I'm curious as to how an AI might handle "Francine Voortrek, laying corpse-like in a coma". Will it auto-correct "laying" to "lying", which I assume is what Caplan meant? Or will it try to render her as a comatose chicken?
I doubt I'll attempt this- it sounds like a lot of work for very little chance of pay-off (I might have a go at one chip just as an experiment), but a few thoughts:
* I strongly suspect that, for the moment, if you find an artist whose work you like, and money is no object, a human artist is still superior for this sort of task, for the moment.
* That said, I also suspect that, for most people, given the cost of hiring a human artist, AI art generators would typically be the way to go.
* In working this out, one ought to factor in the time getting suitable work out of the AI, which could be decidedly non-negligible.
* I have done vaguely similar (but somewhat simpler) projects myself with AI art generators, and I was mostly happy with the results. These were using previous-gen models, as well (DALL-E-3, for example); the latest models are materially better.
* It's not entirely clear whether you insist that the art generator do *all* of the work. Would it be OK if the image came from the art generator, but the lettering and border were added manually?
* This is a difficult task, as it requires the ability to keep consistent style, use multiple reference images in specific ways, and (debatably) getting text correct and in correct positions with consistent layout. Oh, and the art has to be really good as well. The best models at composition (GPT-4o and GPT-5) can do most of this, but GPT-4o at least doesn't produce very pretty results (I haven't really played around much with GPT-5 yet). Meanwhile the models that are best at producing very pretty art (Midjourney, say) cannot do composition at anywhere near this level.
One time I don't want to restack an article I like
I'm curious as to how an AI might handle "Francine Voortrek, laying corpse-like in a coma". Will it auto-correct "laying" to "lying", which I assume is what Caplan meant? Or will it try to render her as a comatose chicken?
Wait, are your twin sons already in graduate school? Or are they doing undergrad?
They start their Ph.D.s tomorrow!
Good lord, but that makes me feel older than anything for a while :)
Congratulations!
So how did the experiment turn out?
I doubt I'll attempt this- it sounds like a lot of work for very little chance of pay-off (I might have a go at one chip just as an experiment), but a few thoughts:
* I strongly suspect that, for the moment, if you find an artist whose work you like, and money is no object, a human artist is still superior for this sort of task, for the moment.
* That said, I also suspect that, for most people, given the cost of hiring a human artist, AI art generators would typically be the way to go.
* In working this out, one ought to factor in the time getting suitable work out of the AI, which could be decidedly non-negligible.
* I have done vaguely similar (but somewhat simpler) projects myself with AI art generators, and I was mostly happy with the results. These were using previous-gen models, as well (DALL-E-3, for example); the latest models are materially better.
* It's not entirely clear whether you insist that the art generator do *all* of the work. Would it be OK if the image came from the art generator, but the lettering and border were added manually?
* This is a difficult task, as it requires the ability to keep consistent style, use multiple reference images in specific ways, and (debatably) getting text correct and in correct positions with consistent layout. Oh, and the art has to be really good as well. The best models at composition (GPT-4o and GPT-5) can do most of this, but GPT-4o at least doesn't produce very pretty results (I haven't really played around much with GPT-5 yet). Meanwhile the models that are best at producing very pretty art (Midjourney, say) cannot do composition at anywhere near this level.
I’d be fine with combining AI with manual lettering.