AI was arguably the most contentious topic in the world of art and design last year, and looks set to retain the same honour in 2023. Text-to-image generators have been causing controversy for a while now – but perhaps lost in all the noise is just how much they've developed in the last 12 months alone.
An Instagram post has revealed the stark contrast between the text-to-image capabilities of AI model Midjourney in March 2022 and March 2023. And the difference is, somewhat terrifyingly, night and day. (Wondering what all the fuss is about? Check out the weirdest images created with AI, and see our pick of the best AI art tutorials and our guide to how to use DALL-E 2 for more on text-to-image generators).
The images were created one year apart, with the exact same text prompt: 'Donald Trump and Barack Obama playing basketball'. And while the first image is a nightmarish blob of barely distinguishable flesh, the second is practically photo-realistic.
"They’re gonna have to make it illegal to have AI image without a watermark saying that it’s AI generated, or this is going to become far out of hand," one worried user comments, while another simply adds, "Insane progress."
It's one of the most tangible signs yet just how fast AI is moving – perhaps it's no wonder that even Elon Musk thinks the tech's progress should be paused. And realistic images isn't the only thing causing controversy in the world of AI, with issues around ethics and copyright plaguing the tech. It's perhaps unsurprising, then, that Adobe is at such paints to point out that its own new AI model, Firefly, is only trained on stock or copyright expired imagery.
The real question is how things will look a year from now. Indeed, if the Trump vs Obama basketball game has changed so dramatically, perhaps that AI-generated Will Smith eating spaghetti video will become terrifyingly real in 2024.