By now you have probably heard of Dall-E and Midjourney, etc.

I have been watching the images coming out of the advanced AIs and I’m pretty impressed. Of course there are problems with some of the outputs, and it’s an iterative process between the AI and the human directing it – but, damn.

I’m not super thrilled about the interface, but I suppose it works – it uses Discord. So, you send a request to a ‘bot, which then posts its reply inline. In the “beginner” channel, you’re squawking out your orders along with dozens of other people (or maybe ‘bots) and it scrolls pretty fast. When you go into a private chat with the AI then you can have more of a conversation.

First off: the AI does a really good job of understanding what you want.

It presents you a few candidate images and you can ask for successive “takes” on any one of them and it’ll iterate a few more versions. Then when you’re happy you can re-render a final version. The discord interface also allows the images to be updated as they are developed, so you get successive iterations of the basic composition.

Of course it’s not perfect. But it’s still amazing. Think about it this way: last year, AI artists were creating pretty mediocre sketches, but now that they have absorbed all of human art and had a few million subjective years to study it, there has been some improvement.

I’m most impressed by the model’s language skills. I experimented with the text-only version before I was able to get into the image generation model, and had it write me stories and poems. Some of them were better than I have seen coming out of today’s “journalists”, which makes me wonder if any of them actually bother to write, anymore.

It’s less good at specific stuff but when a user decides to really drill into a particular look, it can get amazing.

Charles Darwin Riding a Velociraptor in the style of Rembrandt

There are a few aspects of the system I think are flat-out stupid. Apparently the engineers thought they would protect the world from false imagery, so they have a list of concepts that they block. I.e.: “Donald Trump getting peed on by hookers in a Russian Hotel” results in a threat to close my account. So did “Voltaire snorting a line of cocaine off a copy of Newton’s Principia”

It invites infinite farting-around, though:

“The mist of wind”

Part of what I enjoy is that it never gets annoyed by increasingly control-freakish requests:


“An ancient souk on an abandoned planet full of debris and junk”

For some reason, it seems to do exceptionally well when asked for “concept art”:

concept art for 1900s glass mansion, steampunk style

You can play with this all day. I nearly did.

giant robot samurai with infantry march across a snow plane – in the distance the sky is burning

This has gotta be burning some cycles.


  1. Reginald Selkirk says

    Very interesting. There are some aspects of the Darwin image I do not understand. There appears to be a wheeled vehicle – a bicycle? – at the lower right. Are there multiple velociraptors? There appears to be two tails. Or is it three? Is the dark shape extending to the left a shadow, or another velociraptor head?
    I have to suspect the A.I. does not know how a velociraptor is constructed.

  2. says

    @Reginald Selkirk:
    I puzzled over that, too, until it occurred to me that we are seeing a mix of “velociraptor” and “velocipede”.

  3. Dennis K says

    Neat as it is, it’s also bittersweet, this trend to replace all human relevancy. Can’t even do art anymore to distinguish me from a machine.

  4. sonofrojblake says

    Visual art is all very well, and on one level these images are impressive. But as for “replacing human relevancy”? Call me back when an AI can write a tight five minutes for an open mike slot at the Comedy Store. Not just string together some jokes it looked up on t’internet – an actual original comedy routine, with a structure. I won’t hold my breath.

  5. says

    I feel the same as you, but with a slight twist – I don’t see these things as challenges to human creativity: they free it. Imagine if all bullshit “news” was written by AI. Apparently most citydesk articles already are. It devalues the corporate crap to zero and it makes an artist of, say, Caravaggio’s stature, even more notable. We just ignore the corporate crap, as always.

    Some photographers used to complain that photoshop was devaluing their art. My observation was that their problem was that their art already had no value, if photoshop threatened it. If an AI novelist threatens a writer’s screed, they needed to stop writing Harry Potter, anyway, etc.

  6. Dennis K says

    So “AI” is somehow limited? As long as we keep our creative engines revved up we can stay ahead? As far as we know, humans are just biochemical machines themselves. What stops silicon from winning the same race? I don’t see a limitation that a machine with enough flops and programming couldn’t come up with every creative pathway of a typical human brain — including the ability to judge a thing as art or believable fiction, neither of which can be expressed quantitatively anyway.

  7. says

    @Dennis K:
    I think it’s axiomatic that anything one machine can do, can be done by another. At the speed with which these systems are improving, I expect machine DaVinci is booting up on some cluster or other.

  8. says

    Also, this pretty much addresses the question of machine creativity. The image of the wind trees was suggested by the AI. Creativity appears to not be an exclusive attribute of humans. Nor is language.

  9. consciousness razor says

    There are a few aspects of the system I think are flat-out stupid.

    I’m sure I could list more than a few.

    Apparently the engineers thought they would protect the world from false imagery, so they have a list of concepts that they block. I.e.: “Donald Trump getting peed on by hookers in a Russian Hotel” results in a threat to close my account. So did “Voltaire snorting a line of cocaine off a copy of Newton’s Principia”

    There it is, the divine hand of creativity in action.

    It’s funny. If you want to commission a (no doubt very tasteful) sculpture of Trump pissing on hookers, maybe an entire multimedia experience so the kids can really enjoy it, then you don’t actually need to ask an AI to do it of course. I’m sure someone out there would be willing to do that, possibly even for free … or maybe it would be to make another one, since I bet shit like that is already out there. (What’s that you say about “originality”? When did we ever exhibit that?)

    The problems seem to arrive if we act like people need to compete with AIs, at a technical or creative level, economically, or in some way. But who says you need to go out and consume stuff produced by an AI? Go to somebody who will provide what you’re really looking for. If you’re already worried about it, then it seems like you have nothing to worry about.

    Do you think that when I write a piece of music that was commissioned by someone else, I’m forced to do a bunch of black-box calculations based on a one-sentence prompt I don’t understand? Because that’s not how it works at all. I can communicate with clients intelligently, comprehend any feedback they may have afterward, and do appropriate revisions until everyone’s happy (or at least until I’m tired of it and/or they’re not paying enough).

    If such fine people don’t know better than I do, since the customer is not always right, and try to ask for shit that just won’t work or is literally impossible, which does happen, I can tell them as much (diplomatically) or simply refuse to make some changes if I think it’s important enough. Actually, when it comes down to it, I don’t even need their input or to work with them at all, if I don’t feel like it, although it is generally better to adopt a collaborative type of attitude and to accept a lot of work even if I don’t believe it’s the greatest thing ever….

    But are you collaborating with AIs in that way? No, you’re not. However, is that more or less the way you should approach the creative process, if you’re going to involve yourself in it (as you are in fact doing)? Yes.

    So, where’s the problem? I mean, if you do prefer to work with dumb AIs, then more power to you. If that offers you something, okay. I don’t care. I don’t want to be involved in that kind of slave work anyway. If that’s what you think is required of “artists,” I’ll be working with something else who doesn’t think like that. Just make sure that if do go with the other option they’re not actually sentient AIs, because then it would just be plain old slavery.

    Why would you want them to be sentient or whatever, you ask? Can’t you cut corners and do without that? Well, do you know anything about what art is or how it gets made (by people)? That’s why.

  10. Tethys says

    I had not heard of either of these things, and find the art they make is entirely derivative and lacking any artistry. I’m sure I’ve seen those night scenes of illuminated stained glass windows in some famous painters body of work. I call this graphic design rather than art. (But I also don’t consider many of my attempts at making ‘Art’ to rise to that sublime level above craft)

    Those Samurai’s robots have a striking similarity to giant Daleks. I suppose the AI could be a Dr Who fan.

  11. says

    This suggests all sorts of experiments, which I’m sure are being attempted as we speak or have been attempted already. Enter typos and see how that changes output. Push it into gibberish iteratively: “The chartreuse dog wants fritos” – “Teh charntruse dog want fritoes” – “Techtr3us d0dwan frIIIto” – “Tec321us d0d0fri11105.” Give it bad feedback on sensible prompts to see how warped it can get. Program a bunch of bots to give it one bad feedback as an attack on its abilities, causing everyone’s outputs to get worse. Program one to give itself its own prompts and own feedback, so that free from human input it can degenerate, and see what images and words that chaos creates.

    As an artist I see these as an excellent time-saving tool. Say I had the Caravaggio assignment, I give the prompts, I get the best result I can. It’s still a fucken mess so I clean it up to make it actually work. The only downside is that these bots are producing this content by rehashing from digital information of bazillions of images – many under copyright – and the more specific the prompt the more likely you are to accidentally stumble into copyright violation.

    That’s for commercial artist use. For fine art, this swings wide the gate of a vast frontier of formalist experimentation, and also throws out the gatekeeping of education, so regular people can exercise their aesthetic skills without being shut down by their weaknesses. Not just AI, there are a lot of tools right now that can allow people with hidden talents to come out of their shells. The upside of the cyberpunk dystopia.

  12. consciousness razor says

    As an artist I see these as an excellent time-saving tool.

    Yeah, it’s like a calculator, as opposed to a mathematician. It’s not a replacement for the mathematician.

  13. Tethys says

    I’m sure many artists would use this as a tool, but it’s still just taking actual artwork from various actual artists and pixelating it into graphic renderings.

    They have no vitality or expression. Many of them are creepy in an uncanny valley way, even though I usually enjoy surrealism. Those melted off hands in the ‘style of Caravaggio’ are horrific.

  14. consciousness razor says

    I’m sure many artists would use this as a tool, but it’s still just taking actual artwork from various actual artists and pixelating it into graphic renderings.

    You can say the same thing about sampling in music: you’re “just” taking little bits of a recording by an “actual” performer (possibly a different person from you, possibly not).*

    But saying that would be misleading. There’s really a lot more to it than that for someone who’s good at their craft, and it’s not like it constitutes the majority of the process (much less the whole thing). If you know what you’re doing, sampling can sound fantastic and allows a lot of powerful techniques that just didn’t exist before. Also, the end result may not be very much like the original at all. It could be totally unrecognizable.

    So, a lot of the assumptions people brought with them were simply wrong. The fact is, when you give a real artist tools like that, they’re still going to act like a real artist when they use them, and that’s where all the magic happens. The important and interesting parts of the creative process, which you might have thought had to go out the window, don’t in fact have to go out the window.

    Think of what you might say about some other technology, like a piano, which was once new: “it’s like a string instrument, but all you gotta do is push a button?” No, that’s not really all there is to it. People understand that because we’ve actually seen pianists in action, and many of them are of course pretty fucking incredible. In this case, though, it’s still brand new, so we don’t have much of anything to work with yet. We’ll have to see how it goes.

    * Indeed, that’s exactly what a lot of people did say in the 80s and after, once hip hop musicians started using sampling a lot. It must have either slipped under their radar or just didn’t bother them as much, when other musicians were doing it.

  15. Tethys says

    It is somewhat analogous to music sampling (Ice, Ice, Baby), but in this instance there is no creative artist who is combining their chosen samples into an original artwork.

    All art is derivative. However, painting and drawing also require a great deal of technical skill and practice in order to achieve the inspired original creations which we call Art, rather than architectural drafting, or graphic design, or illustration.

    I can’t quantify exactly what is missing from these renderings any more than I can say what makes something art. Art is subjective, but IME it evokes an emotional response in the viewer?

    I’m mostly distracted by trying to place the original art. I see echoes of Camille Pisarro’s depictions of Paris streets and Lesser Ury’ s Berlin paintings in those nighttime concept works with windows and reflections.

  16. consciousness razor says

    It is somewhat analogous to music sampling (Ice, Ice, Baby), but in this instance there is no creative artist who is combining their chosen samples into an original artwork.

    Not yet. In this instance, Marcus is just posting them, with nothing fancy along the lines of producing a new work which uses this stuff as only one step in a much more extensive process. But it could be otherwise.

    If people are just going to grab some free/cheap, quick, AI-generated images and use them for memes and shit, then it’s the slave labor that I was talking about before. (That’s already not a good market for the artists, so nothing extra to worry about there. But if some weren’t already worried about the effects of capitalism on artists, they should go ahead and started worrying.) Maybe in some cases that will be an economical way to go, if it can scale up to meet demand. I expect at least some of it to come with a price tag, perhaps if you want all the bells and whistles or unlimited access or whatever. But it’s incredibly fast, that’s for sure. If all you really want is a cheap, disposable, fast-food sort of image that can be consumed by the masses, it might suffice.

    But if it’s just one of many tools in the toolbox used by artists, that is something else.

    I would bet on a mix of both, much confusion and gnashing of teeth, and of course capitalist exploitation as usual. (I don’t expect sentient AIs in my lifetime, but definitely more lies about it.)

  17. lochaber says

    I can stick a photo in a photocopier, and copy it. A more skilled person than I can draw/paint/whatever an impressively similar image to that in the photo. And, some of them can do it from the original subject matter, without the aid of a camera.

    I’m not a humanities person. I think looking at visual art can be pretty neat sometimes, and there is a lot of fiction I enjoy reading, and a lot of music I like listening to. I don’t have the background (or even the desire) to be able to adequately critique it, just on a visceral(?) level, I enjoy being able to experience the artwork of others.

    I also enjoy looking at weathered rock formations, weird old trees, and patterns in broken glass. I like listening to birdsong, or a nice thunderstorm (preferably from a sheltered perspective…)

    I don’t think artificial intelligence is going to eliminate human artists. Long before it gets to that capacity (not that I think it will, but for the sake of argument…), it will also have eliminated the need for much of human labor. And, if we haven’t irrevocably destroyed ourselves, our habitat, or our psyche, by then, I think it could be a major stepping stone to a Utopia, like something out of Star Trek or The Culture…

    And, even if it did eventually eliminate human artists, wouldn’t it be better to be a bunch of weird apes lounging around admiring weird computer art in comfort and security, than the fucked up weird apes we are, brutalizing and oppressing each other without reason or end, but, hey, at least we are admiring “human” art?

  18. outis says

    Well it IS interesting, I just cannot see what developments it will have for the arts in the future.
    Maybe some aspects of commercial art will be automated, with illustrations generated by PC and no longer needing a human hand. Think black velvet paintings or illos for an operating manual.
    Meanwhile, artists at a higher level may wish to use something similar as a sketching pad/agent, saving a lot of spadework while composing an image.
    And at the highest level it’s likely nothing will change, except what true innovators and pioneers invent will be incorporated in the next generation of AI.
    I suppose we’ll wait and see…

  19. says

    The request is:
    sam harris standing by the lever that switches a trolley car from one track to another. The trolley care is victorian style with smoke billowing from the engine. realistic 8k cross hatched sketch style

  20. says

    I can’t wait ’till someone comes up with an “enhance this picture” mode: you upload a picture then tell the AI what changes you want. e.g.: upload a picture of PZ and ask the AI to add tentacles.

Leave a Reply