5 ms·

If you take the point of view, that children’s media is only for kids, so quality doesn’t really matter. The best children’s book authors and illustrators are

If you take the point of view, that children’s media is only for kids, so quality doesn’t really matter. The best children’s book authors and illustrators are ones who see children as the most important audience of all and deserving of the most careful work.

ftxbro 4y ago
Maybe the idea is that most children's books are illustrated, compared to other books.
p-e-w 4y ago
Except that the better outputs from the better Stable Diffusion models make 99% of professional illustrators look like they're drawing stick figures by comparison. If you still believe that "AI generated = bad quality", you haven't seen some of the stuff that has popped up lately. SD is still new, so understandably it's taking a while for people to figure out how to use it properly, but some users are clearly getting the hang of it now, and the results are mind-blowing.
- riffraff 4y ago
  Can you get SD to do something like Richard Scarry's Busytown? I'm asking sincerely, if we can I'll be genuinely impressed.
  - SanderNL 4y ago
    I’ve tried once with MJ and it looks OK. I’m not even surprised anymore. The devil, though, is in the details. It’s hard to control the outputs. The scene will look 80% OK, but then you want to add some cat somewhere and that won’t be trivial (at the moment). There are already a lot of impressive tools to help you out, like ControlNet, but it’s a bit difficult still. You also have to do cleanup, because some random animal in the background of the illustration will be horribly mangled. This all basically requires you to actually be a (digital) artist to fix properly. Again, for now. Edit: I wanted to add that this, to me, is similar to AI coding. On a good day, it might be 80% OK, but you actually need to be a programmer to catch its lies and/or incompetenties.
    - riffraff 4y ago
      > This all basically requires you to actually be a (digital) artist to fix properly. Again, for now. right, this is my impression of the "graphic ai revolution": it will be a boon for actual digital artists that can take the output and refine it, much like Copilot can help programmers while not replacing them. for now, anyway.
      - p-e-w 4y ago
        My impression has been that Stable Diffusion has outclassed the vast majority of digital artists from day one. The only reason people feel otherwise is because when you look at human-made digital art online, you tend to only see the best of the best of the best. The one-in-a-thousand artists who make it to the DeviantArt or ArtStation frontpages. But most artists aren't anywhere near as good as those few geniuses, and neither are they as good as what Stable Diffusion generates. So my prediction is that 99% of digital artists are essentially obsolete today, and the remaining 1% will become obsolete some time in the next 3-5 years.
        hammyhavoc 3y ago
        "obsolete" assumes distillation of art down to output and not the value of expression and enjoyment of the process itself. To not consider this is to misunderstand creativity. See starving artists who do what they do because they love it.
        raincole 4y ago
        > today More like yesterday, to be honest.
        SanderNL 4y ago
        It's great, but I think reliability is as big a problem there as it is in coding. It's hard to control the output. I won't go into the various problems these tools have, like them ignoring my instructions if they are outside what 90% of the population is doing, because I think they can and will be solved to various degrees eventually. But the bar will raise. The quality that customers expect will start rising. Eventually the level of control and quality that is demanded will again only be adequatedly matched by people we usually call "artists". Sure, I - some scumbag programmer - can create some fancy generic shots today for a generic company blog post and for now that is impressive, but in the long term this won't cut it. Folks will want highly fine-tuned specific imagery trained on company stuff or whatever. Branding will have to match exactly. The complexity of the design will explode. Good taste and raw visual skill, not to mention creativity - being an "artist" - will win and the average dude will not have the proclivity to keep up those skills. I think it's more than holding a pencil and being good at drawing lines. I do think the raw skills of "holding a pencil" and such is becoming obsolete. That's like having good "punching card" skills. However, knowing what to "punch" is a timeless skill or at least a good part of it is and it's easy to adapt to a new era if needed. Punching code, typing code, selecting code-snippets from AI, it's different for sure, but the underlying high-level skills are adaptable and useful. I think the same holds for artists.
  - nickthegreek 4y ago
    you could if you fine tune a model of create a LoRA on busytown work. This would take under a day.
- numpad0 4y ago
  > make 99% of professional illustrators look like they're drawing stick figures by comparison. I feel the opposite to be the case — people are feeding prompts and sketches and AI generated RGBD images as base, and another RGB for style transfer source, all for just one image. If anything, it seems, it’s becoming a tool increasingly limited to professionals.
  - p-e-w 4y ago
    You can get amazing results without doing any of that, and I can promise you that most of the people posting AI art on DeviantArt etc. aren't "professionals".
  - grumbel 4y ago
    While it's true that you need to feed some source images into the AI to get good results, what you feed into them can be pretty arbitrary and low quality, as all you are really doing is providing the AI with information about positioning and composition that can't be expressed with a text prompt. It's pretty much the "How To Draw an Owl" meme with AI doing the last step.
- totetsu 4y ago
  I am not talking about visual wow factor quality. I would love link to my instagram of amazing dalle2 and SD generated images, but I like to keep my HN account clean of personal info. [1 like this] Children don't care about that, they consume images as a way of creating and expanding their inner world and their understanding the outer world. A stick figure that captures something real to life in an way a child can recognize and be excited by is much 'quality' image for children's media than a 'award winning picture of a cute robot holding a flower 4k documentary footage' descended from a random distribution of Gaussian noise. The Gernative AIs we have today don't make things like this story by Joy Cowley with these kind of illustrations.[2] [1] https://pasteboard.co/VG6XNvbeC4gA.png https://pasteboard.co/VG6XNvbeC4gA.png [2] https://www.slideserve.com/herbst/story-by-joy-cowley-illustrations-by-elizabeth-fuller https://www.slideserve.com/herbst/story-by-joy-cowley-illust...

ftxbro 4y ago

Maybe the idea is that most children's books are illustrated, compared to other books.

p-e-w 4y ago

Except that the better outputs from the better Stable Diffusion models make 99% of professional illustrators look like they're drawing stick figures by comparison. If you still believe that "AI generated = bad quality", you haven't seen some of the stuff that has popped up lately. SD is still new, so understandably it's taking a while for people to figure out how to use it properly, but some users are clearly getting the hang of it now, and the results are mind-blowing.

riffraff 4y ago
Can you get SD to do something like Richard Scarry's Busytown? I'm asking sincerely, if we can I'll be genuinely impressed.
- SanderNL 4y ago
  I’ve tried once with MJ and it looks OK. I’m not even surprised anymore. The devil, though, is in the details. It’s hard to control the outputs. The scene will look 80% OK, but then you want to add some cat somewhere and that won’t be trivial (at the moment). There are already a lot of impressive tools to help you out, like ControlNet, but it’s a bit difficult still. You also have to do cleanup, because some random animal in the background of the illustration will be horribly mangled. This all basically requires you to actually be a (digital) artist to fix properly. Again, for now. Edit: I wanted to add that this, to me, is similar to AI coding. On a good day, it might be 80% OK, but you actually need to be a programmer to catch its lies and/or incompetenties.
  - riffraff 4y ago
    > This all basically requires you to actually be a (digital) artist to fix properly. Again, for now. right, this is my impression of the "graphic ai revolution": it will be a boon for actual digital artists that can take the output and refine it, much like Copilot can help programmers while not replacing them. for now, anyway.
    - p-e-w 4y ago
      My impression has been that Stable Diffusion has outclassed the vast majority of digital artists from day one. The only reason people feel otherwise is because when you look at human-made digital art online, you tend to only see the best of the best of the best. The one-in-a-thousand artists who make it to the DeviantArt or ArtStation frontpages. But most artists aren't anywhere near as good as those few geniuses, and neither are they as good as what Stable Diffusion generates. So my prediction is that 99% of digital artists are essentially obsolete today, and the remaining 1% will become obsolete some time in the next 3-5 years.
      - hammyhavoc 3y ago
        "obsolete" assumes distillation of art down to output and not the value of expression and enjoyment of the process itself. To not consider this is to misunderstand creativity. See starving artists who do what they do because they love it.
      - raincole 4y ago
        > today More like yesterday, to be honest.
      - SanderNL 4y ago
        It's great, but I think reliability is as big a problem there as it is in coding. It's hard to control the output. I won't go into the various problems these tools have, like them ignoring my instructions if they are outside what 90% of the population is doing, because I think they can and will be solved to various degrees eventually. But the bar will raise. The quality that customers expect will start rising. Eventually the level of control and quality that is demanded will again only be adequatedly matched by people we usually call "artists". Sure, I - some scumbag programmer - can create some fancy generic shots today for a generic company blog post and for now that is impressive, but in the long term this won't cut it. Folks will want highly fine-tuned specific imagery trained on company stuff or whatever. Branding will have to match exactly. The complexity of the design will explode. Good taste and raw visual skill, not to mention creativity - being an "artist" - will win and the average dude will not have the proclivity to keep up those skills. I think it's more than holding a pencil and being good at drawing lines. I do think the raw skills of "holding a pencil" and such is becoming obsolete. That's like having good "punching card" skills. However, knowing what to "punch" is a timeless skill or at least a good part of it is and it's easy to adapt to a new era if needed. Punching code, typing code, selecting code-snippets from AI, it's different for sure, but the underlying high-level skills are adaptable and useful. I think the same holds for artists.
- nickthegreek 4y ago
  you could if you fine tune a model of create a LoRA on busytown work. This would take under a day.
numpad0 4y ago
> make 99% of professional illustrators look like they're drawing stick figures by comparison. I feel the opposite to be the case — people are feeding prompts and sketches and AI generated RGBD images as base, and another RGB for style transfer source, all for just one image. If anything, it seems, it’s becoming a tool increasingly limited to professionals.
- p-e-w 4y ago
  You can get amazing results without doing any of that, and I can promise you that most of the people posting AI art on DeviantArt etc. aren't "professionals".
- grumbel 4y ago
  While it's true that you need to feed some source images into the AI to get good results, what you feed into them can be pretty arbitrary and low quality, as all you are really doing is providing the AI with information about positioning and composition that can't be expressed with a text prompt. It's pretty much the "How To Draw an Owl" meme with AI doing the last step.
totetsu 4y ago
I am not talking about visual wow factor quality. I would love link to my instagram of amazing dalle2 and SD generated images, but I like to keep my HN account clean of personal info. [1 like this] Children don't care about that, they consume images as a way of creating and expanding their inner world and their understanding the outer world. A stick figure that captures something real to life in an way a child can recognize and be excited by is much 'quality' image for children's media than a 'award winning picture of a cute robot holding a flower 4k documentary footage' descended from a random distribution of Gaussian noise. The Gernative AIs we have today don't make things like this story by Joy Cowley with these kind of illustrations.[2] [1] https://pasteboard.co/VG6XNvbeC4gA.png https://pasteboard.co/VG6XNvbeC4gA.png [2] https://www.slideserve.com/herbst/story-by-joy-cowley-illustrations-by-elizabeth-fuller https://www.slideserve.com/herbst/story-by-joy-cowley-illust...