AI image synthesis is getting more capable at executing ideas, and it's not slowing down.
See full article...
See full article...
From toy to tool: DALL-E 3 is a wake-up call for visual artists—and the rest of us
- Thread starter JournalBot
- Start date
Upvote
178
(190
/
-12)
This tech is so cool, but man... so many creative jobs may be lost to it, and that would be bad.
Now, we get to hear from the libertarians who will tell us how this tech will actually create 1E12 more new jobs that we can't yet even imagine...
Now, we get to hear from the libertarians who will tell us how this tech will actually create 1E12 more new jobs that we can't yet even imagine...
Upvote
118
(168
/
-50)
So DALL-E decided that the queen of the universe is a white human. Big surprise.
Upvote
135
(258
/
-123)
Most of it looks good. Some of looks really good. But knowing it wasn't actually made by a human with any significant effort doesn't sit right with me. Maybe the ideal use is for prototyping/creating placeholder art. At its worst, it's essentially black-box plagiarism. At its best, it looks cool, but it sort of... has no soul? Unless I can convince myself to believe there's actually creative spark and real intelligence within the AI framework, but that's going down a different path...
Upvote
31
(102
/
-71)
Meh; still prefer actually opensource, ie: SDXL.
https://meincmagazine.com/information...generated-visual-worlds-at-your-gpus-command/
https://meincmagazine.com/information...generated-visual-worlds-at-your-gpus-command/
Upvote
82
(85
/
-3)
As a D&D game master, GPT4 and image generators have been transformative in my hobby workflow. Thinking what it means for the real world, though, is fuel for my actual games. Note that I tend to use Lovecraftian horror, existential dread, and apocalyptic speculation as staples in my world
Upvote
120
(128
/
-8)
"Frosted prick cereal" huh. We should all now be thankful that Dall-E is not British, I suppose.
Upvote
129
(131
/
-2)
Upvote
43
(44
/
-1)
Upvote
137
(143
/
-6)
DALLE-3 is good. Good enough for many things, but I'm finding that AI art generation is - and will likely only be for the forseeable future - only good enough still to prototype an idea...yes, it can generate a squirrel as Mad Max...but it isn't the image I have in my head. It isn't what I want.
Is it good enough? Sure. Does it get a point across? Yes. Does it fill that temporary creative itch to visualize a concept and move on? Absolutely. Does it give me that precision, those details that make me say, "Yes, that's the idea fully realized?" No. And it can't, and it never will. Only a human can do that. And only a human will ever be able to.
Oh, and here's a squirrel as Mad Max. Not my squirrel as Mad Max, but amusing all the same. Stable Diffusion couldn't do it. DALLE-2 couldn't do it. A little unnerving just how far it's come in how short a time:
And what the heck. A cuter one, where this human had to explicitly request the car. :
Is it good enough? Sure. Does it get a point across? Yes. Does it fill that temporary creative itch to visualize a concept and move on? Absolutely. Does it give me that precision, those details that make me say, "Yes, that's the idea fully realized?" No. And it can't, and it never will. Only a human can do that. And only a human will ever be able to.
Oh, and here's a squirrel as Mad Max. Not my squirrel as Mad Max, but amusing all the same. Stable Diffusion couldn't do it. DALLE-2 couldn't do it. A little unnerving just how far it's come in how short a time:
And what the heck. A cuter one, where this human had to explicitly request the car.
:
Upvote
118
(143
/
-25)
I'm sure art will adapt. It adapted to photography, it adapted to image manipulation tools like Photoshop. That doesn't excuse the arrogant 'scrape all the data without bothering to ask permission' way that most AI image generators have been trained and rolled out. Not to my mind anyway.
Oh, and I love the snide 'wake up call' headline for this article. "Hey, wake up call. We've done this thing, it screws you over in new and improved ways, have a nice day.' Not much point having a wake up call if there's nothing you can do in response.
Oh, and I love the snide 'wake up call' headline for this article. "Hey, wake up call. We've done this thing, it screws you over in new and improved ways, have a nice day.' Not much point having a wake up call if there's nothing you can do in response.
Upvote
139
(165
/
-26)
Worth noting is that on the "Gerbil Essences" piece, it still absolutely mangled what a roller skate looks like, creating some sort of a weird skateboard hybrid that would probably give QT nightmares with what it thinks a human foot is shaped like.
Upvote
46
(50
/
-4)
Technology is not that interesting. Pictures still look pretty same, glossy, interesting yet visible remixed from other ideas. This is essentially uncanny valley of creativity.
However economical and social effects is not something technology can answer. It is about what we value more - bullshit idea that everyone can be CEO, that corporations deserve all the money in the world, or we actually ensure all people have basic income and then those with creative minds can make things they want, and maybe even get something on the top.
It is pointless to scare monger, point out shockingly how this gonna replace artists. It is not gonna. Artists will be artists. It will however force question about what we all do this technological development for. If it is for corporations, yeah, sign me out.
However economical and social effects is not something technology can answer. It is about what we value more - bullshit idea that everyone can be CEO, that corporations deserve all the money in the world, or we actually ensure all people have basic income and then those with creative minds can make things they want, and maybe even get something on the top.
It is pointless to scare monger, point out shockingly how this gonna replace artists. It is not gonna. Artists will be artists. It will however force question about what we all do this technological development for. If it is for corporations, yeah, sign me out.
Upvote
-19
(38
/
-57)
Upvote
167
(169
/
-2)
I also find that picture unsettling (and hilarious, in its own way, that the cereal is doing a Dreamworks face), and I think it’s because of the hand. The skin seems unnaturally smooth and moisturized.
Upvote
13
(13
/
0)
The depressing discourse on the impossibility of putting genies back in bottles is valid enough, so here's a playbook for those who think the humanities should be for humans:
1) Patronize and promote human art. Drop a streaming service in favor of a monthly pledge to a favorite cartoonist, musician, whomever.
2) Resist the urge to turn to AI when you want to make or edit a funny picture. Break out the MS Paint. Let the imperfect humanity be part of the joke. Appreciate it when others do this. Normalize a preference for the personal.
3) When commercial media uses AI-generated material as a substitute for the work of people, scorn it and shame it. Even for media from small teams with low budgets, express a strong preference for placeholder assets, even amateurish ones, over AI-generated art.
1) Patronize and promote human art. Drop a streaming service in favor of a monthly pledge to a favorite cartoonist, musician, whomever.
2) Resist the urge to turn to AI when you want to make or edit a funny picture. Break out the MS Paint. Let the imperfect humanity be part of the joke. Appreciate it when others do this. Normalize a preference for the personal.
3) When commercial media uses AI-generated material as a substitute for the work of people, scorn it and shame it. Even for media from small teams with low budgets, express a strong preference for placeholder assets, even amateurish ones, over AI-generated art.
Upvote
50
(101
/
-51)
By definition, it is not possible for every competitor in a field to be considered a major player.
Reading articles like this one is an exercise in suspending basic economic ideas like supply and demand.
Upvote
17
(38
/
-21)
Dall-E still doesn't get hands if you look at the Beet Bros arcade game. How would you even play that? And why would you need to play it differently depending on if you're player 1 or player 2?
Upvote
8
(13
/
-5)
If anyone is under any misapprehension that this is going to result in lots of job losses and job downgrades (lower pay, because now you're not really using any hard-to-train/develop-skill, but just 'generating ideas' for the AI to spit back at you) - the human artist talking about handmade art being important because it's nourishing the soul should be a real eye opener.
Upvote
18
(32
/
-14)
While I know there's a fight on right now to get people to acknowledge that images are AI and how to identify deepfakes, I do wish that AI companies would make it easier to tag or otherwise make clear that an image is AI generated and that it is in the public domain. I do consider anything DALLE or StableDiffusion creates for me to be public domain but I currently have no way to make that explicit or obvious when I post it, other than a blurb in the post itself...which gets tedious, distracts from the point, and shouldn't be necessary on an image made as a throwaway joke.
Upvote
12
(22
/
-10)
The quality is great, although a lot of these images still have a bizarre, unreal quality to them.
I'm less concerned about the loss of jobs for artists than I was before due to movement in the legislative direction but also just movement by people themselves.
In the boardgame space we've seen that projects announcing themselves to be AI-art free (and hence supportive of real human artists) get a very positive reception.
If people need to ask themselves why they would value human-made art over AI-art then the answer is actually pretty simple: A lot of people value the effort other people put in to producing things and don't react as positively to what they see as someone "attempting to make a quick buck", so to speak.
I'm less concerned about the loss of jobs for artists than I was before due to movement in the legislative direction but also just movement by people themselves.
In the boardgame space we've seen that projects announcing themselves to be AI-art free (and hence supportive of real human artists) get a very positive reception.
If people need to ask themselves why they would value human-made art over AI-art then the answer is actually pretty simple: A lot of people value the effort other people put in to producing things and don't react as positively to what they see as someone "attempting to make a quick buck", so to speak.
Upvote
54
(65
/
-11)
While it is repeatedly argued that every new technology spawns more job opportunities than kills old ones, the advent of ai has a unique quality. That quality is "thinking" for us. Don't stumble over the exact nature of that verb. Today's ai might not think in the conventional sense of the words, but it's getting closer. Ai is unlike the creative assistance that conventional software offers. It easily reaches far deeper into our perception of reality than previous technology and it does that semi-autonomously, or autonomously.
A vision of any future with ai needs to include basic human psychology. We are creative. But we are also lazy. It seems to me and this is an entirely personal opinion, that intellectual laziness is far more prevalent than creativity. If true, there is possibly a distopian future awaiting us where ai obviates the need to, or worse the impulse to learn basically anything. If ai serves the species with all that is needed to survive and to enjoy that existence, humanity's impulse to create could wither. We "devolve" into a parasitic co-existence with the ai mother who generates energy, heat, food, as well as entertainment and emotional warmth.
Dismiss this vision as standard apocalyptic fare. You're entitled to your opinion. It's simply a view of a possible future based on what we all witness every day of humanity's near limitless capacity at hubris and selfdeception. It seems to me that our current day societal psych profile does not offer much support to the conviction that we are psychologically resilient enough to resist the temptations that a future ai could offer.
A vision of any future with ai needs to include basic human psychology. We are creative. But we are also lazy. It seems to me and this is an entirely personal opinion, that intellectual laziness is far more prevalent than creativity. If true, there is possibly a distopian future awaiting us where ai obviates the need to, or worse the impulse to learn basically anything. If ai serves the species with all that is needed to survive and to enjoy that existence, humanity's impulse to create could wither. We "devolve" into a parasitic co-existence with the ai mother who generates energy, heat, food, as well as entertainment and emotional warmth.
Dismiss this vision as standard apocalyptic fare. You're entitled to your opinion. It's simply a view of a possible future based on what we all witness every day of humanity's near limitless capacity at hubris and selfdeception. It seems to me that our current day societal psych profile does not offer much support to the conviction that we are psychologically resilient enough to resist the temptations that a future ai could offer.
Upvote
50
(63
/
-13)
I find it curious that several of the images with writing in them have defects (as cited by others above). Some of the defects appear as spelling errors (e.g. "marsmallow" missing the "h"), while in others the writing is illegible (look at the "Gerbil Essences" bottle). Have we identified a DALL-E 3 "tell"?
Upvote
-17
(12
/
-29)
That sounds great - in theory - but how do you police such a claim, especially as the technology improves and it really will be totally indistinguishable from a human-made artwork?
There will be all sorts of unscrupulous vendors trying to cash in on the "hand made" movement, claiming one thing and using AI in the background to maximise their profit.
Upvote
31
(32
/
-1)
It's been interesting seeing the changes in art communities. Sharing sketches and interstitial aspects is now a more commonly seen "proof of [human] work" in art packs and how different gaming communities react to AI assets. It's also a little amusing in a mean way how people online will jump on an artist for using AI and then there's this bashful reveal that they didn't and they're just bad at drawing hands.
Upvote
62
(62
/
0)
I've had a ton of fun lately playing around with DALLE-3. My toddler loves it, my mother loves it, it's a great creative outlet.
Upvote
21
(28
/
-7)
As with any tool, AI generated art has it's pros and cons. For me, one of the big disadvantages is that it doesn't have any form of internal consistency or "artist intent".
To take one example: if I asked for something like "a soldier wearing a lot of medals"[*], then what kind of medals would it pick? Would they be for the right country and/or time period for the uniform it created?
Equally, there's been hidden meanings/easter eggs in art since the dawn of time.
https://www.rd.com/list/secrets-famous-paintings/
Certainly though, it's at the point where it's Good Enough (tm) for a lot of use cases. And for better or worse, the vast majority of people are happy to settle for Good Enough, especially if it saves significant amounts of time and/or money.
Anecdotally though, there's still a lot of work involved in getting it to churn out good quality images.
To take one example, I've found it quite hard to get Dall-e to "zoom out", when playing with it to try and make gig posters/flyers and the like.
E.g. a prompt like "far away skeletons wearing a variety of hats dancing on a huge field in the distance in a woodcut style, zoomed out". That virtual camera stays very much close-up on the scene, presumably because that's what the majority of training material tends towards.
Equally, when trying to create images of "cats with machine guns", it was great at generating images with generic tabby-cats. Asking for a specific breed or hair colour caused the quality of the generated images to significantly deteriorate, presumably because there's simply fewer images of non-tabby cats in the training material.
(The latter may have been addressed to at least some degree; a quick test of "siamese cat waving a chainsaw in the air" was a lot more successful than the last time I tried!)
[*] Ironically, I just tried this prompt in Bing, which then refused to generate it because "we detected unsafe content in the images based on our content policy". And then the link to their content policy didn't work. Top work, guys!
To take one example: if I asked for something like "a soldier wearing a lot of medals"[*], then what kind of medals would it pick? Would they be for the right country and/or time period for the uniform it created?
Equally, there's been hidden meanings/easter eggs in art since the dawn of time.
https://www.rd.com/list/secrets-famous-paintings/
Certainly though, it's at the point where it's Good Enough (tm) for a lot of use cases. And for better or worse, the vast majority of people are happy to settle for Good Enough, especially if it saves significant amounts of time and/or money.
Anecdotally though, there's still a lot of work involved in getting it to churn out good quality images.
To take one example, I've found it quite hard to get Dall-e to "zoom out", when playing with it to try and make gig posters/flyers and the like.
E.g. a prompt like "far away skeletons wearing a variety of hats dancing on a huge field in the distance in a woodcut style, zoomed out". That virtual camera stays very much close-up on the scene, presumably because that's what the majority of training material tends towards.
Equally, when trying to create images of "cats with machine guns", it was great at generating images with generic tabby-cats. Asking for a specific breed or hair colour caused the quality of the generated images to significantly deteriorate, presumably because there's simply fewer images of non-tabby cats in the training material.
(The latter may have been addressed to at least some degree; a quick test of "siamese cat waving a chainsaw in the air" was a lot more successful than the last time I tried!)
[*] Ironically, I just tried this prompt in Bing, which then refused to generate it because "we detected unsafe content in the images based on our content policy". And then the link to their content policy didn't work. Top work, guys!
Upvote
43
(46
/
-3)
You sound like if a supervillian were a redditor, why does every AI booster have to be so lame.
And the "accessbility" narrative is so cynical coming from web3/crypto scammers(like the "artist" mentioned in this article), like whining about the price of art school when art is more accessible than ever, paint some Bob Ross or make a donut in Blender!
Upvote
2
(42
/
-40)
Wake up call for artists? What are they supposed to do with their measly human brains?
Upvote
3
(13
/
-10)
If you've ever made proper buttermilk pancakes at home from scratch you will know that the "buttermilk pancakes" served at most restaurants are nothing at all like them in taste or texture. Yet, patrons happily order them along with their breakfast eggs.Gwendolyn Wood said:
Most people do not know any better, nor care. Upon receiving pancakes, they slather them with butter and syrup. They could be eating tasteless discs of flour that tear apart for all they care...and they are.
Unfortunately, the number of people who understand and care will always drop dramatically.
You see this in all industries. For instance, for decades, there was no such thing as "DLC" for video experiences. Companies such as Nintendo or Capcom which once upon a time sold complete products for one set price now gleefully slice off parts to sell for extra, free money. I have even seen some people comment that they will not purchase entertainment software that doesn't have DLC available. These companies generally target teenagers so there are always new customers as the older players become disenchanted and quit. So now, DLC has burrowed its way into products the way a 3-inch roundworm will burrow into a person's brain.
You always hear how bespoke leather shoes or a suit will fit better than those off-the-shelf, and while such articles were commonplace decades ago, they are no longer something that matters to people.
As a writer, I can see the technology will soon take prompts along the lines of 'Write a 500 page story about a man who has to rescue his dog from a dragon' and you'll have the output shortly. If the reader is dissatisfied with any part, they can receive a new tale almost immediately. Compare that to 'I spent months of my life working on this novel' and soon enough, that explanation will be met with a glazed-eyed, 'So?'
As the popularity of reality television attests, a large portion of viewers are satisfied with filler.
The funny bit is that back in 2022, many artists lacked the imagination to see that the technology would improve, rapidly. They scrutinized the AI outputs carefully and said, 'the hair is wrong' or 'the hands are wrong.'
Since then, many artists have tried to reassure themselves by repeating the line "it's only a tool."
Yes, it's only a tool that can replace dozens of artists working for hours at a time, in minutes. Is the output as good? Certainly not but the bar is generally not "good," it's "good enough" and for most people, the pancakes being turned out fill the stomach cheaply and quickly.
Upvote
163
(178
/
-15)
Yep. It encodes a compressed representation of the training distribution. That means it doesn't just learn biases present in the training set, it actually amplifies them. This is made worse by the fact that future rounds of model training will undoubtedly include AI-generated art, creating a feedback loop.
Relevant articles on bias in image generation AI, with examples, first from Bloomberg earlier this year:
https://www.bloomberg.com/graphics/2023-generative-ai-bias/
And WaPo more recently:
https://www.washingtonpost.com/tech...erated-images-bias-racism-sexism-stereotypes/
And a paper on how including model output in the training of a new model inevitably results in forgetting the true distribution over time:
https://arxiv.org/abs/2305.17493
Upvote
25
(37
/
-12)
I don't like the side effects we see coming to human artists (like a train coming down the tracks right now). But this does bode well for higher visual fidelity rogue-like games.
Upvote
4
(11
/
-7)