Of course, I was saying animal excluding humans.
Some thoughts art was exclusive to humans.
In the end, nothing is really exclusive to humans, we are just a bit better at a lot of things.
Please consider adding keyboard shortcuts for thumbs up/down and pass. I do like looking at the generated art, but clicking like/dislike for each is bit tedious.
I'm not sure that exercise is going to be very productive, though. A lot of time my feedback was "looks cool, but the hands / faces" look weird; except when the art style was more abstract, or the subject was landscapes where the generation errors were less blatant.
I don't think that feedback gives the AI something actionable. An AI trained on this feedback won't learn "I need to draw faces and hands better" (and anyway, we already knew that these were the areas to improve), it's going to learn "humans like it better when I do abstract styles and landscapes".
I enjoyed this. I got to level 13 which was labeled as 'Advanced'. It was not just fun to review and provide feedback, but critiquing the art was also kind of relaxing. Felt like I could do it for hours. Some of the art was genuinely good.
My feedback: this is artificially generated art, and is meaningless. Why would I spend any amount of time or effort on this when there is art made with intent by humans?
In the near future you may have little to no idea whether art was made by ai or not unless you see it happen in person. Maybe contemplating this future is necessary?
Contemplating authenticity (and ways that we can reasonably prove it) is a different discussion than wholesale generating ai "art" that has zero intent or meaning.
That's mass-produced furniture. And yes, the original intent/artist meaning of an art piece are extremely important when looking on as a viewer, even if the viewer ends up giving it a different meaning.
Sunsets don't have intent or meaning. A whole lot of people still like to look at them, or even assign their own meaning to the them, or other random machinations of the universe.
There is meaning in that the artist chose a sunset as the subject of the piece, and they decide the composition and colouring, etc. It's all done with intent. And in pieces with more than 1 element, the various objects -do- have symbolism or were chosen with some theme in mind.
We're talking about art, not phone wallpaper screens.
I personally spent many hours on one image starting with 3D printing armor, photographing it on my wife, and using that image as a base for ControlNet + Stable Diffusion to replace the armor, and then compositing it back in.
Some people just type in “pretty woman wearing armor.”
An Artist* utilizing different mediums & tools to achieve an intended vision is different than someone typing a few words on a whim and re-generating until the magic box outputs something roughly close to what they want. Using ML as part of an overall workflow is much more gray.
The line is where we draw it, and we have to draw it somewhere.
It's clear to anyone that typing prompts into a black box (that likely was trained on stolen art) and expending zero effort other than selecting from the randomly generated images is different than the process you described.
* simply writing prompts doesn't make anyone an artist
> and re-generating until the magic box outputs something roughly close to what they want
That is a form of a creative human effort. Not that different from a photographer who shoots a couple dozen photos on the go, and later selects one or few they like best, to publish without further retouching.
It's gonna be more blurry when someone loops SD derivative with GPT-4V derivative, to have the latter rate the output of the former, and the loop running until the rating model deems the picture good enough.
Someone writing words in a prompt and a photographer setting up a photoshoot are completely different! Different enough that we can easily draw a line between these activities.
In the above case, the photographer had a particular vision in mind and fulfilled it with their camera, choosing the scenery, lighting, model, colors, framing, and overall composition. They set up all of those factors, and then chose the best take from the portraits to use for further work. It was deliberate and done with specific intent. There is no way one can argue this is the same as entering words into a prompt with a straight face.
>There is no way one can argue this is the same as entering words into a prompt with a straight face.
I have no stakes in this game but yeah, I think they are the exactly the same. I always thought of photography as a "phony" art, unlike classical canvas painters. You're using a machine that does all the work for you. In this case, the prompt is the camera positioning, lighting etc.
Not that any of this matters anyway, so don't waste energy with an uphill battle. As long as they like the actual result, people won't care.
> Someone writing words in a prompt and a photographer setting up a photoshoot are completely different! Different enough that we can easily draw a line between these activities.
That's why I didn't write "a photographer setting up a photoshoot"; I specifically wrote "a photographer who shoots a couple dozen photos on the go". Perhaps I should've written "a person owning a DSLR" or "amateur instagrammer" instead.
Point is, when discussing copyright both in general, and wrt. generative AI in particular, people tend to bucket together casual, amateur and professional photographers, making it seem that the former do the same kind of creative work as the latter. But that's not true, and not fair - there isn't much more effort or creativity going into casual photography, than there is into iterating a stable diffusion prompt until one finds something they like. Professional photographers have their generative-AI-user counterpart, but in form of an artist or graphic designer using AI-assist tools as part of their overall workflow.
Anyone can call themselves whatever they want, but others aren't necessarily going to acknowledge it. Just like how software developers call themselves "engineers".
People writing prompts can certainly attempt to call themselves "artists", but they will not be perceived or acknowledged as such by literally anyone who takes making art seriously.
Who is the arbiter of terminology? Your argument is entirely unconvincing. Barring places like Canada where the term “engineer” is tied to actual credentials, no such thing exists for artists. Can you imagine, a board of people to license you for artistic practice? A hell I’d rather not visit.
You can also do the same thing with human art. There are most certainly local galleries near you, and you can interact with an actual human and the meaning they put into that art. Not to mention the amount of art made by humans that exists online.
I clicked "Browse the gallery" and expected to be able to browse the gallery, but was instead confronted with a kind of mobile chat window - not what I expected and not what I came for. I immediately left the site.
The chat style introduction is an attempt to have most people read about the context. If you wait a few seconds you can close it and browse the gallery. I know it wasn’t going to be popular on HN, it’s a trade off.
I'm about a 1:20 like to dislike ratio. Most of the dislikes are do to poor anatomy, ornamental text, overly smooth output(which weirdly people seem to like), unidentifiable objects.
> overly smooth output(which weirdly people seem to like)
A pet peeve of mine as well. The over-processed look is something I can't get behind. Especially recently on TikTok, users apply a very strong smoothing filter to film footage sometimes, like an denoise turned up to 11.
Current state of generated images resemble something that has zero human like attributes. Therefore it is REPULSIVE in every way. It's like telling people to copulate with slimy, gargling octopuses instead of real human females.
It won't work, because we are wired differently. I browse a lot of art and AI art too, and there are about 1-2 pictures among a 1000 that I say is good and has human made characteristics. Most of them are trash.
So we are a long way away from rearranging talent space and giving the talent of pro artists in the hands of average joes (or rather to the elite).
40 comments
[ 3.1 ms ] story [ 95.3 ms ] threadThe computer generated images in a weird way by using random noise and mixing the prompt embeddings.
It results in a lot of broken or corrupted images but also some unique images.
The page contains some technical details and some opinions about generative AI art.
I'm not sure that exercise is going to be very productive, though. A lot of time my feedback was "looks cool, but the hands / faces" look weird; except when the art style was more abstract, or the subject was landscapes where the generation errors were less blatant.
I don't think that feedback gives the AI something actionable. An AI trained on this feedback won't learn "I need to draw faces and hands better" (and anyway, we already knew that these were the areas to improve), it's going to learn "humans like it better when I do abstract styles and landscapes".
You don’t have to give feedback to the AI, you can write whatever you feel.
It’s a bit explorative research. Perhaps the final dataset will be useful but it’s not a given.
Does art require intent and meaning?
We're talking about art, not phone wallpaper screens.
Some people just type in “pretty woman wearing armor.”
Where’s the line?
The line is where we draw it, and we have to draw it somewhere.
It's clear to anyone that typing prompts into a black box (that likely was trained on stolen art) and expending zero effort other than selecting from the randomly generated images is different than the process you described.
* simply writing prompts doesn't make anyone an artist
That is a form of a creative human effort. Not that different from a photographer who shoots a couple dozen photos on the go, and later selects one or few they like best, to publish without further retouching.
It's gonna be more blurry when someone loops SD derivative with GPT-4V derivative, to have the latter rate the output of the former, and the loop running until the rating model deems the picture good enough.
In the above case, the photographer had a particular vision in mind and fulfilled it with their camera, choosing the scenery, lighting, model, colors, framing, and overall composition. They set up all of those factors, and then chose the best take from the portraits to use for further work. It was deliberate and done with specific intent. There is no way one can argue this is the same as entering words into a prompt with a straight face.
I have no stakes in this game but yeah, I think they are the exactly the same. I always thought of photography as a "phony" art, unlike classical canvas painters. You're using a machine that does all the work for you. In this case, the prompt is the camera positioning, lighting etc.
Not that any of this matters anyway, so don't waste energy with an uphill battle. As long as they like the actual result, people won't care.
That's why I didn't write "a photographer setting up a photoshoot"; I specifically wrote "a photographer who shoots a couple dozen photos on the go". Perhaps I should've written "a person owning a DSLR" or "amateur instagrammer" instead.
Point is, when discussing copyright both in general, and wrt. generative AI in particular, people tend to bucket together casual, amateur and professional photographers, making it seem that the former do the same kind of creative work as the latter. But that's not true, and not fair - there isn't much more effort or creativity going into casual photography, than there is into iterating a stable diffusion prompt until one finds something they like. Professional photographers have their generative-AI-user counterpart, but in form of an artist or graphic designer using AI-assist tools as part of their overall workflow.
Being an artist is a self-designated term so you’re only projecting your own biases with this statement.
People writing prompts can certainly attempt to call themselves "artists", but they will not be perceived or acknowledged as such by literally anyone who takes making art seriously.
"AI art" is an oxymoron.
I'm about a 1:20 like to dislike ratio. Most of the dislikes are do to poor anatomy, ornamental text, overly smooth output(which weirdly people seem to like), unidentifiable objects.
Does it get more interesting?
A pet peeve of mine as well. The over-processed look is something I can't get behind. Especially recently on TikTok, users apply a very strong smoothing filter to film footage sometimes, like an denoise turned up to 11.
It won't work, because we are wired differently. I browse a lot of art and AI art too, and there are about 1-2 pictures among a 1000 that I say is good and has human made characteristics. Most of them are trash.
So we are a long way away from rearranging talent space and giving the talent of pro artists in the hands of average joes (or rather to the elite).