can AI really get this so correct

Diddums said:
Dunno what you're talking about. These are two perfectly normal looking women.
Click to expand...

I thought that one had been photoshopped rather being AI generated, but looking at the hand on the steering wheel and the hair of the one on the right, I think it is AI.
 
arknor said:
you could train your own model that can create pictures from words, have it process an audiobook and put a video output, even if it were only still images.
you then put that on youtube or somewhere as "Audiobook graphics Dungeon Crawler Carl"
people listen to their audio book in sync with the video of ai created images.

I'm surprised no one seems to be doing it on a professional level, maybe theres some copyright laws stopping it?
Click to expand...

Yeah, I think you'd run into copyright issues with current works, like you can't just grab a copy of say Game of Thrones and make your own video with audio + AI images but I have seen people upload audio + images of stuff that is out of copyright along with AI-generated art.

There are other issues re: the consistency of the images and creative control too - current stuff still looks like a patchwork of now familiar AI-slop images but I'm sure some startups out there would create such a product in future.

Seems like a complete no-brainer for Amazon to create some AI features in house and add it in as functionality for a Kindle Fire etc.
 
Last edited:
arknor said:
you could train your own model that can create pictures from words, have it process an audiobook and put a video output, even if it were only still images.
you then put that on youtube or somewhere as "Audiobook graphics Dungeon Crawler Carl"
people listen to their audio book in sync with the video of ai created images.

I'm surprised no one seems to be doing it on a professional level, maybe theres some copyright laws stopping it?
Click to expand...
You can’t copyright AI works. So you could do all that work and other people could repost it, make money from it and you would have no legal recourse.
 
Last edited:
AI is ****, it's just some marketing ******** to get you to buy shares in Nvidia.

When AI can come up with a reason why my Mrs is in a strop with me when I've done nothing wrong, I'll buy into it, untill then......
 
Stand up arcade machines did not look like that in the 80's, I'm old enough to have spent an unhealthy amount of time in them
The screen would have been almost 90 degrees ie facing the player, with just the controls at hand level.
The box at the top was usually just a lightbox with the games name on it, or occasionally some naff speakers.
 
Almost all of them have the hand of AI on them (pun intended).

#1 AI. Text is garbage wherever it appears and the pic is clearly stretched from 4:3 to widescreen.
#2 AI. Look at her left hand and pay attention to the knuckles.
#3 AI. Hand on steering wheel and again a stretched image. Also, are they in the process of spinning off the highway? If so, why so happy :)
#4 AI. Hands again. Also lack of brake levers on girl's bike.
#5 AI. Inconsistent text and depth of field, again seems stretched.
#6 Probably AI, grip seems wrong.
#7 AI. Woman's hand on seat is wrong. Guy in background has no nose.
#8 AI. Arm and hand under book has weird foreshortening.
#9 Probably AI, also seems stretched.
#10 AI. Inconsistent motion blur and shadows (some trees have, some don't). Car bonnet looks deformed.
#11 Probably AI, maybe just a rubbish Photoshop (note the seam in the background in line with model's arm).

I'd suggest most of these are the result of feeding screengrabs of 80s TV programmes into generative AI as supplemental reference images (if Midjourney) or possibly a specific training set (if SD).

These have either been stretched from their original 4:3 format to widescreen before feeding into the AI, or the AI has done this. The general blurryness in all of them might be an AI style or postwork in Photoshop.

I squished some of the images back to 4:3 in Photoshop and they looked much more believable (all the AI weirdness aside).
 
BUDFORCE said:
When AI can come up with a reason why my Mrs is in a strop with me when I've done nothing wrong, I'll buy into it, untill then......
Click to expand...

They'll solve human-to-dolphin communication and cure cancer before they create an AI model to solve "why is the missus mad at me?"
 
Chuk_Chuk said:
You can’t copyright AI works. So you could do all that work and other people could repost it, make money from it and you would have no legal recourse.
Click to expand...

True but that's more the case where there is no human input beyond prompting, if you're mixing it with IP then that changes things - like if you had the rights to some fantasy/dungeon crawler story (or wrote one yourself) for example and you were making an audiobook + pictures YouTube video as he suggested then other people can't just rip that video as they don't have rights to the story - they could grab the still artworks though *if* those were only made by prompting.

But then there's another aspect perhaps - if you were doing it properly then you'd want consistent characters and so would be training a LoRA model for those - maybe you hand draw the initial characters or get someone on fiver to do it for you... train the LoRA and then create a load of the AI artwork using those characters - can someone copy and upload that artwork now? Doubtful, those are your characters/IP - likewise if Disney made some Mickey Mouse artwork with AI the fact it's AI is perhaps secondary to the fact that Mickey Mouse is still their IP regardless.
 
mattyg said:
This isn't AI though

boingboing.net

Everything about this “Beautiful Woman Soldering” stock photo Is wrong

Over at MAKE, Mike Senese comments on this stock photo of a "Beautiful Woman Soldering." It's an unintentional "how many things are wrong in this picture?" puzzle. Rather than holding…
boingboing.net boingboing.net
Click to expand...

It's stuff like this that is part of the reason that generative AI makes the mistakes it does. The generative AI models were primarily fed whatever they could trawl from stock libraries (largely watermarked low res previews and an abundance of clichés), art sharing sites (huge variance in quality and of course copyright infringements aplenty), social media platforms (full of Photoshop fakes and cat pics) and so on. Even before the internet filled up with AI slop, they was already a vast sea of slop out there, much of it poorly tagged or labelled. All of which was used to train the AI models. Where some quality control and guidance was applied in the training process results are better, but there's still a lot of baseline slop.
 
Cern said:
It's stuff like this that is part of the reason that generative AI makes the mistakes it does. The generative AI models were primarily fed whatever they could trawl from stock libraries (largely watermarked low res previews and an abundance of clichés), art sharing sites (huge variance in quality and of course copyright infringements aplenty), social media platforms (full of Photoshop fakes and cat pics) and so on. Even before the internet filled up with AI slop, they was already a vast sea of slop out there, much of it poorly tagged or labelled. All of which was used to train the AI models. Where some quality control and guidance was applied in the training process results are better, but there's still a lot of baseline slop.
Click to expand...
Garbage in, Garbage out
 
AI still struggles hysterically with hands and continuity in scenes, and takes massive hardware resources to produce longer continuous video, but some of the AI produced fantasy videos people have been cooking up are crazy especially the stuff based off Lord Of The Rings, etc.
 
You must log in or register to reply here.
Back
Top Bottom