Practically, this sort of thing isn't a creative revolution, but the destruction of economically viable human creativity. Great, it can make art in whatever style you want. It can insert Kermit...
Practically, this sort of thing isn't a creative revolution, but the destruction of economically viable human creativity. Great, it can make art in whatever style you want. It can insert Kermit the Frog into any movie you want. If what you want is the image equivalent of the 30'th Marvel movie, it can churn out the memes and regurgitate the most popular stuff ad nauseam. This feels like an inflection point between the visual art paid for because you want to support an individual artist making something that inspires feelings in you, personally, and visual art made for commercial purposes.
If you're a business owner and you're looking at your margins, why would you have a robust art department? In the article there's a sample of alternative logos as compared to the real logo, and while maybe they're not quite as polished, a few more rounds of generation could easily turn up something good enough, especially as compared to the time and expense of a human putting their MFA to work on it. The only loss to having a computer generate it is the way that it lacks background or context to link it to personal experience. Setting aside the economic reasons to sell art, humans makes art because it conveys something about themselves or their feelings. Nothing DALL-E does can have that, and so it'll split the market even more fully into commercial and personal art.
If nothing else you could probably reduce the number of hired artists. Instead of a company having, say, a team of 5 artists, with 2 senior artists and 3 junior artists, they can just have the 2...
If nothing else you could probably reduce the number of hired artists. Instead of a company having, say, a team of 5 artists, with 2 senior artists and 3 junior artists, they can just have the 2 senior artists, and a DALL-E subscription. You still need some expertise, and a human decision maker at the end, and touch ups to the final product...
but just as the advent of Excel meant that you can replace 10 accountants with paper books to 1 accountant with an Excel spreadsheet, you can reduce a lot of the labor that can only be done in volume with computerized alternatives.
Yeah, and accounting is a good comparison. I've been in the industry for more than a decade, and automation is functionally pulling up the ladder when it comes to pathways into the industry for...
Yeah, and accounting is a good comparison. I've been in the industry for more than a decade, and automation is functionally pulling up the ladder when it comes to pathways into the industry for people without 6+ years of training. The low end stuff that used to require a lot of headcount is automated now and there's still plenty of trained personnel from before, but the senior people will age out one way or another and then what will you do? Depend on people who have more training but less experience, mostly, and that comes with its own challenges. Both in increased required education on the part of the employees, who have to spend even more before they can get a job, and reluctance to hire people for that senior level in the absence of experience. But when the years of work to get that experience have been wholesale removed by the implementation of automation, there's no functional pathway left.
Sorry. I've got a lot of feelings about this problem and what it'll do to us all in the next few decades.
It's so fun to ask DALL-E to play out your imagination, but its actual usefulness seems closer to the suggestion at the end that it be used more as inspiration for human art.
It's so fun to ask DALL-E to play out your imagination, but its actual usefulness seems closer to the suggestion at the end that it be used more as inspiration for human art.
I think it could be pretty useful for the "stock image" market, which is fairly large. Y'know those seemingly obligatory images that go with every article, no matter how banal. For example, an...
I think it could be pretty useful for the "stock image" market, which is fairly large. Y'know those seemingly obligatory images that go with every article, no matter how banal. For example, an article on some trivial iOS bug would have a stock photo of someone looking frustrated at their iPhone.
Well, now you can type "person looking frustrated at iPhone" and get "tailor made" pointless article images practically for free. Unlike stock photos, these can be in cartoonish style, or something wild like medieval japanese inkbrush, not like anyone actually cares about the image here, it just needs to break up the text visually.
You joke, but how do you know that isn't already happening? Review top 10k Google searches per language and build a basic website framework using the search terms as the prompts. Put some ads up....
You joke, but how do you know that isn't already happening? Review top 10k Google searches per language and build a basic website framework using the search terms as the prompts. Put some ads up. Given the low cost per site, it'd probably pay for itself. Of course, it'd be corrosive as far as the value of an internet search is concerned, but it's free-ish money.
Oh the jest wasn't on websites wanting to so that, but websites that already do that or websites that care so little their manual work is barely distinguishable from auto generated crap. Think...
Oh the jest wasn't on websites wanting to so that, but websites that already do that or websites that care so little their manual work is barely distinguishable from auto generated crap. Think Twitter and its bots shouting at bots or the 'New Media' crapshoot.
I dunno. If someone was wanting art made cheap, I don't see why they wouldn't pay DALL-E or one of the other art generation programs instead of paying an actual human artist. Generating a dozen...
I dunno. If someone was wanting art made cheap, I don't see why they wouldn't pay DALL-E or one of the other art generation programs instead of paying an actual human artist. Generating a dozen variations to pick from is a matter of minutes, where a human could take days and hundreds of dollars. Best case, they pay someone to tweak it slightly so it could be trademarked/copywritten. Grim case, that sort of use case is written into the agreement for use of the software package, you pay an additional fee, and no human ever touches it.
Not something I could see having any real use until AI is much more advanced. For all the funny meme images it's able to create, most of it can be firmly placed in "Lovecraftian horrors that...
Not something I could see having any real use until AI is much more advanced. For all the funny meme images it's able to create, most of it can be firmly placed in "Lovecraftian horrors that mostly resemble the characters/things you specified." I can't see it being of any use to anyone unless they wanted free abstract art
Really? Maybe you're thinking of ] DALL-E mini,a free but heavily scaled back clone responsible for those grids of generated images with the prompt people have been posting for fun. The actual...
Really? Maybe you're thinking of ] DALL-E mini,a free but heavily scaled back clone responsible for those grids of generated images with the prompt people have been posting for fun. The actual DALL•E 2 is imo extremely impressive and many of the generated images are completely plausible if not indistinguishable from reality. The instagram account @neural9000 has a bunch of examples in various styles (and the ones that look like pictures aren't actually pictures they are AI-generated).
DALL-E has serious limitations, like being really bad at recognising and drawing human faces. Cartoons and anime are less bad. Someone put "anime girl jesus" in DALL-E and it produced some very...
DALL-E has serious limitations, like being really bad at recognising and drawing human faces. Cartoons and anime are less bad. Someone put "anime girl jesus" in DALL-E and it produced some very accurate images.
Honestly, it has more meme potential than anything else. I've seen a lot of funny shit in political meme groups on Facebook generated using AI.
I believe this is intentional as iirc OpenAI tried to exclude faces from the training data in order to limit nefarious uses of the AI. on their website they say "Preventing Harmful Generations...
DALL-E has serious limitations, like being really bad at recognising and drawing human faces.
I believe this is intentional as iirc OpenAI tried to exclude faces from the training data in order to limit nefarious uses of the AI.
We’ve limited the ability for DALL·E 2 to generate violent, hate, or adult images. By removing the most explicit content from the training data, we minimized DALL·E 2’s exposure to these concepts. We also used advanced techniques to prevent photorealistic generations of real individuals’ faces, including those of public figures."
Practically, this sort of thing isn't a creative revolution, but the destruction of economically viable human creativity. Great, it can make art in whatever style you want. It can insert Kermit the Frog into any movie you want. If what you want is the image equivalent of the 30'th Marvel movie, it can churn out the memes and regurgitate the most popular stuff ad nauseam. This feels like an inflection point between the visual art paid for because you want to support an individual artist making something that inspires feelings in you, personally, and visual art made for commercial purposes.
If you're a business owner and you're looking at your margins, why would you have a robust art department? In the article there's a sample of alternative logos as compared to the real logo, and while maybe they're not quite as polished, a few more rounds of generation could easily turn up something good enough, especially as compared to the time and expense of a human putting their MFA to work on it. The only loss to having a computer generate it is the way that it lacks background or context to link it to personal experience. Setting aside the economic reasons to sell art, humans makes art because it conveys something about themselves or their feelings. Nothing DALL-E does can have that, and so it'll split the market even more fully into commercial and personal art.
If nothing else you could probably reduce the number of hired artists. Instead of a company having, say, a team of 5 artists, with 2 senior artists and 3 junior artists, they can just have the 2 senior artists, and a DALL-E subscription. You still need some expertise, and a human decision maker at the end, and touch ups to the final product...
but just as the advent of Excel meant that you can replace 10 accountants with paper books to 1 accountant with an Excel spreadsheet, you can reduce a lot of the labor that can only be done in volume with computerized alternatives.
Yeah, and accounting is a good comparison. I've been in the industry for more than a decade, and automation is functionally pulling up the ladder when it comes to pathways into the industry for people without 6+ years of training. The low end stuff that used to require a lot of headcount is automated now and there's still plenty of trained personnel from before, but the senior people will age out one way or another and then what will you do? Depend on people who have more training but less experience, mostly, and that comes with its own challenges. Both in increased required education on the part of the employees, who have to spend even more before they can get a job, and reluctance to hire people for that senior level in the absence of experience. But when the years of work to get that experience have been wholesale removed by the implementation of automation, there's no functional pathway left.
Sorry. I've got a lot of feelings about this problem and what it'll do to us all in the next few decades.
It's so fun to ask DALL-E to play out your imagination, but its actual usefulness seems closer to the suggestion at the end that it be used more as inspiration for human art.
I think it could be pretty useful for the "stock image" market, which is fairly large. Y'know those seemingly obligatory images that go with every article, no matter how banal. For example, an article on some trivial iOS bug would have a stock photo of someone looking frustrated at their iPhone.
Well, now you can type "person looking frustrated at iPhone" and get "tailor made" pointless article images practically for free. Unlike stock photos, these can be in cartoonish style, or something wild like medieval japanese inkbrush, not like anyone actually cares about the image here, it just needs to break up the text visually.
I demo'd this exact use case to a startup working with GPT-3. I think we'll see DALL-E images along side GPT-3 blogs before too long.
Why stop there? Auto generate stuff with GPT-3, auto generate image with DALL-E, auto generate website code through Copilot, ???, Profit!!
/s
You joke, but how do you know that isn't already happening? Review top 10k Google searches per language and build a basic website framework using the search terms as the prompts. Put some ads up. Given the low cost per site, it'd probably pay for itself. Of course, it'd be corrosive as far as the value of an internet search is concerned, but it's free-ish money.
Oh the jest wasn't on websites wanting to so that, but websites that already do that or websites that care so little their manual work is barely distinguishable from auto generated crap. Think Twitter and its bots shouting at bots or the 'New Media' crapshoot.
I dunno. If someone was wanting art made cheap, I don't see why they wouldn't pay DALL-E or one of the other art generation programs instead of paying an actual human artist. Generating a dozen variations to pick from is a matter of minutes, where a human could take days and hundreds of dollars. Best case, they pay someone to tweak it slightly so it could be trademarked/copywritten. Grim case, that sort of use case is written into the agreement for use of the software package, you pay an additional fee, and no human ever touches it.
Not something I could see having any real use until AI is much more advanced. For all the funny meme images it's able to create, most of it can be firmly placed in "Lovecraftian horrors that mostly resemble the characters/things you specified." I can't see it being of any use to anyone unless they wanted free abstract art
Really? Maybe you're thinking of ]
DALL-E mini,a free but heavily scaled back clone responsible for those grids of generated images with the prompt people have been posting for fun. The actual DALL•E 2 is imo extremely impressive and many of the generated images are completely plausible if not indistinguishable from reality. The instagram account @neural9000 has a bunch of examples in various styles (and the ones that look like pictures aren't actually pictures they are AI-generated).
DALL-E has serious limitations, like being really bad at recognising and drawing human faces. Cartoons and anime are less bad. Someone put "anime girl jesus" in DALL-E and it produced some very accurate images.
Honestly, it has more meme potential than anything else. I've seen a lot of funny shit in political meme groups on Facebook generated using AI.
I believe this is intentional as iirc OpenAI tried to exclude faces from the training data in order to limit nefarious uses of the AI.
on their website they say
"Preventing Harmful Generations
We’ve limited the ability for DALL·E 2 to generate violent, hate, or adult images. By removing the most explicit content from the training data, we minimized DALL·E 2’s exposure to these concepts. We also used advanced techniques to prevent photorealistic generations of real individuals’ faces, including those of public figures."
Probably a good thing. I don't even want to imagine the kind of things people would be able to deepfake without these limitations.