VOGONS

Common searches


Reply 20 of 102, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
Shponglefan wrote on 2024-09-30, 13:32:

These images highlight a lot of the limitations of current iteration of AI generative art. It's basically just copying surface level details without foundational understanding of geometry, perspective or form. Hence, some of the shapes of the 'computers' look half-melted and distorted.

All the images show understanding of geometry, perspective and form. Ask a human to paint photorealism alla prima and you get the same effects.

Reply 21 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t
vvbee wrote on 2024-10-16, 06:37:
Shponglefan wrote on 2024-09-30, 13:32:

These images highlight a lot of the limitations of current iteration of AI generative art. It's basically just copying surface level details without foundational understanding of geometry, perspective or form. Hence, some of the shapes of the 'computers' look half-melted and distorted.

All the images show understanding of geometry, perspective and form. Ask a human to paint photorealism alla prima and you get the same effects.

I'm afraid I can't agree. In photorealism, artists are capable of extremely tight geometry and sharp lines even in traditional media. See examples here, here, and here.

In contrast, examine the second image OP posted. The computer keyboards and what I think is supposed to be a tower (?) to the right of the monitor are clearly geometrically distorted. The keys in particular are highly distorted compared to the underlying objects they are attached to. And the "tower" (?) has a strange overall geometry and I'm not really sure what it's supposed to be.

When it comes to alla prima paintings or sketches, I wouldn't expect super precise geometry or lines. That's typically because the artist is working quickly to capture an overall composition or mood, in comparison to a detailed rendering. But even that will have consistency to it.

Again, using the second image as contrast, just compare the sharp lines of the picture frame or monitor versus the distorted lines of the window blinds or keys. Again, these are things that an artist would render more consistently.

I'll admit the AI images do have a certain surrealistic quality to them. But they don't strike me as either alla prima nor photorealistic.

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards

Reply 22 of 102, by kingcake

User metadata
Rank Oldbie
Rank
Oldbie
doshea wrote on 2024-09-28, 04:48:

It certainly made up some interesting things (is that an Apple computer/copier/fax/scanner?) but they all look melted and broken like something someone found at the bottom of a pile in a barn and cleaned extensively despite being completely broken.

You guys have to remember that what silicon valley wants us to think is real AI is just regression on steroids with massive data sets. Neural nets are just regression models. The estimates that regression models produce are averages. So you end up with a weird, averaged blob of a computer based on what is in the training set(s).

Reply 23 of 102, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
Shponglefan wrote on 2024-10-16, 11:26:
I'm afraid I can't agree. In photorealism, artists are capable of extremely tight geometry and sharp lines even in traditional […]
Show full quote
vvbee wrote on 2024-10-16, 06:37:
Shponglefan wrote on 2024-09-30, 13:32:

These images highlight a lot of the limitations of current iteration of AI generative art. It's basically just copying surface level details without foundational understanding of geometry, perspective or form. Hence, some of the shapes of the 'computers' look half-melted and distorted.

All the images show understanding of geometry, perspective and form. Ask a human to paint photorealism alla prima and you get the same effects.

I'm afraid I can't agree. In photorealism, artists are capable of extremely tight geometry and sharp lines even in traditional media. See examples here, here, and here.

In contrast, examine the second image OP posted. The computer keyboards and what I think is supposed to be a tower (?) to the right of the monitor are clearly geometrically distorted. The keys in particular are highly distorted compared to the underlying objects they are attached to. And the "tower" (?) has a strange overall geometry and I'm not really sure what it's supposed to be.

When it comes to alla prima paintings or sketches, I wouldn't expect super precise geometry or lines. That's typically because the artist is working quickly to capture an overall composition or mood, in comparison to a detailed rendering. But even that will have consistency to it.

Again, using the second image as contrast, just compare the sharp lines of the picture frame or monitor versus the distorted lines of the window blinds or keys. Again, these are things that an artist would render more consistently.

I'll admit the AI images do have a certain surrealistic quality to them. But they don't strike me as either alla prima nor photorealistic.

Your post sounds AI generated to be very honest, but that's the world we live in now. The easiest way to get up to speed on current state of the AI art is https://www.midjourney.com/explore, probably skewing toward better case scenarios. Better yet, set one up locally and generate various images to see where the strengths and weaknesses are.

Reply 24 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t
vvbee wrote on 2024-10-16, 16:01:

Your post sounds AI generated to be very honest, but that's the world we live in now. The easiest way to get up to speed on current state of the AI art is https://www.midjourney.com/explore, probably skewing toward better case scenarios. Better yet, set one up locally and generate various images to see where the strengths and weaknesses are.

I don't use AI generators. I'm of the old school internet discussion/debate, where I do my own searches, compile information myself, and type out responses by hand. 😀

(Or is that something an AI would say. 😉 )

Any any rate, as an amateur artist I have kept a close eye on AI artwork and the developments in that regard. I do occasionally peruse the Midjourney subreddit (among others) to see what sorts of things are being generated and what sorts of outputs it can produce.

This is why I'm highly critical of AI art generation because it still has a lot of problems in my view. The fact that AI doesn't understand the geometry of the scenes it is generating results in fundamental errors in perspective and form, and especially lighting. AI art is often rife with non-distinct light sources or other weird lighting issues.

AI art still a long way to go before it can ever replace real artists in my view.

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards

Reply 27 of 102, by Jo22

User metadata
Rank l33t++
Rank
l33t++

They aren’t true A.I.s, but rather *just* Large Language Models (LLMs).
I'm sure most here know this already, but just in case someone doesn't.. 😅
https://en.wikipedia.org/wiki/Large_language_model

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 29 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t
vvbee wrote on 2024-10-17, 03:34:

Many human artists fall under that criticism, could even use it to argue that AI is a real artist.

I disagree strongly with the latter simply because of the difference in how AI generates images versus how a real artist does it. In the latter case, artists typically use underlying structure to build up an image in layers. In contrast, a typical AI is just copying the "surface" of an image based on an LLM (which is really just fancy probability modelling).

And while I agree that artists can also create images with bad perspective and form, or poor lighting, the real, good artists can analyze such drawing and correct those underlying flaws while retaining effectively the same image.

You see this a lot with artists that redraw or correct their old work, or in cases where professionals redraw or correct amateur work.

For a couple examples of the above see:

Jim Lee transforms my Psylocke sketch into a Masterpiece
Critiquing and Overpainting My Old Art

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards

Reply 30 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t
matti157 wrote on 2024-10-17, 07:57:

But only I feel a sense of nausea\sickness when I look at these images? I am forced to close them after a few seconds

I find a lot of AI art has an "uncanny valley" feel to it, so it might be that.

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards

Reply 31 of 102, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

What I mean is your definition of real art excludes genres of human art. It's easy to disagree for starters that naive art isn't real art. If you struggle to exclude AI without also excluding human art then that doesn't speak of AI not being a real artist. What you might want to say is real art requires a human, but let's be real, you don't know what's required to have subjective experiences.

Reply 32 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t
vvbee wrote on 2024-10-17, 16:20:

What I mean is your definition of real art excludes genres of human art.

I never said "real art". I said real artist.

I was talking about the differences between traditional illustration methods and the way AI generates images.

I'm not trying to define "art" nor am I interested in having that debate.

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards

Reply 34 of 102, by gerry

User metadata
Rank Oldbie
Rank
Oldbie
Shponglefan wrote on 2024-10-17, 15:04:
vvbee wrote on 2024-10-17, 03:34:

Many human artists fall under that criticism, could even use it to argue that AI is a real artist.

I disagree strongly with the latter simply because of the difference in how AI generates images versus how a real artist does it. In the latter case, artists typically use underlying structure to build up an image in layers. In contrast, a typical AI is just copying the "surface" of an image based on an LLM (which is really just fancy probability modelling).

I'd agree with that. an artist (like most people) have a model of the world. we know that placing a cup on a table means it will rest there and may cast a shadow in relation to the source of light.

A generator 'knows' nothing, it simply 'looks' at what has gone before and uses that with some algorithmic weighting of color/pixel choices, mixing of sources and some randomness to produce something according to various inputs

that's what makes it amazing to me, that you can say "a person on a beach" and from that it can synthesise a superficially reasonable image showing a distinct person on what looks like a beach, for instance, without the generator needing to understand (in any human sense) the concepts of 'person', 'beach' or any intuitive knowledge of up/down, gravity or light at all.

the more often a word is 'image associated' the better the representation; very famous and much pictured objects like certain buildings, people and so on are well represented. apparently 1990's computers weren't photographed enough though, as the range depicted seems more limited and 'wrong' somehow.

while there will undoubtedly be improvements i dont think they will happen in a linear fashion, i.e. the improvements in the last 3 years won't be equalled in the next 3, from now its just refinements and more and more data

Reply 35 of 102, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

If you negate the idea of "generator" in a human then you end up with coma, it's an implied label. If you remove the effect of "'looks' at what has gone before and uses that with some algorithmic weighting of color/pixel choices, mixing of sources and some randomness to produce something according to various inputs" from human cognition then where does the 'knowing' come from? I think to realistically separate AI from the human experience you'd have to show that the subjective experience can exist in the physical world without being replicable there, or believe that you can share other peoples' internal worlds beyond their physical effects.

Reply 37 of 102, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
leileilol wrote on 2024-10-18, 23:15:

human artists can't consistently fuck up drawing two+ different people.

Think some had too much altman koolaid...

Sure about that?

Henri_Rousseau_-_The_Football_Players.jpg

If an artist fucks up consistently it's called a style, so in that sense it's ass backwards to think of it as fucking up.

Reply 38 of 102, by Shponglefan

User metadata
Rank l33t
Rank
l33t

Came across a video the other day that talks about detecting AI artwork. Rather than looking at the obvious like wonky anatomy, geometry, etc., it talks about things like channel noise patterns and compression artifacts.

For example, they mention how AI art will include random JPEG artifacts because it's training on JPEG images. So it includes artifact generation as part of its image generation, even if the image itself isn't a JPEG.

How to spot Generative AI? (even if it has all 10 fingers and toes)

Pentium 4 Multi-OS Build
486 DX4-100 with 6 sound cards
486 DX-33 with 5 sound cards