What do you think of AI art generators like Dall-E?

[CSJones]

I don't think anyone will be paying any artists no matter what tier when this becomes mainstream and efficient enough.

Yeah. There are too many responses to get to all of them, and although some like @DokiDokiTsuna present great counterpoints, I think many of them are looking at it from a standpoint of where the tech is today—not where it'll be in 10 years. Who knows what kind of godlike VR worldbuilding powers AI could have by then: Perfectly imitating someone's art style could be nothing.

[Legendofgenii]

Do they just grab images off Google?

[CSJones]

I think the current ones do, yeah. Or uploaded images.

[NickRowler]

I have seen some Dall-E 2 images that do look nice but overall I feel like people have mostly just used it for memes.

I do think it could be a useful tool for some people. I was thinking how I am not very good at making backgrounds and having a program that generates random ones which would be public domain could be helpful. CGI artist do use algorithms to help generate 3D backgrounds and other assets like hair. Trees in 3D spaces are randomly generated.

I could also see AI images helping artist get ideas. Like I would not be surprised if down the line, there is an art challenge where people are redrawing or adding onto AI images

[TedGravesArt]

They don't make me nervous at all. AI generators are fascinating though.

There are certain things that robots have a hard time doing that humans do effortlessly, that we don't even think about. Grabbing something without looking at it is something we do all the time, but is extremely hard for a droid. Similarly, imagining things is something we do passively, even when we're unconscious, but that requires a huge amount of effort for a robot to attempt to replicate. When you draw or paint, you probably don't even realize the depth of comprehension required to do that job, because it just comes so naturally to you.
Dalle and similar AI image generators are impressive and promising, but the fact remains that they are built for a single purpose, which inherently limits their creativity, and they pull from a database (albeit a very large database) of existing images for training models. These bots repurpose the composition, color, and style choices from humans, they don't invent their own. While what they produce may seem fairly competent, they're only copying the pixel patterns of something they've seen before. They have no concept of anatomy, or exactly what a hand is, or emotion. They can't deliver a subtle performance through art, or extrapolate a scene in 3d from previous real life experience, then redraw it from another angle. That's not the goal of the programmers. But these are the kinds of skills that are required of artists to do our job.
AI images are fascinating in that they look right at first glance, but the second you try to resolve details about what you're looking at it all falls apart. Look at this one:

Teddy bears doing chemistry, right? But is it...? What exactly is that thing he's holding? In fact what are any of those things they're working with? Is the bear on the right wearing a tie or...wtf is that? An ascot? And what's in his pocket?

AI art is really cool in that it can create images that completely fool your brain into thinking they're describing actual objects and shapes, but the closer you look the more you feel like you're tripping out. It has zero actual concept of what it's creating, it's just using keywords with associated existing images. Here's another example:

Something as simple as that is just too hard for it. I'm sure it has complete images of stop signs in it's database, but it doesn't understand the simple concept of text or language. Even just asking it to write any kind of text on something is too specific for it.

So no, this will not replace artists, because it's not designed to have our skill set. We have real brains, this is just a photobash machine. Impressive for sure, but very simple comparatively.

All that said though, I am interested in it as a tool. If you get stuck concepting a thing, maybe it can photobash some stuff together to spark some kind of idea. That's very compelling!

[TheLemmaLlama]

Honestly, even if the technology advances to the point where perfectly imitating someone's art style becomes trivial, I'm all for it. I got into art because I need a vehicle through which to execute my vision, and if I could just press a button and have a fully illustrated comic in front of me that perfectly matches my vision, I'd do it in a heartbeat. This is literally exactly what I wanted all along XD

Many, perhaps even most people who commission artists (or at least the big companies hiring an animation team etc) don't see you as individual craftsmen with your own artisanal flair. They're hiring you as grunt workers to execute their vision exactly according to their design. If a machine could do this instead (and we somehow fix the issue of letting only a few people own all the machines and keep all the fruits of their labor), it frees up us actual humans to focus on our vision.

This is not a world where no-one needs to create anymore. This is a world where we don't need to waste time on the tedious, rote aspects of creation so we can focus on true creation, the parts only a human can do. If you're born to create, you can now create more than was ever possible with your one puny human body and its carpal-tunnel-ridden wrist.

I guess though many artists aren't like me and actually value the craftsmanship of art and not just the vision. Yes, the finished product of a machine would be indistinguishable from the finished product created with your own two hands, and if that sucks to you, who am I to say your feelings are invalid? I guess if machine art becomes the norm, there will be an increased demand for live illustration viewings/speedpaints, where people see your actual process since it would be so rare in those days. I'm not sure if that's any comfort for you. I can't claim to understand how you feel. I'm just saying what I think, since that's what this thread is about :'D

[DokiDokiTsuna]

I got into art because I need a vehicle through which to execute my vision, and if I could just press a button and have a fully illustrated comic in front of me that perfectly matches my vision, I'd do it in a heartbeat. This is literally exactly what I wanted all along XD

Somehow I never thought about it from this angle, even though I 1000% agree-- if I could someday just train an AI on my artstyle and get it to draw comics for me, that would take nearly all of the stress out my artistic life. ^^ I'd be free to do fun art and write more stories while a computer does all the grunt work I don't wanna do. That's the dream! =D

[BoomerZ]

Just remember the smartest computer is dumber than a cockroach. Yes it can navigate a room, but change the room around and to the computer it's a completely new room. A computer understands what a chair looks like but not the concept of a chair. You literally have to feel it every type of chair or it just doesn't "get" it. It sees every different chair as a completely different object with no relation to each other. We are so far away from actual AI that its not going to happen within your lifetime.(seriously, read the hard science behind it. most of the smart computers don't actually think. they use short cuts to do a specific task.)

Also remember that mocap did not replace animation. It replaced the first rough pass. Animators still have to go in and clean it up. They add weight, anticipation and follow through. The little things that humans expect in their animation. Studios that don't do this make very stiff animation that looks unrealistic.

It's like a traced 3d model vs a Jim Lee comic drawing. Sure one might be more realistic and in proportion. BUT its boring. There will always be a place for artist that know their craft.

[allenT]

Dall-E version 2 is IMPRESSIVE.

Future concept artists are done for

edit: a new one called Google Imagen:

[CSJones]

Although writing and art are disconnected, I see whether a creator has taken the time to learn the craft or not as an early indicator of how good their work is. If you actively don't want to learn to draw, then a drawing based medium like comics isn't for you. There are plenty of great writing-based ones: novels, webnovels, web serials, scripts, etc.

And if you really are a great comic writer but have no aptitude for art - e.g. the original One Punch Man comic - you should start by publishing your writing, then try to find an artist to work with later once you've practiced and built a following. Which One did, and the result speaks for itself. Nothing against you personally, but if someone's motivation for having an AI generate a whole comic for them was that they're too lazy to learn art, I wouldn't be super inclined to trust their writing either.

[darthmongoose]

I'm fairly certain we'll see a scandal break soon that a lot of these "Dall-E" created images are actually fakes, or that a lot of the truly impressive seeming results actually came about because it found an image of pretty much exactly that thing somebody had made on Google and just put it up pretty much unedited.

My reasoning is that some of the really amazing looking results go way beyond a computer's ability to understand reality, never mind to understand abstract drawn representations of it. Think about how we can barely make a chatbot yet that can pass the Turing Test and come across like a real human in a typed conversation... you want me to believe a computer can't handle a basic conversation in English, but it can produce a perfect cartoon image of a "panda mad scientist" understanding how 2D lines are a symbolic representation of boundaries of space, and stylistic shorthand like cel shading, and that "panda mad scientist" means "a panda who dressed as the fictional archetype of a scientist who has gone mad with power" and not "a scientist who is mad about pandas" or "a scientist panda who is angry"? Nah, I'm calling bullcrap. Computers can't even reliably translate Japanese to English or tell the difference between peaches and bums, so any of these that look a little too good and like the computer understood the prompt a little too well or seemed to understand human perception of space and light, composition or aesthetics in a way a computer just doesn't, I'm highly skeptical of.

It's like those "WE FED A COMPUTER EVERY EPISODE OF FRIENDS AND HAD IT WRITE A SCRIPT" memes... those are mostly fake too. You can tell the ones that are fake because they're just a little too funny or random in a way that's pleasing to a human rather than how randomness actually occurs.

[NickRowler]

I'm fairly certain we'll see a scandal break soon that a lot of these "Dall-E" created images are actually fakes, or that a lot of the truly impressive seeming results actually came about because it found an image of pretty much exactly that thing somebody had made on Google and just put it up pretty much unedited.

Oh, honestly I thought the eventual scandal would be data mining.

I guess this also goes into why people like using it for memes. It's good at recreating images of things that there are already a lot of examples of. That is why it is more efficient at recreating characters from Star Wars or Homer Simpson than things like signs for business like Waffle House. While there is a good number of reference for Waffle House signs, there are not really enough consistent images for machine learning to develop a pattern.

[MartaTod]

This doesn't worry me. Until someone comes up with an AI that can execute every whim of a client while having the ability to read what it is that they are actually asking for plus process what they're asking for through artistic terms to give them what they actually need - no human is getting replaced.

If what I just wrote sounds confusing, here's an example. There are logo generators out there. And yet, clients like to hire designers to make logos. Because:

1. Clients ultimately don't know what to do with generators. And generators can't say "the client commented they want this bolder but my skill, knowledge and experience tells me that will be crap so instead of bolding it I will accent it by making it a different color".

2. To the clients, the art they are paying for matters. Their logo matters. Their OC matters. The comic they're hiring you to draw matters. And so they want to know that there is another person there, giving their best to create this thing that matters to them.

[darthmongoose]

I guess this also goes into why people like using it for memes. It's good at recreating images of things that there are already a lot of examples of. That is why it is more efficient at recreating characters from Star Wars or Homer Simpson than things like signs for business like Waffle House. While there is a good number of reference for Waffle House signs, there are not really enough consistent images for machine learning to develop a pattern.

Oh yeah, it's a lot of fun for making memes and stuff! Like this one where I had Dave from Homestuck face off against Darth Vader!

This is roughly what I'd expect. It's still impressive, but you can see how the computer can't... quite... do it? Like you can see how it's understood:

• Darth Vader
• Dave Strider
• Fight

So get pictures of Darth Vader fighting. Okay, plenty of images of Darth Vader fighting and Dave Strider fighting, I made it easy here. Now you just need to paste in Dave as the person fighting, which it's made a solid attempt at!

Anything that looks too perfect, like where it understands context to too human a level, assume it's faked. For example, if you write "cat wearing panties" and you get a picture of an anthropomorphic cat wearing underwear, it's definitely just found a drawing of that off google, because a computer will probably give you "woman in knickers with head of cat in place of her head" or "picture of a cat with knickers on top". The computer just doesn't have enough capacity for abstract thought for anything else right now.

[DualDragons]

Yes

[rajillustration]

I'm fairly certain we'll see a scandal break soon that a lot of these "Dall-E" created images are actually fakes,

for reals, the stuff we see actually coming off of Dall-E on twitter vs the stuff advertised is hugely different. Like one will look like wads of clay and the other will be somehow a perfect illustration? hmmmmmm

That and if they are using artists work to make Dall-E, they're gonna have to pay for that copyright down the line youknow, you can't just use whatever art you want. And like even classical art that you'd assume is copyright free is often owned by like different inheretors. It would be a mess of legal work.

[DokiDokiTsuna]

Although writing and art are disconnected, I see whether a creator has taken the time to learn the craft or not as a litmus test of how good their work is. If you actively don't want to learn to draw, then a drawing based medium like comics isn't for you. There are plenty of great writing-based ones: novels, webnovels, web serials, scripts, etc.

And if you really are a great comic writer but have no aptitude for art - e.g. the original One Punch Man comic - you should start by publishing your writing, then try to find an artist to work with later once you've practiced and built a following. Which One did, and the result speaks for itself. Nothing against you personally, but if someone's motivation for having an AI generate a whole comic for them was that they're too lazy to learn art, I wouldn't be super inclined to trust their writing either.

...I think the point of their statement wasn't that they're "too lazy" to learn to draw, it's just that drawing comics is a task they'd rather delegate than do for themselves. You're making way too many assumptions...

And anyway, it is completely, totally normal to want someone or something else to perform a task for you when you don't want to do it. The fact that you aren't interested (or even just LESS interested) in one area of creativity doesn't imply that you can't be trusted in others.

I'm willing to bet that there are tons of professional comic writers out there who are literally too lazy to learn art and hire other people to do it for them, only they don't get criticized for it because that's just called collaboration. o_O Happens all the time; an entire industry was built off the concept.

[VibrantFox]

given theyre only generating images based on whatever data their being fed as opposed to having full grasp of concepts and the ability to create with intent im not concerned

plus an ai is only as skilled or as smart as the person or people who program them and since they (the ai) lack true sentience its never gonna match up to a human the way some people think they might. it could come close maybe but i highly doubt it'll ever be 1:1

[TheLemmaLlama]

I see whether a creator has taken the time to learn the craft or not as an early indicator of how good their work is.

Well yeah, if you say "I want to make X" but never actually get out there and learn how to make X, you're not really a creator. But a dislike of doing things is not an unwillingness to do things. Honestly, I don't like writing either, so I guess I should just give up on creating altogether XD But I learn to write AND draw despite how tedious I find it, because I'm really excited about my ideas and want to make them exist. (Well okay, I do like the learning process when I take it at my own pace, it's more the rote execution of stuff I already know which I find tedious :P)

What if a comic artist does digital art instead of traditional because they're too lazy to color in a large area with a brush, or draw a similar panel from scratch twice when they can just copy-paste and make minor edits? Does that also make you less inclined to trust their writing than a traditional comic artist? :]

And if you really are a great comic writer but have no aptitude for art - e.g. the original One Punch Man comic - you should start by publishing your writing, then try to find an artist to work with later once you've practiced and built a following.

So delegating to people is okay but delegating to machines isn't? I assume what you're getting at is "if humans are willing to execute your vision, that proves your vision is good, but if you get a machine to do it, it's going to do what you say even if your writing sucks".

I get the logic behind that, at least. But also, it's easier to communicate your vision with visuals than without it, especially if you're envisioning a project with a visual component. Did the creator of the original One Punch Man comic publish a piece of pure writing first, or did they only gain traction after communicating their vision with their own attempt at the comic?

[CSJones]

I assume what you're getting at is "if humans are willing to execute your vision, that proves your vision is good, but if you get a machine to do it, it's going to do what you say even if your writing sucks".

Yeah, that's a better phrasing than I came up with, actually. I know it's not that straightforward since hiring artists is expensive, but yeah, overall I think if you can find real people to work with, it means your vision has at least reached the level of professionalism where you're willing to present it to other people and they'll agree to spend hours drawing it. Granted, it could just mean you have enough money to get artists to keep working with you no matter how bad your writing is, which is admittedly a really unfair advantage, but least it helps human artists earn a living instead of outsourcing yet another job to machines.

What I'm afraid of is if it ever reaches the point where it's impossible to tell a comic (or whatever medium) that the artist put years of work into learning to execute, from one someone dashed off a script for on their lunch break then hit "generate," by appearance. It'd be the same problem writers face: there's no barrier to entry and it's hard to tell a good novel from a bad one at first glance, so the market is flooded with an overwhelming amount of low-quality content and people tend to just ignore books by default unless it's popular or they're specifically given a recommendation.

Re: One Punch Man, I guess a better phrasing for this would be "publish what you have," not just "publish your writing by iteslf." If someone's bad at drawing and wants to do a comic themself, that's totally fine, and it might still find an audience on the story's strength - as he did - but the price is that it won't look as good as you want to. If you want a professional-looking result, you have to earn it by either learning drawing (or 3d modeling, rendered comics are valid) to a professional level, or hire one, which is how it should be, because that's how you get good at visual storytelling as well as writing. Maybe this is a reactionary standpoint, but the idea of there being a shortcut so powerful it takes away your having to earn it entirely boils my piss. It would be like if an MMO suddenly switched from starting new players off at Level 1 to 80.