviously, but it’s improving every day and when all you have is poor quality media and no superior ‘master tapes’, AI does a damn fine job.

Well I’ll leave it alone, I’ve said what I’ve said. One final note though, I will evaluate what I mean by the term “AI slop” if you think about what you think the term “Boomer” means. I was born in the 1990s. I think I might well be younger than you if I was to have a guess.

[gerrydelasel]

These days boomer is just a casual insult for anyone who isn’t ‘hip’ and ‘switched on’ and ‘down with the latest tech’. If you prefer, I will substitute crypto fascist.

[Unrumble]

Still beats the Series X mask

[gerrydelasel]

Holly on the monitor, and the multitude of other AI generated crimes in these images… Your AI version looks nothing like them. It’s a bit weird you don’t see that.

I think maybe we’re looking at this with different expectations. Take this image; it’s a model Starbug and a model human figure. So if I were to create an AI upscale, would you expect me to aim for it to look like a real human-scale Starbug and an actual human Chris/Rimmer, or like a higher-res photo of the model starbug and model figure?

[Frank Smeghammer]

.

[Frank Smeghammer]

Edit – said I’d leave it so I will

[gerrydelasel]

Don’t go!

[Podey]

I’m sorry if we’re not impressed enough that you got a computer to waste an unreasonable amount of natural resources to up-scale an image you didn’t create, using AI trained on the work of unwilling artists and photographers and then fixed all the bits it got wrong. 

The issue for me is it just doesn’t feel “transformative”, but maybe I am underestimating exactly how much work you’re putting into the edits – maybe you could share a comparison of what AI spat out vs your final version, purely out of interest?

For what it’s worth, I don’t *entirely* agree with the people who say there is absolutely no skill in being an AI prompter. I couldn’t open one up right now and create generative “art” from scratch without some kind of learning curve, which implies there is skill involved. I don’t believe that is the same skill as is involved with being an artist – I see it as more like coding – but I’d admit to finding, say, a shot of Starbug built entirely from the ground up vaguely interesting, even if the tech is currently putting me and many other artists I know out of a career. Nor do I deny there is tonnes that’s absolutely fascinating about AI and it’s uses.

[International Debris]

For what it’s worth, I don’t think you’re going to find many – or any – people here who think that the various issues around AI – be they natural resources, theft, or the fact that most of the AI companies financially support fascists – are offset by the quality of the work. Most of this stuff is really not much more interesting than publicity shots and stuff that we’ve seen. Sharper? Yes. The occasional angle that makes it look more interesting? Yes. But actually worth all the compromises involved in supporting an appallingly damaging and corrupt form of tech? Very, very unlikely. 

Not starting an argument with this, but I think it might be worth just rescaling your expectations in terms of the kinds of response you’re hoping to get. 

[Jimboid]

Well, I like them.

[Flap Jack]

I have mixed at best feelings about AI overall, but this use seems more or less fine, especially as it’s siloed to its own dedicated thread. Gerry doesn’t seem to be selling these images as anything other than they are. Hopefully they don’t end up getting spread around as actual HD set shots…

They’re probably being a bit too ambitious in trying to recreate full model shots though. I mean you need 25 sequential images just to get a second of footage.

[Jenuall]

I don’t believe that is the same skill as is involved with being an artist – I see it as more like coding 

I do get where you’re coming from, but speaking as someone from the software development industry I’d just like to clarify that the architecture, design, and development of good software requires a hell of a lot more knowledge, skill and, dare I say it, creativity than prompting AI does!

[Podey]

Apologies, I was thinking specifically in terms of basic website coding (my own experience being limited to HTML… !). 

[Frank Lewis]

Yeah, but at least in yours, the light above Lister’s bunk didn’t read “IFC TOGA ANRONT []”

Gonna be slightly cunty and point out how incredibly easy this would be to fix with Photoshop or GIMP or whatever.

Still funny, innit’.

[Technopeasant]

I don’t believe that is the same skill as is involved with being an artist – I see it as more like coding 
I do get where you’re coming from, but speaking as someone from the software development industry I’d just like to clarify that the architecture, design, and development of good software requires a hell of a lot more knowledge, skill and, dare I say it, creativity than prompting AI does!

If AI does anything it sure makes people argue over what is or isn’t actual talent or worthwhile work.

[gerrydelasel]

[gerrydelasel]

[gerrydelasel]

Something spooky happened last night. I asked AI for an image of the officers’ quarters and I was happily tweaking it when I noticed something. Check out the mug on the table, it has the Red dwarf logo on it, which was not in the original image and was not added by me. Ignoring my edits, this image was generated in a new chat thread where I had never mentioned I was uploading images from a TV show, I had never used the words ‘Red Dwarf’ or any other noun specific to the show, and I have absolutely never referenced the Red Dwarf logo, ever. Not only has it somehow recognised the room, it has also pulled the logo from somewhere. Is AI self aware? Is it a fan of the show? This is giving me the heebie jeebies. (Original screenshot for comparison)

[Flap Jack]

Presumably the AI just has images from Red Dwarf in its dataset and so did associate the image of the officer’s quarters with the show. Totally within the expected behaviour of these things.

If the mug, say, had the name of a departed relative you never mentioned, above the words “They’re in hell now, and you’re next”, then you could be concerned.

[gerrydelasel]

Hmm, it is exceptionally unlikely to recognise a set from an obscure, 20+ year old British TV show without key words in the prompt, let alone to pull the logo and apply it to an object without being instructed. Not expected behaviour at all. I suspect a rift in the space time continuum.

[Flap Jack]

Red Dwarf is not obscure.

[gerrydelasel]

[Dave]

It’s true, only a handful of people in the world possess any actual images of Red Dwarf, so for AI to be able to identify the show based on a photo of one of the main sets is unthinkable.

[Podey]

Newer models are able to remember your previous chats and apply what it knows about you across new chat threads – I would guess it has identified from previous interactions that you’re upscaling images from Red Dwarf and so is using that as an unspoken parameter for new requests.

You could try asking it to reset your conversations and forget all previous data and see if throws up something like that again.

[Dave]

Either that or start upscaling images from The Good Life and see if all the mugs have Red Dwarf logos on them.

[Unrumble]

[Jonathan Capps]

It’s a shame that we’ve provided a pretty comprehensive set of LLM training data about Red Dwarf with the SmegaDrive. 

[gerrydelasel]

[Author]

Replies