Is OpenAI’s DALL·E an existential threat to illustrators and photographers?
OpenAI announced this week that they have trained an AI to take simple one-line briefs and generate corresponding illustrations and photo- realistic images. The results are stunning. But what are the implications for the creative industries and the creative people who work in it?
FIRST OF ALL, WHAT IS DALL·E AND WHAT CAN IT DO?
Which is fine, but falls into the parlour-trick end of AI. Impressive but ultimately pretty useless. After all, how many times have you only had part of an image and thought, ‘if only I could make up the rest of this image’?
DALL·E takes this one big step further. Built on top of OpenAI’s GPT-3 language model, which can produces text that mimics human- generated language, DALL-E can not only create imagery, it can take a simple written brief.
OpenAI describe it as “creating anthropomorphised versions of animals and objects, combining unrelated concepts in plausible ways, rendering text, and applying transformations to existing images”.
But really, you need to see it to get it and here are a few I pulled out to give you a flavour. There are hundreds more to play with on the OpenAI website.
Prompt: a professional high-quality illustration of a flamingo turtle chimera. A flamingo imitating a turtle. A flamingo made of turtle.
Prompt: an armchair in the shape of an avocado, an armchair imitating an avocado
Prompt: a male mannequin dressed in a black leather jacket and navy jeans
Prompt: A store front that has the word ‘OpenAI’ written on it. OpenAI storefront.
Of course, being AI, it sometimes gets it pretty wrong/weird (and obviously, sometimes wrong can be more interesting than right).
Prompt: a cross-section view of a human head
USE-CASES: WHERE COULD DALL·E BE USED?
Currently, DALL·E is limited to 256x256 pixels, but assuming it will get to a decent level of resolution, there are many industries which regularly use still imagery and for which this type of program might be useful:
Advertising — print ads, out-of-home, digital display, mock-ups for pitches
Publishing — book covers, children’s book illustrations, graphic novels, humour, encyclopedias
Journalism — illustrations for newspaper/magazine/online news site features
Social media — images to accompany posts/stories
Product design — stimulus for new products (what would a soap dispenser shaped like a sardine look like?)
Web design — royalty free, bespoke imagery for websites
But just because these industries use still images doesn’t mean that DALL·E is something that would be taken up by these industries. Adoption of a new technology is often complicated and takes time. So I asked some people in the business to give me their thoughts.
THE PROFESSIONAL REACTION
Simon Mannion is a creative director who has worked with brands like
Shell, adidas, Bacardi, Sony, Pepsi and Nectar.
“I wouldn’t use the output for final imagery. You would want a style that’s unique. It can’t feel generic, which a lot of this is,” he says.
“So, it’s not a game-changer, but I can see it being a really useful tool like photoshop or Getty images in mocking up and scamping ideas together, which is something creatives do all the time. For instance, the shopfront example — I could get that made up by someone in the design department, but it might take 24–48 hours and it’s another person in the loop. It would be quicker if creative team or an art director could just do it themselves.”
“It’s interesting that the more imaginative the brief, the better result. Mixing a flamingo with a tortoise gets you to results you might not have thought of and wouldn’t have found elsewhere on the internet. But cross- section through a brain looks like stuff I would have found doing an image search.”
It’s not a game-changer, but I can see it being a really useful tool like photoshop or Getty images in mocking up and scamping ideas together — Simon Mannion
“It looks like it could be a great muse, to inspire. But if i were an illustrator, I wouldn’t be sweating just yet, it seems to be more of a fun, throwaway kind of thing at this initial stage. But I’d be keeping an eye on innovations like this. As we know these things can move fast.”
One younger illustrator I spoke to, who wants to remain nameless, had three objections to the tech:
- This is so derivative, you can get most of this kind of thing on the internet already
- This is just tech people infringing on the creative space to make money so it’s wrong morally
- To get anything interesting, they will have to rip off existing artists’ styles. Isn’t that illegal? [Probably not, as copyright doesn’t protect ‘style’, but that’s yet to be contested]
A BIT BETTER THAN GOOGLE IMAGE SEARCH
As it stands, DALL·E is not ready for the big time. As pointed out, the look is often derivative, and it could potentially be quite expensive and time consuming to run the algorithm, which generates 512 images to then give the user the best 30-ish.
And of course it’s not very ‘tweakable’. You can’t go back to the AI and say, “I like that one, but can you make the flamingo a bit happier?”
But it could be useful for non-professional uses such as school projects, making Spotify playlist cover art, internal Powerpoint creation, Squarespace-style web sites, and so on.
Given that licensed imagery is currently pretty cheap (eg £20 for 10 images a month on Adobe Stock) and also the ready availability of royalty- free imagery, there will be a limit to the size of this opportunity.
But technology doesn’t stand still and I’m sure that it will it ultimately be more than a better clip-art-slash-Google-image-search.
As Mario Klingemann artist and commentator on all things data and art put it “I thought I would have at least 10 years left to prepare for the future where everyone has their instant anything digital on demand at home. Looks to me like it’s down to something like five years now.”
Originally published at https://www.linkedin.com.