Tiemen Molendijk
December 5, 2022
These were the words of Pink Floyd’s David Gilmour and Roger Waters, welcoming you to the new way of living, with ‘the machine’ in control. And these are the words we’re using to welcome you to a new piece of state-of-the-art technology, which baffles us till this date.
We’re talking about text-to-image generators. You put in a textual prompt describing whatever you want to see, and the computer generates the image you asked for. But are these images good enough to replace designers as a whole? Find out in today’s article!
To start off, the procedure of how text-to-image generators and designers do their work are pretty similar. They both receive textual “prompts” which they then convert into images that accurately display the textual prompt. A creative's arsenal includes a Shutterstock subscription, thousands of hours of Photoshop experience, and a bachelor's degree. The machine’s arsenal consists of the millions of pieces of data that it’s been trained on, and the algorithms that allow it to link those concepts together to create realistic images.
No, the computer does not understand the entered text, but what the software can do is link text input to a meaningful image. And the programme gets more sophisticated every time it is fed new data, so basically it teaches itself to get better at generating the right images. And the interesting thing is that those programs are also becoming publicly available and getting more and more astonishing features. With even TikTok getting in the race, it seems that a lot of big (mainstream) companies are starting to believe that AI image generation can be the next big thing!
Today we’re going to compare the both of them based on a couple of challenges. Is Artificial intelligence ready to take over the creative playfield? Man vs. machine, find out.
Challenge 1: Winning awards
Artificial intelligence sure isn’t able to win art awards that are meant for human artists, right? …Right!? Well, we’ve got news for you, because this year's Colorado State Fair's annual art competition had a startling winner!
A contestant, Jason M. Allen, didn't come in with an artwork he made himself. He made it with the software Midjourney. Jason’s work, “Théâtre D'opéra Spatial”, was awarded the first place in the Emerging Digital Artists Exhibition Competition – making it the first artwork being completely autonomously and artificially created to win an award. But then again, man has been doing this for years now so, both score.
SCORE: Man: 1 | Machine: 1
Challenge 2: Speed
Dall-E 2 recently released a new function called ‘Outpainting’ which can be used to expand existing images like Johannes Vermeer’s ‘Girl with the pearl earring.’ The tool gives the user the ability to change elements of the painting while expanding the canvas, so that different works can be created with the same image. Now we have to admit that a human artist could’ve done this too, but Dall-E generated this result within a minute. We were so impressed by the results that we have to take sides with Dall-E here. Score for the machine!
SCORE: Man: 2 | Machine: 1
Challenge 3: Precision
An interesting question is, whether text-to-image generators can also generate text in images. So let’s put it to the test, here you see a poster generated by the prompt “Soviet realist poster advertising a drink that turns you into a duck.” Again a very impressive and realistic outcome. Just one downside, the text in the poster doesn’t make any sense. This reveals another weakness of the AI software, it generates the text based on images it has seen, and not on logical combinations of alphabetical characters, so basically it recognizes the shapes of letters, but it doesn’t understand the letters themselves or the meaning of the letters, which is needed to create a logical word. So that’s a score for man.
SCORE: Man: 2 | Machine: 2
Challenge 4: Artistic vision
As a tool, it’s incredibly easy to use — just type and wait. Extracting a satisfying image from DALL-E, on the other hand, is no simple task. Without giving it any stylistic guidelines (“digital art” or “surrealist” or “sci-fi” for example) the images DALL-E generates tend to be reminiscent of tragically terrible stock photography.
When generating abstract images based on a specific theme, AI tends to give out surprisingly interesting content. However, without human input and vision, the painting would come without any context. For this challenge, it’s a point for each.
SCORE: Man: 3 | Machine: 3
Ultimately, are creatives at risk?
How will it affect the digital agency world when the software becomes commercially available? We believe it will cause a revolution in the creative field. Anyone could generate professional imagery with a simple search prompt. But we think the question should not be whether AI will fundamentally change the creative workfield. Instead, we should ask ourselves how can we successfully use AI in ways that enable, not replace, the human workforce? How can machine learning make us faster, more efficient, and more productive at our jobs than ever before? We think the efficiency of creatives could be greatly improved thanks to AI, leaving more time for ‘real’ human creativity instead of losing time to simple ‘you ask, we deliver’-creativity. We think it will take some adaptation from creatives to adapt to pieces of software like this rather than creatives being replaced as a whole.