That is interesting!
A bit like how teenagers now think that TV in the 90’s looked like it does on the YouTube rips which are based on MP4 compression in the 2000’s.
Some of its points of reference will be interesting, untraditional. It’s really interesting the things it doesn’t understand too, that you think it would - sometimes connections can seem obvious but aren’t.
My So Called Life looks pretty grainy in my memory too. Most days it feels like my mind is a stuck VHS head.
I don’t see any stretched faces? Where do you get this information from anyway?
I have eyes. Bye, random internet person.
Actually I love that we get all those 1080p and 4K rereleases of old/older movies, because that’s how I remember them, not from substandard VHS tapes and bad TV reception 
Just block me. I don’t want to talk to a person who uses the word “mansplaining” on women.
n/a
It’s an inevitability based on the search material - you find when using the AI’s that other such quirks crop up - like ‘signatures’ appearing in the corner of the image. It’s not that it’s lifting or copying a signature from anywhere etc. - but it’s made a connection of sorts somewhere on its journey and thinks that little line squiggle is inherent to the context it’s working with.
So you could imagine with all the poorly ripped 4:3 movies and captures that exist that there could become a sort of association in the AI’s dataset with distorted proportions. From its own perspective that’s an aesthetic quality of that data. One day it’ll be smart enough to do as we do and say, ‘oh, that’s not how people look, it’s stretched a bit’, but currently those wrinkles are still being worked out.
That said, it’s also doing its job well. If you want a convincing screenshot of an 80’s movie then that’s a quality that might make it look more convincing. Little flaws like that can be powerful in making things seem real. I used to work with CG and a lot of photorealism is introducing imperfections.
I think this is the point, since I can’t see any distortion on the faces and we have really no evidence that early 2000’s pictures with distortion or poor quality have influenced those final images.
But this aesthetic could also very well be fully intentional and obtained as part of the process? I mean I’m sure the AI is capable of creating Dune images that are wholly modern, futuristic and not vintage at all?
For those who are interested, I took a few images and corrected them by eye (right side). It does not affact all of them and some are mixed, like sometimes faces are partially stretched and partially correct.
Note how on the last image, the face in the foreground is now correct but the circle and the person in the background is now wrong.
n/a
Those are my fav so far! Care to give some details about the system used and the prompts?
These were from Midjourney and the top two I was trying to get it to interpret the idea of sound being rendered as light
Thanks a lot! Did you use the beta upscale to reach 2048*2048px? Also did you mention “black and white” somewhere to produce a monochrome?
This is maybe the last one I did that I like. Will probably end up using it as cover art.
Love it!
Yes this was using beta upscale I think (it doesn’t say in the archive) - I included some other attributes like minimalism, contrast etc. I was actually just asking for ‘white’ not black and white.
Still not there but good stuff tho! 
interesting textures!
Yeah it’s kind of weird I’ve had the little sphere on a pile of rocks several times without even mentioning it at times. This is what I had just for the word “Persistence”:
It’s funny to see the AI come up with reoccurring patterns.
Also that one was so derivative it even included a watermark lol.
