Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How a Stable Diffusion prompt changes its output for the style of 1500 artists

gorgeous.adityashankar.xyz

Post reply on HN
Print view View original

141–150 of 205 posts

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#141

Post by logisticpeach »

p1esk wrote: post #78
LeoPanthera wrote: post #73

Earlier quoted context omitted.

> People said this about cameras. About digital cameras. About digital photo editing software. Did they? Because I don't think they did. I think most people were amazed by all these technologies.

Most people haven't heard about recent advancements in image generation. When they do, I expect they will be amazed.

True, but then we've essentialy had limitless image generation capabilities since we've had the tools to make marks. I guess this is faster, and in other ways it offers promising new opportinities for people who can't / don't want to learn to create stuff directly.

Others are interpreting my original comment as "this is not art", but I'm not really trying to make that argument. Art is entirely subjective and i don't presume to define what is or isn't art.

I guess my point is more specifically "what itch does this scratch"?

It's really cool, and that may well be the answer tbh.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#142

Post by astrange »

spijdar wrote: post #33
jhbadger wrote: post #25

What frustrates me about Stable Diffusion is there doesn't seem to be any documentation as to what artists or vocabulary it understands. Generally people say "look at existing prompts or use various prompt generators" but that doesn't really solve the problem. I don't want to just look at what other people have randomly discovered; I want to know what the program really knows.

The simple answer is that there is no clean cut list of artists that it "understands". The model has no explicitly programmed concept of artist or style -- just the CLIP based text encoding used to train the conditional autoencoding part of the denoiser network, trained on (AFAIK) caption data recorded with the image. So in practice asking for art "in the style of " is sort of limiting the denoiser to statistical pat…

You can create (or discover) explicit vocabulary in the model using “textual inversion”, or train more into it using fine tuning.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#143

Post by swid »

pmoriarty wrote: post #112
pessimizer wrote: post #18

Earlier quoted context omitted.

A good measure for whether you're more of a celebrity or an artist is how much of your face a google-image trained AI thinks belongs in your work.

> A good measure for whether you're more of a celebrity or an artist is how much of your face a google-image trained AI thinks belongs in your work. That this happens at all is evidence that the training data hasn't been curated, cleaned, or labeled well enough.

While it is the case the dataset isn’t well curated or perfectly labeled, it could just mean that grammar is not understood - the labels could be clear to a human, whether the image is a picture of Bob Ross or a painting by him. But the training misses that relationship. Even with poorly labeled data, I suspect AI will eventually figure out which labels are more likely to be poor and deal with it appropriately.

In the reverse direction, you can try:

A horse rides an astronaut

And you will probably generate an astronaut riding a horse. It’s not a poor description of what we want; our assumptions about how grammar should work aren’t being honored.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#145

Post by odessacubbage »

intelVISA wrote: post #128
educaysean wrote: post #71

Earlier quoted context omitted.

I'm of the opposite opinion. AI assisted art is simply the natural next chapter for "art" as a whole. It will finally kickstart the public discourse about what being an artist means in the perspective of artistic vision vs execution. Most artists spend their lives not refining their brush stroke, but rather their eyes. The way I see it, the impact of curation and artistic direction will matter more and more in the fu…

As with programming, an AI model cannot replace the key parts but can help automate the monotony. For me it's exciting to use as placeholder art and then have a 'real' artist review it.

i find myself mentally unable to comprehend people who believe that the drawing part of drawing is monotony. discovering that this mindset not only exists but is widespread has been equally as disturbing as any ai advancement.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#147

Post by Aeolun »

hecanjog wrote: post #4

Funny that the Bob Ross version just makes them look like Bob Ross. Maybe there are more pictures of Bob Ross in the training set than his actual paintings.

Walt Disney makes the pictures look like promotional Disneyland pictures.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#148

Post by nl »

krumbie wrote: post #67

What's kind of crazy is how the images tend to have similarities in small features that become very apparent when flipping back and forth between images, but which are not obvious per se. For example, I flipped back and forth between Beatrix Potter and Paulus Potter. A rounded white bonnet in one picture becomes a couple of blossoms in the other. The roof of a house becomes some shadowy wall with plants in the other.…

If you play with Stable Diffusion enough this behavior becomes very apparent. Changing the seeds will give different results, but even relatively significant changes in the prompt will still find similar themes or layouts.

Re: How a Stable Diffusion prompt changes its output for the style of 1500 artists

#150

Post by eru »

pmoriarty wrote: post #113
kjeetgill wrote: post #85

Earlier quoted context omitted.

Frida Kahlo comes to mind.

She's a rare exception in that she's mostly known for her self-portraits. Most other famous artists are mostly known for other things. Again, if the training data was labeled well enough, confusion about this sort of thing shouldn't happen.

Either labelled well, or you have enough data and a good enough algorithm so that the computer can figure it out.
Post reply on HN