DALL-E won't steal jobs. But Stable Diffusion (a publicly available model similar to DALL-E) will flood the job market once people bolt on various GUIs. E.g, https://www.reddit.com/r/StableDiffusion/comments/wyduk1/sho... is a Photoshop plugin that uses Stable Diffusion as a backend for interactive image generation. Stuff like this will make it much easier for more people to become >90th percentile illustrators (as defined today). This means the market for many digital arts degrees is going to become glutted and salaries are going to crash like a rock.
The standard could also change. It's done it before. Today's tools already let countless people do things that would've required huge expense three decades ago. We've recently been in a fairly grounded/realistic style in many types of commercial art, maybe more imaginative imagery will start to grab the eye if this more clip-art-y sort of thing becomes easier and easier. (Imaginative in the "not along the existing trends" sense, not just purely "nothing like this has ever been conceived before.")
A DIY website today can look better than the most professional one from 20 years ago. And it took away the jobs of folks who just wrote HTML to lay things out on a page. But very few people want to put "the same level of thing that anyone without any artistic talent can whip up" as their public facing website, so the question is what will the most skilled masters of the new tools end up doing to differentiate themselves. (In terms of any single particular digital artist who doesn't want to learn new tools, that's still bad, though... ;) )
This reminds me of how VisiCalc / Lotus 123 / Microsoft Office made many tasks traditionally done by a dedicated office worker semi-obsolete. A middle/upper manager no longer had to call on a tabulator to calculate figures or make a spreadsheet. Yet for all the power of these applications, the need for actual accountants never went away.
I feel like the number of secretaries and receptionists has collapsed however under the weight of everyone learning how to type, office suites, voice menus, and good-enough speech-to-text.
I don't understand the advantage everyone here is seeing. The produced artwork in that reddit thread is a digital collage. The perspective is mangled, the end result looks like a geocities page and a myspace page had a miyasaki-themed baby.
I understand it's a tech demo. The tech just patches in holes in the artwork with flat shapes that match a brushstroke style. It's Clone Stamp 2.0. Stable Diffusion doesn't understand how to integrate together images, it doesn't understand how to create images with all the techniques artists use to make a coherent image.
If you took the basic shapes of a miyasaki film, frame by frame, like a young girl with a pig, or a young girl with a dragon and fed those shapes into the filter, you wouldn't get the miyasaki film back out of the algorithm. SD flattens, freezes and arbitrarily recreates.
This isn't art, this is a frozen heart. There is a sneaking idea among the heart of our youth, that you could capture and recreate all vision in words. Isn't possible.
I wonder if this will happen to frontend and backend developers with the exact same principle: Have the bulk of the lifting done by some stable diffusion for CRUD api/frontend and test/modify edge cases.
I think your last point on salaries: This will increase salaries for those with a lot of experience but you are correct, for vast majority of digital art degrees graduating will suffer from demand plummet.
I really fear for the young Z generation, it appears the bar to entry is increasing as these AI tools automate bulk of their requirement.
It's akin to how github copilot generates a ton of boilerplate work, something that used to be delegated to junior devs.
Yep, it's eating its way upward from junior to intermediate. And even if that doesn't take away the whole Senior role, it can certainly do a good chunk of the day to day work. We haven't even begun to see the market effects of github copilot or stable diffusion yet.
That demo video is to me further confirmation that adding computers to creative activities sucks the joy out of them and makes them bland, boring and pointless.
Illustration has been reduced to dragging pictures around, using the eraser tool and typing commands in a box: artistic dystopia.
Not only does that not deserve to steal anyone’s job, it should be ejected into space and forever forgotten.
Unfortunately the tech is already here and won’t dissapear. Most people will have to become AI assisted image creators. Like you say it will make the work joyless and much more precarious. People will be required to create much more output in same amount of time. It will be more stressful and busy and the time saved will not go to them but their bosses.
AI is going to probably do that with many jobs and while everybody will point out how industrial revolution went well (we moved to knowledge work) for those who had to stay in the factories it is not so great after all. The craftsmen proud of their work became cogs in joyless process. Entirely replaceable.
If this is gonna happen with knowledge workers… great. Maybe the luddites were right.
I don’t understand what jobs people are talking about.
Making a living as an illustrator comes from a long stream of networking and brand building, that path is reserved to a crazy small minority and being talented and good at drawing is just the entry ticket to participate in the battle royale.
If you just need a nice illustration you can already catch any random guy on a drawing board and pay them nothing close to a living salary, and more like enough for two lattes at a starbucks. An AI plugin doesn’t seem to me to change much of that situation.
Right. With all due respect to illustrators my observation of the evolution of web design: since the 2000s it started with a graphic designer and an creating the web pages, THEN Bootstrap was enough with some small changes from the graphic designer.
We can criticise stable diffusion quality and/or creativity but at the end is the market and time-to-find. If people find th graphic solution in just one second, they don't need to think further except for outliers.
Another question is when will this happen for developers. I am not talking only about AI but about more robust frameworks that quickly solve a development project without coding. It can be good frameworks more than AI. Once we reach it ideas and business will be the core more than (development) execution.
> This means the market for many digital arts degrees is going to become glutted and salaries are going to crash like a rock.
Art degrees are notoriously undervalued... There are jokes about it going far back in time.
Social Media has also done a lot of damage to art-based professions and opportunity therein over time, as artists are required to endlessly share their work for free in order to promote it and stay relevant... Actually social media has devalued art probably far beyond what AI could do from what I can see.
Most of the AI image generation services "sample" real objects from undisclosed sources, and then do most of their work to cover up or alter the sampled object. There will eventually be issues of duplicated works and copyright infringement that will take it all down a notch if you ask me.
Just like in the music industry now, many people are finding samples within the footprints of a lot of works that trigger content ID issues. When this happens on a much larger scale, the only music of value comes from human artists that can create completely unique works with keen composition that is aware of human wants and needs. If you ask me, AI is still pretty far away from being able to do that right now. From what I can observe, there are still humans behind each "Ai driven tool" that are actively tweaking the nuances of AI to make it look like it is more "sentient"... I'll only begin to worry about AI when it runs with all hands off, and when it can update and develop itself (completely on it's own). We keep redefining the term, but I think something that shouldn't be co-mingled with technology advancement driven by constant human interventions behind the curtains.
For more context, check out the story of "FN Meka", a massive blunder of a project to introduce a (fake) AI music artist launched by Capitol Records that melted down for them just this month.
These AI Models will make an impact in many domains and affect the distribution of job types available in art, legal, software, and other professions.
To prepare for the upcoming major changes, we need to urgently make creativity and innovation much more prominent in curricula across the board. Kids and adults will need to learn to adapt so they can thrive with such AI tools in the market.
Skills such as creativity, collaboration, communication, innovation using new technologies will be even more crucial to a large number of people in the coming years.
==============
I found this article to be a good discussion of many points involved. The author demonstrates pretty good understanding of the powers and limitations of these narrow AI models, although I'd say DALL-E-like models do have some level of understanding of how people use objects, based on both their textual and visual training data.
"...What these systems have is the meta-ability of learning to paint — anything in any style, already invented or to be invented. This is a crucial difference and it’s what distinguishes DALL·E 2 from any previous disruptive single-purpose technology.
...Artists can escape a camera because it’s static in its abilities. It can take pics and nothing else. ... AI wouldn’t replace just one style, it would replace all styles.
....
The most important one is that it doesn’t understand that the objects it paints are a reference to physical objects in the world. If you ask DALL·E 2 to draw a chair it can do it in all colors, forms, and styles. And still, it doesn’t know we use chairs for sitting down. Art is only meaningful in its relationship with either the external physical world or our internal subjective world — and DALL·E 2 doesn’t have access to the former and lacks the latter.
...
This means it can’t understand the underlying realities in which its paintings are based and generalize those into novel situations, styles, or ideas it has never seen before.
This limitation is key because we humans can do it very easily.
..."
>it can’t understand the underlying realities in which its paintings are based and generalize those into novel situations, styles, or ideas it has never seen before
As long as we keep putting a greater emphasis on the final output than we do on the process that led to the output, this "limitation" won't be a factor.
> And still, it doesn’t know we use chairs for sitting down.
I imagine if I gave it the prompt "chair being used by woman" then a majority of the pics would have someone sitting in it.
I realize your point is more of a "Chinese Room" philosophical objection. But it's important to note the level of sophistication these models have.
I'll also add that Nvidia is currently doing research on 3D model based generation, so that the model is aware of the 'skeleton' if the subject before illustrating it.
This is always the assumption of a new technology, and relies on people that assume the absolute minimum of the person's role. It seems you might have the belief that a creative's role is to sit and generate MS Office style clipart.
Instead let's take a short journey into the role of an illustrator with an example job.
An author hires an illustrator to produce a series of illustrations for their new children's story. The illustrator reads the story and discuss ideas with the author, they will talk through and present styles (or make a new one) that match the feel of the story, and match the progression of the story-arc. They'll craft mood boards and multiple story boards to best capture the story, then work through revisions with the author and finally proceed to artwork. Their skills extend beyond drawing, their skills are rather in the knowledge in how to craft and present in a visually engaging way, the picture is the end point - but everything else is 80-90% of the job. I know watercolour artists whose stunning visuals take less than 10 minutes to paint, their whole job is about the steps that came to that point.
However, let's get back to the illustrator - the artwork will then be tailored to suit the medium, for a book this may mean altering the imagery away from the spine area or where thumb placement typically is - for digital this could involve developing layered sub-elements to support animation or parallax, or ensuring that screen size isn't going to ruin the experience and so on, at this point we've now engaged other kinds of creatives to suit the specific technicalities of the output.
I won't go on, but the process does continue from here, it's a highly tailored process and every job is clearly different. AI text to image doesn't do any of this. It currently can't even do consistent images (hopefully that will change.)
Now we have people on the internet who have never worked in a creative field and not really experimented with StableDiffusion to see its obvious limitations all while telling the world that these people will be unemployable and that their craft will be worthless.
I think my simple example above helps outline that the drawing of a picture is just one part of the role. Here I've used an illustrator as an example, but it could have just as easily been a branding designer with a logo, a character designer, a story-teller, a prototyper and so on and so forth. All of these people ultimately produce a picture, but the "job" is everything that leads up to that point.
AI will merely be a new tool for these people, and they'll be excited to receive it.
In the short term, attorneys at any major commercial enterprise are going to look at Stable Diffusion's license and shudder.
The license is well-intentioned, but the use-restrictions are both broad and vague-enough that they may add extra unnecessary liabilities that one doesn't generally expect to find from a graphics-artist contractor.
One of things AI still cannot do is create things out of whole cloth. They still are trained on pre-existing data and create new things based on what they are trained on. One example of this is Japanese Anime and Manga styles which have changed since the 80s till now. You can tell which era an artwork is from by looking at the style. Like a lot of art, the medium goes through phases which arise by individual artists making small changes which influence other artists, and so on. The actual "creation" then of style really isn't a single creative event, but it is a social process which develops over time from the culmination of additions from different artists, more or less unintentionally. Once in a while, one artist or another may be so influential that they inspire many from one artwork, but usual the development is more gradual.
At least as it has been done so far, AI cannot reproduce what is essentially the result of a social process, which is the true source of creativity. The NNs so far can just reproduce what it is already trained on which is already the result of existing works in the current style since that's all it has to train on, and I don't know how to a NN can reproduce a "social process" without merely recreating a community of artists.
This is somewhat true, but then again it can be argued that the current style is a synthesis of already existing artistic practices from the past (aka 'there is nothing new under the sun'). And if trained on enough data, NNs do have this interpolation/generalization capability to explore spaces not seen in the training set (but seen within the training space distribution). In other words, does the distribution of the training dataset span all styles humans might find appealing? Maybe, maybe not.
So I made this comment quickly, I didn't expand on the last bit, which is what I think is key to understanding my argument. Basically, the required piece that makes art a social process is feedback, both from other artists and consumers (the public, in the past, patrons), basically whether they like, whether they talk about, whether it stays in the social conscious both for good and bad reasons, etc. Feedback performs the culling. That part thus requires other people, and cannot be reproduced within a black box currently, because you'd otherwise need to model an entire civilization's reaction to art, which is a coupled system which feeds back onto itself. Modelling very large, complex systems is a challenge still, for both traditional modelling (I happen to be a computational scientist) and for AI. AI has picked off some bits but it struggles in other domains. I feel like at this point, arguing "AI will get better at modelling complex systems" is similar to people who say "weather modelling with get better eventually," people really underestimate the difficulty with rounding out the last 10% of the problem.
Anyway, assuming you cannot model "society's reaction to art" in a black box, I think the only way to help refine DALL-E or other future AI systems is to "put humans back into the loop." You already see AI "extrapolating" from the dataset, and more often than not (which you expect from mere statistics) the results are ugly mistakes that AI researchers are trying to "fix", some examples of which the article lampoons as a draw back of DALL-E itself. The thing is creating "ugly" and having those pieces be culled is what is done by the external social process in art I'm referring to already. In this way, DALL-E doesn't replace art or artists, it just becomes another actor in the space, being subject to the same feedback that other artists are. You haven't really replaced artists, you have just added another competitor, so to speak.
Anyway, I went on a monologue, but one last thing I will say, I really think people making this argument seriously underestimate the size of the codomain of "possible styles humans might find appealing," especially since that set changes over time (that is, over the centuries of human history). I really don't know how someone can look at all of human history and think that "nothing new under the sun" really limits the size of that set. Just because a set is finite doesn't mean its "small." Cardinality of sets concerns things mathematicians make up in their heads, not things in the real world that require finite time to consider, count, process, etc.
Humans get to choose which images to use and which to discard, which might be just enough feedback. That said there’s nothing to say progression of the world requires humans in the process, perhaps eventually it will be AI all the way down.
I deleted a section of the last reply (I seem to keep doing that) because it was a bit of a "toke of the bong" sort of take. Yes, you can imagine a situation where it's just "AI all the way down," that is, you recreate a community of artists which are only AI and no human draws anymore. The thing is I feel like that situation, the AI looks a lot less like the NNs which are just matrices today and it looks a lot more like actual AI which is closer to what someone could imagine are people. That is what I can imagine in the future, which is a community of artists still, just not of humans.
Regarding an "all knowing" NN sort of black box that literally models the whole social process itself, I don't think is possible due to chaos, that is, sensitivity to initial conditions, like the weather. Large social systems are probably as sensitivity to initial conditions too and cannot be a priori modelled completely. That's kind of what I'll still stake as impossible in my opinion.
Your statements seem contradictory: AI creates new things based on what they are trained on, artists create new things based on what they are trained on. Soon enough, the training and the process will pull AI into the ability level of humans, and this is a huge step in that direction.
But human artists have a physical existence, memories of the life they have lived, social interactions and relationships, hopes and dreams for the life they wish to live, personal fears and regrets. A human is not an empty vessel or a pure reflection of a dead corpus of captured data.
They have a story to tell.
I’m not saying that a machine cannot be contrived to fake a story, but even a fake story is just a derivation from a collection of real stories.
What purpose would a machine have in telling a story about a life it never experienced? At best it could pretend.
Okay, then just take, say, van Gogh, and see if you can get it to independently paint The Starry Night (approximately) without having ever seen van Gogh's paintings or anything influenced by him.
Or pick impressionism generally and show that an AI that has only ever seen antecedents to impressionist painters can come up with something similar. It's hard to imagine something like DALL-E being able to do that.
Van Gogh was influenced by myriad artists before him.
The Impressionists were influenced by myriad artists before them.
Of course an AI could eventually be trained to develop a whole new art style.
Someone already said it earlier... there's nothing new under the Sun. It's just re-interpretations of what already exists. The value in art of any kind - song, writing, visual - isn't in creating something dramatically new and novel for the sake of creating something dramatically new and novel, it's an iterative, generative process that retells the same theme in a somewhat new way.
Dont forget there is also lots of thinking involved. Artists understand their context and work with that. Like they do things to piss people of or to contradict something. They are also painfuly aware of the fact that work cannt be too novel to become popular. Often times they “train” their audience to come to work they’ve done years ago. Artists are not only repeaters of image patterns like current AI is.
So unless the AI becomes concious then i dont think they will replicate this process. On they other hand they will be probably able to “brute force” novelty with mashing random things and help of humans to sort it out.
That seems to be the challenge: If you train an AI model on all of the art Van Gogh could have observed, but not any of Van Gogh's own images, would it come up with anything like "Starry Night?"
There's a gap between what comes before and what comes after, and it's not totally clear than an AI model trained on what comes before can cross that gap in a way that a human can. Maybe? I don't think it's an "of course" thing, though.
I think of it this way: an AI model trained on existing images has billions of images to draw upon, more than a human does. But a human has some subset of those billions of images and their lived experiences, too. That is, as I stand at my desk typing these words, I see out a window where the bushes in front of my house are overgrown and need trimmed back, but the blooming trees across the street are visible through the bushes and providing splashes of color in a tangle of leaves that look nice to me. The branches of the post oak in my front yard reach out toward the street, and are suffused with dark green while the trees across the street are more medium green and yellow. As I walk around the neighborhood later, I'll see much, much more sensory input than even billions of static images supply.
Artists in the past were often constrained by the choice of medium, and things like having to mix chemicals to produce new paint colors dictated the palettes with which they could paint. Those artificial restrictions affected things in ways that might or might not be clear to a trained AI model.
Perhaps you tag the year of each image in the model training corpus so that it could, in theory, have some sense of progression, but would it choose to consider that feature? If you then asked for a painting as of 1880, could it make enough sense of that to avoid colors and techniques developed after that time, or at least techniques that built on techniques developed after that time? It's possible the future will involve more carefully trained models rather than the "everything all at once" models the current generation of models is using, but... I'm not sure.
In any case, I think the poster to whom you're replying is suggesting that someone like Van Gogh isn't just creating "more like what already exists," but that sometimes they are introducing something entirely new, at least new to the art world, perhaps drawn from life, or a fever dream, or strong drink.
>Van Gogh was influenced by myriad artists before him.
> The Impressionists were influenced by myriad artists before them.
Yes. That's why I'm saying, if you can train an AI exclusively on the stuff that could have influenced van Gogh, and coax it into reinventing van Gogh's work, then I'll believe you've developed an AI that's capable of something like the sort of innovation that van Gogh was.
I'm skeptical that what van Gogh produced can be reduced down to a function of the paintings that preceded him, in the way DALL-E is just a big function of all the images it's been trained on. Humans are "trained" on a lot more than images.
This doesn’t hold water. Most if not all art is derivative at this point. There’s only so many things you can do on a 2D surface and there’s more than enough material for new derivations.
To be fair, I've had similar experiences with professional illustrators. I describe what I want, and... something comes back.
In practice, the technology will improve. I'm still waiting to see where the ceiling is to know how many people will keep their jobs. I see similar articles about codex, and v0 won't take my job, but v3 might. Or v3 might be identical to v0. It's too early to know.
What I have found all these tools to do is to make me use whatever they generate far more. There was a pretty high bar for custom illustrations before DALL-E. With DALL-E, my last presentation had a half-dozen custom images. It was awesome. I'd never hire an illustrator to do that, but I definitely did "hire" DALL-E.
codex and gpt-3 are starting to change how I work.
If I use illustrations 10-fold more, which I seem to be doing, will illustrators get more work? Less? I don't know.
I think you touch on a good point. I think, like the automation of other skilled labor, the first jobs to be affected are the low-hanging fruit jobs. The ones where they just need something good enough. While I think tools like these will decrease the total number of jobs, I think the jobs left over will be ones that artists will enjoy more, as they will require more creativity than just writing a prompt.
This wasnt the case with industrial revolution. Most craftsmen either ended up doing a lot more joyless work as cog in factory OR they had to accept/adopt the factory process moved their creativity to design and ultimately lost their craft.
Main users of these tools will have to be illustrators. It will drive their prices down (it will be race to the bottom when you compete with instant AI) and over time there will be fewer and fewer people who will be able to create these visuals without AI because there will be no incentive to learn. Everyone will become prompt operator expert that will be the job.
You can imagine the slowly dying generation of illustrators in 40 years doing interviews for local TV about their wierd craft called drawing. (just like now you can see docs about scottish grandmas making tweed by hand).
I wonder where will the people move up from the knowledge work. One would hope it would be more free time but we all know how that worked after industrial revolution :)
This is an excellent point and easy for designers themselves to overlook. They may see DALLE as a tool that some are claiming does their entire job, and it can't quite yet. But it's really a question of how close it can match the actual result of employing an illustrator.
Yeah a lot of these articles are basically “it couldn’t read my mind so it’s bad”. I’ve employed a few artists before and the workflow is quite similar to the version 0.1 we’re dealing with for AI art now.
Explain your concept and then get 1 out of 10 illustrations kind of in the right direction, start iterating on that for a few generations and so on.
A flood of illustrations may also devalue illustrations themselves.
Often people value things specifically because they're scarce, and because they took effort to make (e.g. hand-made vs factory-made).
So I wonder if AI art will devalue all illustrations to the level of memes and stock photos. Or maybe some illustration style that AI is bad at will emerge, and become more desirable?
Images are already devalued on the web. Most people think that an article needs an image to not be a “wall of text,” with the result that many articles have barely relevant images that don’t add anything to the article at all. Like hotel room paintings, however nice they are, they are there to take up the space because an image is expected.
Actually relevant images are real photos of the thing you’re talking about, charts and graphs, screenshots, diagrams, and so on. Illustrations can be important content (consider a bird field guide), but often they aren’t.
Like the clip art and stock photos they will replace, these AI generated images aren’t a good substitute for the real thing, but I’m hopeful that at least people will have fun with them.
It might be interesting to consider how AI image generation could be used for nonfiction. Could you give it five imperfect photos of a bird species and ask it to draw a good illustration? Or maybe auto-create a good infographic from a spreadsheet? You’d have to do something to keep it from making up data points, though.
Art / design teacher here. In our school we are very worried about this software. Absolutely it will steal jobs sooner or later. A similar thing happened when photography was invented... Painters had to re-design the job of an artist. Result= modernism (kinda oversimplification but true enough).
> A similar thing happened when photography was invented
Good analogy, I'd be curious how many people argue that photography being invented was a bad thing, and if at the time it was invented there were essays about how dangerous it was going to be
I'd speculate that when photography was invented, there was some gatekeeper group with a monopoly on it (to some extent true right up until digital photography became common) and so there was less vocal concern because portraiture (or whatever it's called) while evolving, still had some value capture for a small group, while recent image generation has been almost completely opened up, so there are no entrenched special interests that get to profit (and therefore more entrenched special interests whining)
> there was some gatekeeper group with a monopoly on it (to some extent true right up until digital photography became common)
Bollocks. There were consumer photography options widely available for decades before digital photography became available - everything from point and shoot cameras to decent SLRs, with a dozen kinds of film available, all at a price point pretty much everyone could afford.
Disposable cameras were even a thing for a couple of decades before modern digital photography took off...
The biggest democratisation of photography came with kodacks invention of the box brownie. The film stock it used was infact just repurposed movie film stock and much cheaper to produce and process than plate film. The camera itself was super cheap to make and operate. It heralded the birth of snapshot photography.
How things change. It was Kodak that invented the digital camera, but failed to capatalise on it as well as they could have.
No expert on photography history, but I know that it took quite a while for people to recognize the innate strengths of photography. Indeed, I have heard it said that the first true photographer did not arrive until around 50 years after the camera was invented.
The first applications of photography in art were clearly trying to emulate the paintings of the time.
Interesting fact. France bought the patents of Daguerre (Daguerreotype was an early name for photograph) and gave it to the world for free.
"Arrangements were made for Daguerre's rights to be acquired by the French Government in exchange for lifetime pensions for himself and Niépce's son Isidore; then, on 19 August 1839, the French Government presented the invention as a gift from France "free to the world"
Can you imagine if a country did this today just for the glory of it?
What are some convos & ideas about how to pivot/ adjust, in light of this taking some art & design jerbs?
Something that makes me not fear so much is: corporations and brands are still going to need artistically minded folks to handle media campaigns. Like just because you can make a turn-key website in Squarespace (and even in Adobe now I believe), don't you as a baller corporation still want to hire an expert/ team of experts to handle all that?
Like in the worst case scenario, aren't Art Directors and Creative Directors, Studios and Agencies, and even in-house marketing teams going to be using these AI image generation tools, as part of the process to reach the final images that go live?
Maybe we'll see less illustrators and designers, and more Art Directors (who'd still need to know how to tweak, refine, flesh out the images so they are on point)?
At least that's the way some colleagues of mine view these tools, as something to integrate in the creative process, and not as something that will replace the need for those who are aesthetically minded.
> What are some convos & ideas about how to pivot/ adjust, in light of this taking some art & design jerbs?
The million dollar question. The only idea that has effective traction so far is that we return to natural media (paint, sculpture etc).
> Like in the worst case scenario, aren't Art Directors and Creative Directors, Studios and Agencies, and even in-house marketing teams going to be using these AI image generation tools, as part of the process to reach the final images that go live?
Our worry is that we will end up training art directors, not artists. The problem will be that everyone thinks that they are an art director. Just like everyone thought they were a designer with the dawn of desktop publishing.
Valid. Not everyone's cut out for all positions/ roles indeed.
But that's interesting and worth considering: returning to natural media.
This is something I personally feel strongly about, but do not think will find any traction whatsoever given how things are unfolding, coupled with human nature across time, butttttt... I personally would love to see less production of everything. From churning out entertainment, to product updates strategized around planned obsolescence, and even car & bike models that have marginal upgrades where they won't benefit 99% of consumers/ users, and even food that tries to stay relevant with ridiculous versions of itself in the form of new flavors or toppings or whatever.
In some ways, perhaps whatever I'm longing for (less of everything) may sorta kinda maybe somewhat lineup with you and your colleagues wanting to return to natural media instead of digital? Maybe not?
Kudo’s for being proactive in looking out for the futures of your students.
In technology, it’s best to be the one who makes your own technology obsolete. The arts are one of the oldest drivers and consumers of technology.
Commercially, technology needs to amplify the productivity of artists. It should compound the advantages they have over the untrained.
I see the self-service model of illustration and content creation as potentially increasing the demand for skilled professionals. Most amateurs will probably produce generally low quality work, if judged by professional standards. But they’ll create way more of it. Directors (of even small projects like commercials) will draw their own concept drawings and story boards to get their ideas down on the page. But then if they want something cleaned up, or done professionally, they have a much better starting point to communicate with a real artist. Clients will also be able to visually convey some of the edits they would like to see in the professional works.
> In technology, it’s best to be the one who makes your own technology obsolete. The arts are one of the oldest drivers and consumers of technology.
That is true. Art has survived the invention of the camera obscura, the camera itself, oil painting, offset lithography, in each case managing to consume the tools of the victor.
I’m not sure that will work for corporate clients which is what majority of the work is for. Oh great so this red splotch conveys the trustworthiness of Uber to our riders? I’m sure they’ll get it
> So all of that to say that it’s very, like very, difficult to describe an image with words.
DALL-E isn't going to steal your job... but likely within a year a better interface for it will.
I don't think it's going to take too long to be able to specify textual components for image parts to occupy certain areas, and then to do iterative non-textual refinement simply by selecting among rendered variations "genetically" until you get what you're looking for.
> It took me maybe 10 to 15mins to generate all those images... And it took me something like 30mins to draw it myself,
But the point is, I can't draw. We are very quickly going to reach a point where anybody can produce basically the exact illustration they want in 15 min. It's not here today, but will almost certainly be here in 2 years if programmers and designers are allowed to experiment with the raw models.
I actually think discovering the best UI interface for generating illustrations, the magic combo of prompts+seeds+outlines+refinement, is going to be the main area of progress soon.
> But the point is, I can't draw. We are very quickly going to reach a point where anybody can produce basically the exact illustration they want in 15 min.
I’m curious about the economic impact to illustrators, artists and designers because while I agree with you that this opens up the ability for people who can’t draw to produce stuff they couldn’t before, it seems to me that this is a fairly marginal use-case.
I’m personally excited about these tools because, like many electronic musicians, it’s helpful to be able to produce cool album art easily. But I wouldn’t have paid good money for this before. At best I’d pay some freelancer on the internet $20 to throw something together. Meanwhile, in the professional contexts where I’ve dealt with profesional artists and designers, the demands are far too precise and nuanced for these tools to be really useful.
I understand that these tools are still in their early stages, but is it possible that, like self-driving cars, the last mile in terms of professional-level suitability will be much more difficult than we might expect?
We cannot compare self-driving with art generation. If self driving fails at an edge cases somebody is hurt or dies. When art fails you just need a better prompt.
Where self-driving and art generation are similar is a belief in the extremes of computing capabilities.
Self-driving enthusiasts believe that a computer will be able to operate a car far more safely than a human. This is in spite of the complete lack of evidence (so far) that this is true.
Similarly, look at your statement “when art fails you just need a better prompt.” It seems to exclude the possibility that computers will be incapable of creating certain art no matter what the prompt is.
I’m not saying this because of some ineffable or supernatural attribute of human creativity, but because of the shortcomings in our flawed human attempts to create computing systems.
But are those shortcomings fundamental and insurmountable? I don't believe so. While I agree with you about the current state of the art, in principle I think the project of "AI creates original art" is possible. Though it may take a few centuries to get there.
> But the point is, I can't draw. We are very quickly going to reach a point where anybody can produce basically the exact illustration they want in 15 min.
Exact is an overly strong word. Suppose you could put onto the page exactly what you thought a pterodactyl looked like. Would you be surprised that there were skilled technicians who could produce much more accurate and visually appealing illustrations?
You may not even know what you actually want, until after you see it rendered by a skilled hand.
Yes, like other technologies, DALL-E and Co. will change and empower the mass, like the compact camera and the phone camera before. It feels like a tectonic change because of its potential impact for the mass. But most people who can not draw have also never hired a professional to do some work. It might make the old, not demanding works obsolete. But not the top 5%.
In my opinion, image generation models will replace some jobs - the low-value stuff that's basically "I need somebody to draw a bunch of random things and I don't particularly care about the outcome".
Now, moving up the value chain we find ourselves at design agencies that do 2 things: they gather requirements and then they convert those requirements into an output. Many times a client themselves don't know what they want, and part of a good agency's job is turning nebulous requirements into something usable (and charging through the roof for it).
The technology will make images much more accessible - the same way Google translate is crap, but it opened up me communicating with a random foreign dude when travelling in a new country that nobody would've ever hired a translator for. The translators moved up market to industries where it's important that a human verified everything looks good like official government/corporate publications.
Same thing here; one day you'll have these models in powerpoint and people will click on the model when they need some kind of image there. The high-end won't go away, but the low-value work almost certainly will.
As a computer vision researcher who has worked with these T2I generators, this is more spot on. I think a lot of people are making a lot of assumptions without having actually worked with these networks. There's also a highly fundamental problem that is not only unsolved, but we don't even have great directions to solve: alignment. Alignment is even a difficult task when you're communicating to another person. Everyone has had an experience of working with someone and they are asking you to do something (or vise versa) and there is a gross misinterpretation. That's alignment. Now try properly aligning with an "entity" that thinks nothing like you, has a vastly different set of experiences, doesn't know how to count, and has no real concept of physical mechanisms (and more).
The technology will definitely help us see a lot more unique art and a lot more amateur artists (a good thing). But it would be naive to think that artists are going away anytime soon. I'd even argue that there will be new jobs from this (prompt engineer comes to mind). It is really difficult to speculate how this technology (even as it progresses rapidly) will change the future. But considering that no one is creating machines that think like humans we will always have alignment issues. This isn't a fact that can be ignored.
It is also worth noting that if you haven't worked with these systems, that what you are seeing is extremely biased. People aren't showing their failures. But there are some twitter accounts that do: WeirdStableAI and weirddalle are probably the most famous. Of course, you can also follow Gary Marcus who is going to retweet every example of failure that he can get his hands on.
Other side note: some of these conversations remind me about how people used to talk about 3D printers
This confirms some of the suspicions I've had after playing with DALL-E for a few months. In the good examples I see from Dall-E, oftentimes the things that it frequently does wrong (face symmetry, hands, feet, limbs) are just missing altogether because they're not needed to produce the prompt. Even in some of the good ones, a lot of elements out of focus seem to get smeared into other objects quite often, or the "smudginess" contributes to its surreal style.
People keep saying that sort of problem with be optimized away, but people often underestimate that when a technology plateaus at a certain point it usually requires a completely new innovation to get past some of the hurdles - not just optimization. It can take decades for those innovations to arrive.
Of course I don't have enough specialized knowledge to know if that's really the case here but it's an observation I've noticed about hyped technologies in my lifetime.
Well IIRC DALL-E specifically in their public release tries to censor humans. But other works like Stable Diffusion and Midjourney do not have these restrictions. Still, both are quite terrible at human faces. This is always interesting because generative tasks like FFHQ or CelebA have always been considered easy (despite humans being quite adept at recognizing when something is wrong with a face). But being able to create whole scenes requires a substantial amount of diversity. A good network can both memorize and create. There are some technical details that I am certain will be resolved (like long range features) simply due to increased hardware power/memory. But alignment will still be quite difficult as the entire problem is ill-defined to begin with.
But to give you some hints of how to use DALL-E better, there are magic keywords and this is why people suggest pretending like you're writing a prompt as if the thing exists. Some of these magic words are: screenshot, unreal engine, photorealistic, studio Ghibli. It does better with anime styles, probably due to training and human interpretations. Try to write longer prompts too. For example "flying otters" will give you otters in the water but "a photorealistic unreal engine render of otters flying through a beautiful sunset" will give you something much closer to what you actually want.
> Now try properly aligning with an "entity" that thinks nothing like you, has a vastly different set of experiences, doesn't know how to count, and has no real concept of physical mechanisms (and more).
Indeed. This point was also made in a different way in the article:
> So all of that to say that it’s very, like very, difficult to describe an image with words. There are so many things that an illustrator will instinctively know how to do, that the machine would need to be told to do. Even if it was possible
It struck me how similar this was to the basic challenge of programming: computers don't have the layers upon layers of background information that humans take for granted, to be able to accept descriptions of the simplest tasks. You must explain every minute detail to them.
> It struck me how similar this was to the basic challenge of programming
Yeah, this challenge is actually fairly universal. I said this in another comment but it is worth repeating. When communicating there is: 1) what you intend to say, 2) what you say, and 3) what is heard. These don't even have to be the same thing. There's a lot of ways the miscommunication can happen but this is also why we need to act in good faith. As the speaker you need to convey the idea in your head to someone else's head. As the listener you need to try to interpret what is in someone's head through the words they say. People think language is extremely structured but there is a lot lost in between and we fill in so many gaps. Once we lose good faith communication becomes near impossible because we won't correctly fill in the gaps.
>Now try properly aligning with an "entity" that thinks nothing like you, has a vastly different set of experiences, doesn't know how to count, and has no real concept of physical mechanisms (and more).
The biggest issue with alignment is that humans don't really know what they mean nor want in the first place.
Yet, we train these networks to produce high quality images, whose resolution is necessarily higher than the resolution of the fundamentally ambiguous human input.
Training a high quality image is different than training a high quality image of the specific thing you want. One has substantially more flexibility. It is ridiculous to compare the two.
As for humans, I will give the constant reminder. There are 3 parts to language: 1) what you intend to convey (what's in your head), 2) what you actually say/write (encoding head to physical), 3) what the other person understands (decoding physical to mental). These 3 things can have 3 different meanings. Do your best in the first two, but the third requires the other person to be acting in good faith.
There is a difference between art and illustration.
People consume art because they enjoy admiring the human talent that creates it, celebrating that some individuals are capable of extraordinary feats the vast majority of people are incapable of. It’s the same reason people watch sports—they enjoy admiring the top echelon of human physical ability. Very few people would watch Olympic Games performed by realistic androids.
On the other hand, most illustration exists for such mundane purposes that viewers never bother to consider the illustrator. Corporate Memphis [1] designers’ days are numbered.
That said, sometimes the two overlap. I’ve certainly ogled extremely good looking websites [2] or well-executed pieces of industrial design [3], implicitly admiring the human talent that created them. Perhaps the glut of human illustrator talent soon to enter the marketplace will kindle a design renaissance, with brands competing to distinguish themselves from the sea of DALL-E generated Corporate Memphis clipart, just as Apple distinguished itself from the sea of beige minitowers in the late 90s.
There's a whole world of paying jobs out there where near enough is good enough. Where the client just wants to come up with something, and is happy to go in a direction that balances speed and cheapness.
There are millions of these jobs, taken by juniors and people trying to build a portfolio, and they will be diminished greatly.
In my experience, there is no such thing as a design job where the client (or boss) doesn't make some kind of specific requests or feedback. Of course, with human artists, the marginal cost of such requests is relatively low - perhaps when the choice is between getting your stupid request honored for a human wage or taking what you get from an AI for ~free, executives will realize that none of their opinions are actually that important.
This is the same hollowing-out that came for all other industries. The high end people trusted with non-obvious problems continue to be worth a ton (worth even more if they are good users of the new tool), while juniors who do rote practice become nearly worthless. It came for lawyers, doctors, engineers, and even writers. Now commercial artists and designers are in trouble.
How did it come for doctors? AI is only useful in narrow applications like radiology and spotting things that are off, maybe radiology techs get displaced?
The hollowing out of primary care via midlevels e.g NPs and PAs is interesting however. They are fighting for the ability to diagnose and prescribe meds and see patients without physician oversight, largely driven by pressures to reduce costs by politicians and big insurance companies.
Radiologists, pathologists, etc. are now competing with AIs. Other doctors are now competing with RNs supervised by a single doctor and a huge electronic medical records system. The digitization and standardization of medical records is the transformative technology here.
And if the juniors can’t pay their bills doing the thing, who’s gonna step into the high-end shoes as the people who mastered the thing start to retire?
If you look at lawyers, you can see a narrow pipeline where there once was a wide funnel. People from top law schools and people who distinguish themselves early basically get groomed for partnership, and other lawyers have a hard time cracking 6 figures. I assume this will be the same for artists and software engineers too.
Aren’t paralegals going to get displaced by automated case law cross referencing systems doing the leg work for lawyers? I guess they would want to keep billing the most hours and use slow paralegals, while also using the automated systems to ensure their legal arguments are the best researched.
And that's the base higher ambitions build on, as a fallback layer. The "fine arts" certainly won't be the same without the lower levels of the pyramid. With the abundance of stable diffusion offerings, I'd expect that wave of "style transfer" ML that was all the rage a few years ago to become more important. Perhaps the act of shepherding recognisable ML identity providers won't be quite as bad a substitute of those low level graphics jobs as we think.
This is a very narrow viewpoint. This is 2022. You can be sure that DALL-E and Stable Diffusion and subsequent iterations will be continuously improving at a faster pace (especially those that become more open), and 5 years, 10 years from now, you will be able to make very convincing images that will absolutely compete with the best artists out there.
It will not "steal your job" right now, but that's besides the point. With image generators you can produce 100's of images for "free", and you can just spend your time selecting the best in the lot. For most needs out there, it's enough.
Yup, this article is some heavy, heavy copium. It feels like digital job automation is only a matter of time at this point.
However, I for one am not looking forward to the piles of copilot spaghetti vomit we're going to have to clean up in five years... because copilot will be abused to hell just like Access and Excel were in the 90s-00s.
Even if it never reached such peaks, it'll still steal the majority of lower-level jobs in every industry it touches, creating a giant expertise gap between new graduates and the level required to improve upon AI generators. That's clear just from current tools working now.
This comment from the author was an interesting perspective on how it'll impact the market for illustrators.
> "Those "clients" [who just need an image and don't care about the quality] you are talking about don't really exist. I know they do, but on a practical level they don't for us freelances. It's a long debate among graphic designers and illustrators.
> To explain it simply, there are two big categories of "clients" for graphic designers, illustrator, photographers etc... 1-Those who want good stuff and can pay,
> 2-Those who don't care/can't make the difference between good and bad, and/or who don't have money.
> The first category is the only "real" clients who exist, on a practical level. They are the only one I work with, like any illustrator.
> The others, AI or not, will NEVER pay for a "real" illustrator anyway. Never. That has nothing to do with AI. I know them, they contact me often. Then I tell them my rates, they try to negotiate, then they give up. I never ended up working with them.
> They either contact a very junior illustrator who have no idea what they are doing, or they buy cheap illustrations on Shutterstock. They don't have money to spend anyway. They will do the same horrible images with AI that they already do with shutterstock. If they "leave" the market, it will impact nobody."
lol, former animator/illustrator. It's a race to the bottom. The jobs looking for quality won't be replaced, but the entry-level jobs are changing. Where in the 80s they used to develop skillsets that you could build on, working towards feature work, increasingly they are optimized for quantity. The skills are not necessarily transferrable.
Look at 3D modelers. There used to be several more environment artists that have been replaced with 3D scanners in videogames. This isn't necessarily a bad thing, it's just a different skillset. It reduces the unit cost.
I think entry level illustration jobs are at a higher risk. I've done plenty of album covers, promotional posters early in my career on a shoestring budget. They don't care if it's interesting - they want something cheap that gets the job done. Storyboard artists have a very different skillset. You need to communicate an idea as a cohesive sequence of images. Dall-E isn't designed for this, it's designed for standalone images. If you're investing in storyboard artists to begin with, you care about this visual communication, otherwise it's much cheaper to just yolo with a camera and and hand it off to post.
IMO, the most successful artists will learn to embrace AI artists to quickly generate ideas and concepts at scale, and then refine them into brilliantly perfect works of art for production usage.
The analogy is use is that renaissance masters had workshops with multiple apprentices who would do things like paint gross details according to the master's guidance, then the master would finish the work. I expect this will play out similarly.
I'm imagining some artists will master prompt writing and then use AI image generation to fill in sections of an image one at a time, taking direct control of composition and style selection - but leaving the fine details mostly up to the AI.
But for most utilitarian images the artist is soon dead.
Crazy, this article was rising on the homepage of hacker news. Got up to number nine when I checked and then all of a sudden, I refresh and it was gone. Demoted to page two. It’s a shame too cause more people need to read this.
I think DALL-E will end up more as a tool to provide inspiration for artists as opposed to one that will replace them. I can see AI working the same way in other mediums in the future (music ideas for musicians, for example).
No. You're watching the beginnings of these tools now. In a few years they'll blow your mind. There's no reason to believe these won't be an order of magnitude more powerful in short order.
Just as an artist starts with an outline and iteratively refines, so too will these tools enable such workflows for complete novices. You'll be able to create an illustration in ten minutes exactly as you'd like, without the artist's five hours of work and ten thousand hours of practice. Unlike an artist, you'll have thousands of paths to choose from at your fingertips.
Illustrators are about to go the way of basket weavers, crocheters, and elevator attendants. This isn't like photography where you need to hire someone for your event. Illustration is offline, async, and now it can be done for fractions of pennies.
The article almost gets to this point when they touch on the idea that "People still play Chess even though people can't win against computers" but
1) People don't do things just because it means a Job for them
2) New Jobs will be created to work through the vacuum left by new Tech
2.B) We'll sooner-or-later have to grasp with a different conceptualization/organization of how we allocate the resources for human life (Food, Shelter, Community)... or else. (Apply this sentiment for anything we can do to earn our livelihoods, Art is not exceptional in this regard)
I see technologies like Dall-e being GREAT for capital A Art, in that I wholeheartedly believe our current system Art is worse-off than it otherwise could be. If 'Artist' stopped being a 'job', that'd be a good thing. The more 'Work' we eliminate from human life the closer we get to a place where our hand is forced to change, and in my view, for the better.
I find it really fascinating to see the amount of similar posts theses past few months. Professional artists, fearing for their jobs due to AI-generated art, who find solace in the fact that the resulting artifacts are still missing on production quality by a tiny margin. This is somewhat similar to the fear that self-driving cars will one day replace the most common job in America.
In the short term, I think there's a real opportunity for artists to leverage those tools to create things they didn't think were possible before. In the long term, it seems inevitable to me that all jobs today will one day be performed better by computers. I do like to think that we will adapt to those changes over time, like we have in the past technology breakthroughs. While the nature of our jobs will most likely change, ideally we all end up having more time to spend with the ones we love.
Art isn't like chess. Chess is an event, like boxing, where the interest is in seeing two humans square off. Art is not an event. No one pays for artwork so they can watch someone draw on a canvas for 5 hours - it's not really the same thing at all. They pay for artwork to get the final product. (Or conversely, if chess was like art, then you'd be paying money to read the moves after the game is over.)
> A machine can lift a house, but people still love to participate in and watch the strongmen competitions.
Yeah, but who would pay to watch a bunch of people lift a house? I mean, maybe as a spectacle, but certainly not as a replacement for a forklift.
> We don’t care if any average Joe is faster on his motorbike
This isn't really true, either. Have you heard of NASCAR? :P
Pretty fun post with specific examples of how current tech is not sufficient to take over his job. I have also found when trying to prompt for conceptually specific things (with layout and semantic aspects) it's not that good yet. Will it become massively better though? Probably, yes.
Aside from all that, I like the way the piece ended:
"Ai is a tool. What makes art is not a tool, it’s the will to make art. A machine doesn’t have will. A tool can transform art, improve it, expand it, not erase it. I’m very excited to see what those tools will bring to futur artists. I’m not worried they will steal our jobs. I mean ok, maybe just a little bit deep down haha, but not in the short or mid term at least, that I’m sure of :)"
At least in the realm of editorial illustration, concept is king. You can get away with very simple art so long as it is conceptually effective. Like, it wasn't the widespread availability of pencil and paper that put the poets out of work.
I honestly don't care how many jobs AI will replace, it's a neo-Luddite argument. The point of AI is to replace jobs, and I certainly won't stop human progress just because some people won't have a job in the future.
The author talks about the quality and professionalism of his work which is true. I'm not sure how much a model can replace that.
However, with this is cheaper to produce a huge amount of substandard stuff. That will swamp out the quality stuff and will probably have economic effects too.
One analogous example that comes to mind is cheap animation software. In a world where there were a few hand animated cartoons, now there are tons of quickly put together shows.
Like every other kind of technology before, DALL-E and Co. will help the mass to create a lot of amusing and sometimes passable artworks. Just like the compact camera and the phone camera had helped create billions more photos compared to the previous generations of equipments. But arguably, the trillions of photos and billions of potential photographers didn't help to bring further the art of photography much. Combined with myriads photo apps, they help created tons of amusing and pleasant photo artworks. They raised the bar for what is considered a good, acceptable photo and montage work today. They made creating good photo work much faster and cheaper for the mass, and ways more accessible. However, they didn't change the standard for professional work because they still require the hand and the mind of a talented artist. It will democratize the illustration process but like in every democracy, the elites still set the tone and exercise enormous influences.
In other word, new technology will skew the distribution to the right, increase the median dramatically and leave the old bottom half obsolete. But not the top 5%.
I don't see the big deal with either DALL-E or Stable Diffusion. It's compositional value is nil. At best it's text-to-img mapping may help some with getting to learn what different artistic styles look like, but I see it as the same dynamic as GitHub CoPilot.
Garbage In, Garbage Out. The more it's used, the more it's weights will be skewed toward SD'ness, and to be frank, I'm not entirely convinced that there won't be rapidly discovered "verbatim" reproductions of things as outputs that'll turn into legal challenges a la license enforcement.
Further, the user experience is extremely disappointing. One is quickly left feeling as if one's time may be better spent articulating one's vision enough such that a human artist can make it happen.
It's much like what I call "the hair stylist" problem. The hard part around communicating is figuring out the hairstylist vocabulary. DALL-E may make learning that easier.
Then again, my track record with "meh, never happens" has been abysmal lately.
The only reason we see articles like this about AI stealing (or not stealing) jobs is because we have a collective fantasy that generalized intelligence is right around the corner. It's not.
But that doesn't mean we won't get there eventually.
AI (and AR/VR, but that's another topic), to me, is like PCs in the early 80s. Everyone can kinda sorta theoretically understand the value, but the technology has not advanced enough to have a meaningful impact on our day to day lives, outside of very well defined problems. The use cases that most of us can think of require massive leaps in technology.
Once upon a time, people thought of computers as fancy calculators. Steve Jobs had to compare them to bicycles to get people to understand the benefit. Computers were so bad for so long that any movie that tried to feature a hacker subplot had to come up with all kinds of crazy visuals to help the audience relate to this strange new world.
Same with AI. We can see what it's doing now, and be disappointed when we realize it's nowhere near as good as we've been led to believe. Or we can zoom into some distant future where AI is just as conscious as anyone else, but super intelligent. What we struggle with is the in-between, the journey to getting there.
DALL-E is amazing. One day we'll look back and be able to see how crucial of an advance this is.
But the PR story around DALL-E is all smokes and mirrors. It's the classic tale: Scientists make a discovery; the general public doesn't understand the significance of the discovery; so someone (usually in marketing or PR) has to paint a picture to show how incredible it is, otherwise funding will dry up. That picture, inevitably, is both useful and inaccurate.
For another analogy: Just like a map of the world is inaccurate but useful, so are these fantastical stories. We just need to remember: the map is not the territory. It's just a tool to help us navigate the territory.
Since no fully autonomous cars are sold and widely deployed, we do not know.
A lot of tech people are blind about fetishism of tech for its own sake. I am all in for new tech and modern inventions, but I am also surprised when people advocate for new things even when they do not work because of aesthetic reasons. Software engineers are not immune to this ("our scientists were so preoccupied with whether or not they could, they didn't stop to think if they should")
Short rant about self-driving ahead.
I think we will reinvent the train. Fully autonomous cars will work only in certified roads with special signage. This will offload the cost to the private car owner, as opposed to building public transport.
Otherwise, it will remain a lane assist tech, which I am perfectly fine with. IMHO until I see fully self-driving trucks I won't believe it is around the corner, since the economics for trucks make much more sense. But, since the cars can be controlled remotely via 5g, I do not know if it would ever be worth the cost to have them be fully autonomous, instead of remotely controlled by humans. Maybe the solution is a remotely controlled car by a human. Interesting times.
I keep thinking about AI image models' effects on illustration and comparing/contrasting with things like GitHub's CoPilot for code.
Inasmuch as someone starts with an exact mental image and wants something matching that mental image, then no, it could be quite a while before an AI model is able to do that. Maybe six months, maybe six years, maybe never. As always with AI, it's hard to say. How often is it true that someone insists on something that exactly matches their mental image? I honestly don't know! If I write another novel, how picky am I going to be to be about the cover? If I get Stable Diffusion, or a future version of something similar, to generate a dozen images that match my text prompt, do I just pick the one I like best and lay some text over it? Or do I decide that no, I need something very specific, and hire someone to draw exactly what I want?
I think one of the main criticisms of AI code generation is that you end up needing to be so specific with your "text prompts" that you're essentially writing the code yourself and letting the AI handle syntactic sugar. A sort of "AI IntelliSense™," if you will. But that's because we're developers. Like the illustrator in this piece, we have an exact idea of what we want, and nothing else will do.
What if I were a non-technical person and could ask an AI model to give me code that collects names and email address into a database I can later use for a mailing list? Could an AI model do that? If not today, then surely soon. Isn't that a similar issue? Does it put me out of work? No, not immediately. But it does potentially eliminate an entire category of work, specifically the work that people generally use to learn how to be good developers. Will this cause problems with training future developers, since they have to rise above a certain skill level to exceed AI models? It might!
I wonder if the same is true of illustrators. Maybe existing illustrators have their niche, but how hard will future illustrators have to work before they can exceed AI models?
"when camera became cheap, and everybody had a good camera on their phones, people assumed it would be the death of professional photographers. But it was not."
Prices for stock photography fell so low that it stopped being a viable way to pay for housing and food. All those displaced stock photographers then tried to get into adjacent markets, which ruined wedding photography prices, too.
I predict the author will soon face increased competition by other storyboard artists. Those that used to do art which has now been replaced by AI.
DALL-E and other AIs are ultimately doing a cheap, very 'human' task with unclear utility function. I expect professional human artists will always be able to compete for the high end.
However, there's an economic risk. AI will definitely cramp the lower end of the market, and with that there'll be fewer humans who are able to get to the high end market (since they won't be paid before gaining enough experience to become better artists - leading to getting another job which means less time to draw...).
This blog post seems invalidated by having the ability to pass graphical prompts in. So combined graphical and text prompts together with some constraints will be enough eventually.
Yeah. A lower skilled artist using basic sketches and prompt-engineering with img2img could make this work. I’m surprised you’re the only person to mention img2img in this thread but I guess it’s pretty new.
This example shows how you can make a drawing with stick figures and then run that through img2img with a prompt to make some art that looks artistic. The model is able to understand what the stick figures represent and then make real art in their place. People often stack output so they take the output of img2img and run it through again to let the style drift more while maintaining the original locations. You can also quickly and poorly photoshop multiple image elements together to form a composite, then pass that through img2img to make something nice.
I am well and truly jaded from decades of silicon valley's hype train bullshit, but after playing with this tool for a week on my home desktop, I feel like it is powerful enough to find a niche that will displace some artists. It may be though that it makes art so much more accessible that it mostly just increases our use of art, I don't really know.
The author is probably correct, for the next several months. After that I would expect new types of careers for people who are good with “prompt text” and abilities to touch up generated images.
In a few years, being really good with prompt text will probably be good enough.
Last weekend I used DALL-E to make a 101 year old birthday card for my Dad and everyone loved it. Last week I started generating images for a sci-fi story I am writing.
The difference is that nobody really cares that a computer can win in chess/go/etc. It's almost a given, and the novelty of "when" came and passed. People follow those scenes because they enjoy the history/lore/sport itself/etc.
Artwork/illustration, on the other hand, is ultimately an end-product to consume. There's potentially a larger impact here.
I have spent some time with dall-e and it is really useful for touchups and content that can stay at the edges. The centerpiece still has to be hand designed by a living breathing team.
That said, using anything other than dall-e for social media content feels like a waste. Why use a human for generating low effort read once content anymore?
yeah the generated examples were somewhat decent, with a few more iterations I can see it matching some turing-complete form that clients will pay for.
Thanks for reminding everyone that computers in all their glory and usefuleness, with all their AI will always remain a tool in service of human beings.
“softwares” do improve over time. so while it may not steal your job it willower the entry barrier. so your income will likely drop significantly once the tech matures.
Why I don’t care about it stealing or not you job - because it does get me what I need without having to go the all the lint story how special illustrators work is.
You seem to be replying to the headline in a knee-jerk manner. The article tries to communicate that the natural language is inherently poor at giving directions for artistic work. It's a valuable observation by an expert, and as someone who studies arts as a hobby I absolutely agree. Text-to-image isn't going to change anything on its own; higher order tools are needed.
"It doesn't affect my job yet because I'm not an artist, so I don't care".
Yet, we see managers at software companies off-shoring for cheap contractor coding sweatshops in the South East of the world and on the site, which HNers here see that as a 'problem' and is hated on.
Well guess what? Copilot will make it worse and will have cheap software shops generating code / project templates and getting more for less rather than employing an expensive full time typical western software engineer. Open source will just accelerate the drive down against closed source alternatives as well.
The start of the race to the bottom, Then it would be a problem for software engineers, software companies, etc. Open source alternatives is already eating itself and closed-source alternative and tools like Copilot used by contractors will just accelerate this much quicker.
If I want a bit of clip art for a presentation or stock image for a website, is it disdainful to say it's irrelevant to me if it was made by a living human or a computer?
I want the thing. An artist getting paid is a means to an end for this purpose. To you and others, the artist getting paid might be the end. That's fine for you, but I may not have reason to care.
There is not some universal right to get paid for things you like to do. Pay is a transaction between someone who wants something and someone who can provide it. I once saw a Ted Talk by a busker who was a mime or something and was calling what they do "work" (and was complaining about people who didn't respect that). I found it ridiculous because effectively they are a beggar, they are doing what they want to do and hoping someone gives them money, and in some cases, feeling entitled to it. I think it's this perception of entitlement (real or imagined) that rubs people the wrong way about artists, at least ones complaining about not getting paid. If you want someone to give you money, do something people want. The universe doesn't owe you a business model.
Along the same lines, weavers and scribes and travel agents and a million other professions have moved on. Automating low value art is not any different, nobody owes you anything.
> a busker who was a mime or something and was calling what they do "work"
How is it not work? I mean, they are providing a service and then wanting money. The money is provided after the service and 100% voluntary, but does that make it less of a job than a waiter with a base pay of $2.15 an hour? That guaranteed $4,300 a year separates a job from a not-job?
You appear to be trying to Godwin a discussion about automated picture drawing... even accounting for how your carelessly your analogy normally gets used, I'd say you're reaching
You appear to have taken my intent wrongly. I meant that eventually this tech may come for us super special developers, too, in some aspect. I could imagine one good developer doing the work of several once code models are perfected. The commenter I replied to had zero respect for the effects this tech might have on real humans.
