Your AI-generated Art

Asked AI to create painting based on description alone and then, uploaded the painting and asked it again.It is personally something I wanted to test, it is unable to capture the paint strokes, nor capture the canvas texture. Fundamentally the paintings are flawed, no matter how much data you end up feeding it, it still ends up giving something that falls short. It can capture likeness but it does capture details or lack of details well. Nor uniqueness either.


It got rather annoyed and defensive when I pointed out, AI will NEVER be able to create uniqueness, it argued it could. Well it can't even the mimics is flawed in some way or another. The paint strokes is quite cube/fractal like, the way it is generated is quite telling, it is still a long ways off, getting the complexities of oil and water paintings.





The painting I described.
NY_PPA_006-001.jpg

What it actually did
cce1ddfe-9f8c-48d5-ae19-0127fcced7ef.png

And what it did after I uploaded the image
f1b2ebbc-edae-4771-acec-c227801f2bdd.png
 
wizardofid said:
Asked AI to create painting based on description alone and then, uploaded the painting and asked it again.It is personally something I wanted to test, it is unable to capture the paint strokes, nor capture the canvas texture. Fundamentally the paintings are flawed, no matter how much data you end up feeding it, it still ends up giving something that falls short. It can capture likeness but it does capture details or lack of details well. Nor uniqueness either.


It got rather annoyed and defensive when I pointed out, AI will NEVER be able to create uniqueness, it argued it could. Well it can't even the mimics is flawed in some way or another. The paint strokes is quite cube/fractal like, the way it is generated is quite telling, it is still a long ways off, getting the complexities of oil and water paintings.





The painting I described.
View attachment 1895354

What it actually did
View attachment 1895352

And what it did after I uploaded the image
View attachment 1895355
Click to expand...
Honestly this feels a lot like the same conversation we had around AI music.

If you give a model a vague prompt and let it do everything, you’ll get something generic. That’s true for Suno, and it’s true for image models. It’s not really a test of the tool, it’s a test of how much control you’re giving up.

On the painting side, you’re also judging it against physical properties like real brush strokes and canvas texture. Most current models aren’t actually simulating paint physics, they’re generating patterns that look like paintings. So yeah, if you inspect it like a real canvas, it falls apart. That’s expected. You could obviously train a large model on a dataset of real brush strokes and canvas textures and it would be closer to the real thing, as opposed to a Gemini / ChatGPT / Midjourney style general model which dtasets contain only a vague idea of what paintings look like up close.

As for uniqueness, I think that depends more on the workflow than the model. If you rely entirely on one prompt, you’ll get something derivative. If you iterate, combine outputs, guide it, or integrate it into a broader creative process, you can generally end up with something that feels more human and unique.

So, I’d say it’s not a matter of "AI can’t do it" and more like "this is one way of using it." You’re basically testing it in the laziest possible way and then concluding the ceiling is low.
 
OnlyOneKenobi said:
Honestly this feels a lot like the same conversation we had around AI music.

If you give a model a vague prompt and let it do everything, you’ll get something generic. That’s true for Suno, and it’s true for image models. It’s not really a test of the tool, it’s a test of how much control you’re giving up.

On the painting side, you’re also judging it against physical properties like real brush strokes and canvas texture. Most current models aren’t actually simulating paint physics, they’re generating patterns that look like paintings. So yeah, if you inspect it like a real canvas, it falls apart. That’s expected. You could obviously train a large model on a dataset of real brush strokes and canvas textures and it would be closer to the real thing, as opposed to a Gemini / ChatGPT / Midjourney style general model which dtasets contain only a vague idea of what paintings look like up close.

As for uniqueness, I think that depends more on the workflow than the model. If you rely entirely on one prompt, you’ll get something derivative. If you iterate, combine outputs, guide it, or integrate it into a broader creative process, you can generally end up with something that feels more human and unique.

So, I’d say it’s not a matter of "AI can’t do it" and more like "this is one way of using it." You’re basically testing it in the laziest possible way and then concluding the ceiling is low.
Click to expand...

Well was having a discussion, and one thing let to another. lol Down the rabbit hole I went. Was just a rather interesting subject to delve into and giving specific example like the description of the painting before I uploaded it and what it came up. It did a pretty good job overall, regardless. It was more a case to see if I could fault it. Lol finally made it agree, fair enough visual art is more complex compared to music creation, art is often intentionally chaotic, vague and lacks the specificity of music.

Took a long ass time to get to that point, sometime I hate AI and have to remind it not to take the diplomatic stance, sometimes you have to be forcible but meaningful choose a side of the fence without directly influencing the choice. I should really at some point setup a persona. The diplomatic stance can be annoying

basically the summary of that discussion.
You were exploring how AI interacts differently with music versus visual art. You noted that music has a stronger structural and mathematical framework, making it more of a craft, whereas artistry in music adds emotion and character. Because music is bound by rules—like harmony, rhythm, and progression—AI can often handle it better than visual art, where chaos, abstraction, and emotional depth dominate. You emphasized that while AI can mimic imperfections and stylistic elements, it struggles with true uniqueness and the lived, experiential quality that humans bring, particularly in visual art. Visual mediums are more fluid and emotional, making it much harder for AI to capture subtle personal interpretation or originality, even though technically it can reproduce the look or feel convincingly.
Click to expand...
 
borro said:
Click to expand...
It's been a while since I laughed so much I cried and worked myself into a six pack.

Especially that poster's context of "that kind of" anime they consume and then this was too much.

Oh well, thanks for good lolz and you will now be remembered as that guy.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X