For the record, this man looks absolutely nothing like me 
1 Like
I don’t think, I know I can tell AI art. Did you miss the 20 or so replies where we were pointing out tells and making fun of them?
Topic removed due to directed bullying.
You can stop banging the table any time because nobody here is as dumb, naive, or delusional as you seem to be intent on seeing us as, and we heard you the first time. We have already had several long threads about this very subject, which are a lot more respectful even though people disagree on a lot of issues surrounding the subject; something you might want to learn if you want to have actual conversations.
I can wrap my head around it. It’s an impressive achievement on the part of AI researchers, but it’s not utterly mindboggling like quantum mechanics. Some of your statements remind me of people who’re amazed that Madame Sybil, the Great Psychic could tell so much about them, like how could she have known Grandma had been so ill before she died!?!?? Madame Sybil isn’t psychic, she’s just good at asking leading questions and reading people’s unconscious tells. Once people know how she does it, they’ll be much less impressed.
Let’s assume this was your actual prompt: “a BlenderArtist moderator panicking looking at a Blender screen deciding whether it is A.I. generated or not”
Not a good prompt, btw. AIs don’t do well with negation at this point, and “deciding whether it is A.I. generated or not” is something even a human graphics artist would have serious trouble depicting without being ham-handed about it. Very likely the entire last part of that prompt got discarded.
The first thing I noticed was that the AI did NOT give you what you asked for. This is a random guy, not one of the BlenderArtists moderators. The AI clearly also had no concept of what BA is, or what a Blender screen might look like; you had to fake those things. I suspect it can’t really determine what a moderator in general is either since that would be fairly hard to derive from a database of images of people with that occupation – how do you tell it from any number of other professions? It’s a guy not because all actual BA moderators are guys (remember, it knows nothing about BA) but because “moderator” isn’t gendered feminine (we know AI are biased by the biases of the training model – human images come with a huge slew of such biases). And “deciding whether it is A.I. generated or not” I already covered; you got nothing there.
So we have “a man panicking looking at a screen”.
The man doesn’t look like he’s panicking. He looks like he is concentrating. This AI isn’t particularly good with emotions. I get a lot of panicked imagery from an image search engine, though most of it looks very fake – but the AI wouldn’t care. It just can’t do it for some reason.
So we have “a man looking at a screen”. Much, much less specific already.
You sound amazed how the AI went further and what, you feel it captured what you envisioned so perfectly even though you didn’t even tell it? Yeah, no. Madame Sybil strikes again.
It didn’t invent that angle (you didn’t give it any directions about it).
It didn’t come up with how the man is seated and how far he should be away from the screen based on your direction (you didn’t give it any).
It didn’t calculate that perspective (you didn’t give it any directions about it).
It didn’t “create” an office-like building (you know you didn’t tell it to).
Type “a man looking at a screen” into your image search engine of choice and see what you get. Oh, look, a bunch of images of guys facing a computer screen. Many of them are silhouetted in front of a window, often in an office environment. Many of them are youngish and bearded.
This is highly simplified (you can learn elsewhere how this actually works in more detail): The AI was trained with a database of existing images of “man looking at screen” and derived angles, distance from screen, perspective, composition, colours, and the amazing office-like windows because a huge number of existing images show such things. They show them because that’s what a graphics artist would give you if you gave them this prompt, and so there are tons of these on the internet with that description associated. And that’s what the AI scrapes, from what it builds that “latent space” as it’s called – an abstract, multidimensional space in which representations of features and relationships between data are encoded – within which it knows everything (limited to the model) about how we depict “man looking at screen”.
That latent space is why people who state that the AI doesn’t take bytes or pixels from existing images aren’t lying liars pulling wool over your eyes, it’s because it does not take bytes or pixels. It just “learns” from decoding existing images and descriptive text how we semantically represent visual data, and it can use that information to make new images that represent the same data in various ways. It’s not just like you learning from all the images you’ve ever seen, it’s much more methodical and thorough than humans can ever be, and it has better recall – but it’s more akin to how we learn to synthesize from analyzing existing art than it is to making a collage from pieces of other people’s images. And even the latter is legal as it stands right now, as long as the new work is transformative enough.
I think that is why artists who complain they weren’t asked for permission don’t have much of a leg to stand on – no artist themself has ever asked artists who came before them for permission to learn from their images. And styles cannot be copyrighted. But the sheer scope of an AI to produce new images in the style of a human artist, and potentially flood the market and drive the original artist out of business, that’s sure frightening, and I can understand why people fear it and rebel against it. But I notice that you’re not boycotting the technology for their sake either – how many people are gonna put their money where their protesting mouth is? Technology marches on, and capitalism will exchange people for machines any chance it’ll get.
5 Likes
I am done talking to your bullying, mean, condensing attitude towards me for the last time @piranha4D.
A lot of those issues could be photoshopped out, but that would require effort! And AI artists seem to be allergic to effort.
Actually, I’d like to see people use these tools in a creative way, but unless they can find another AI to do it for them, they won’t.
These tools are really powerful and have a lot of potential in projects, but everyone seems to have the impression that it’s an all or nothing case.
3 Likes
Yes, you are right. Since the beginning, I always said it would be nice to use AI to generate textures. Things to make work easier where it’s necessary, not a “tool” that makes all the work for me.
2 Likes
Noting that the original poster of this thread has removed the content of their posts, I think it’s time for me to step in here.
There is one thing I’d like to clarify first, though. It’s not the moderation team who selects the works that show up in the Featured Row. These pieces are suggested and voted upon by forum community members. It would take a lot for something like any of these images to make it through. I’m not saying that it won’t happen; I’m just saying that it’s unlikely (as @joseph pointed out) and that it’s not the moderation team that makes these selections.
As for the actual topic at hand. There are quite a few other threads already in existence on this topic. If you want to continue to discuss your opinions of a world with machine-generated art, I suggest you jump into one of those threads. I’m going to close this one.
6 Likes