3 ms·

One aspect of being an artist is to get precisely what you want, down to the last detail. Currently most generated images seem to need significant touchups to

One aspect of being an artist is to get precisely what you want, down to the last detail. Currently most generated images seem to need significant touchups to be a good image. For future generators to meet the specificity requirement they will need the language to describe it.

coffee_beqn 4y ago
But for people like me I can never get what’s in my head out on paper. So is this better then me trying to explain to a third party what I want and then checking in a few days later if they got it? I think it very easily could be
- _carbyau_ 4y ago
  Better for results that don't require absolute precision maybe. A nice picture to accompany something else of substance. But if you want precisely a thing. Then using a generator is unlikely to give it to you. As a tool for inspiration or for supplying foundational material to tweak from there, it could be brilliant. But I don't think it will fully automatically supply end game content for those with specificity in mind.
forsythe_ 4y ago
I agree. This is one of the big factors that I haven't seen thoroughly addressed on this topic: no matter how capable AI tools are at interpreting broad declarative prompts there's still a wide degree of complexity in expressing detail and subtlety that I don't think they'll be able to match anytime soon, if ever. The author gives the example of needing dozens of images to accompany scenes in his book. If you feed DALL-E something like: "create a panoramic view of a grey ocean with four mountains and an ominous dark sky", then sure, it can handle that prompt fine. It'll do a great job at broadly interpreting adjectives and phrases that are unambiguous. Mass produced art for advertising/media is going to change drastically because it doesn't need precision. However, once you introduce specificity and detail the tools are going to suck. For example, telling it to generate a portrait of a specific person with subtlety in expression, body language, pose, and demeanor. It'll create something that looks aesthetically sane but not anywhere near what you had in mind. And just for some extra counter-predictions of my own: - The novelty of seeing thousands of "good enough" images in mass media is going to wear off - AI generated images will become obvious to the average person - Society will largely get tired of AI art due to the lack of novelty - The utter collapse of the art profession some are predicting is never going to come

coffee_beqn 4y ago

But for people like me I can never get what’s in my head out on paper. So is this better then me trying to explain to a third party what I want and then checking in a few days later if they got it? I think it very easily could be

_carbyau_ 4y ago
Better for results that don't require absolute precision maybe. A nice picture to accompany something else of substance. But if you want precisely a thing. Then using a generator is unlikely to give it to you. As a tool for inspiration or for supplying foundational material to tweak from there, it could be brilliant. But I don't think it will fully automatically supply end game content for those with specificity in mind.

forsythe_ 4y ago

I agree. This is one of the big factors that I haven't seen thoroughly addressed on this topic: no matter how capable AI tools are at interpreting broad declarative prompts there's still a wide degree of complexity in expressing detail and subtlety that I don't think they'll be able to match anytime soon, if ever. The author gives the example of needing dozens of images to accompany scenes in his book. If you feed DALL-E something like: "create a panoramic view of a grey ocean with four mountains and an ominous dark sky", then sure, it can handle that prompt fine. It'll do a great job at broadly interpreting adjectives and phrases that are unambiguous. Mass produced art for advertising/media is going to change drastically because it doesn't need precision. However, once you introduce specificity and detail the tools are going to suck. For example, telling it to generate a portrait of a specific person with subtlety in expression, body language, pose, and demeanor. It'll create something that looks aesthetically sane but not anywhere near what you had in mind. And just for some extra counter-predictions of my own: - The novelty of seeing thousands of "good enough" images in mass media is going to wear off - AI generated images will become obvious to the average person - Society will largely get tired of AI art due to the lack of novelty - The utter collapse of the art profession some are predicting is never going to come