Ah, AI art. The technology has developed so phenomenally fast that it feels it can't possibly be only around 12 months since we started talking about the latest generation of text-to-image generators. No doubt, in another year things will have advanced even further, ironing out some of those strange artefacts that appear in AI art. But for now, AI image generators are still responsible for some pretty freaky creations.
If you're curious about working with AI image generators yourself, make sure you see our collection of the best AI art tutorials. In the meantime, as we approach the first anniversary of diffusion models like DALLE-2, Stable Diffusion and Midjourney, we round up the freakiest AI art we've seen this year so far.
01. The pope in a coat
OKAAYYY pic.twitter.com/MliHsksX7LMarch 25, 2023
One of the most terrifyingly convincing AI-generated 'photos' of a celebrity we've seen yet has to be the AI Pope in a coat. Created using Midjourney, this image took the internet by storm and provided the perfect example of just how good image generators have got at replicating the likenesses of famous people. Only a year ago, AI images of even the most-photographed celebs looked quite a bit off, and it didn't take too close a look to realise that they weren't real. But this looked so flawless, that many people fell for it, raising concerns about how we're going to be able to tell fact from fiction online.
02. Woman eating salad
One area of photography that AI-generated images could have a massive impact is stock imagery. We've already seen some of the best stock photo libraries start to incorporate AI, and several brands have been using AI-generated 'models'. A classic trope in stock imagery has been the genre of 'woman eating salad', so how well can the best AI image generators handle that as a prompt? Pretty horrifically, it turns out, if you don't do some serious fine-tuning. A Twitter user shared a series of AI images created via the text-to-image prompt 'Woman laughing with salad', creating a whole new type of horror in the process.
03. Will Smith eating spaghetti
The thing I'm enjoying most about working in AI is the sheer unpredictability of it all 😅Today's surprise: this text-to-video generated monstrosity of Will Smith eating spaghetti 🍝 pic.twitter.com/fTOY4LBTRdMarch 28, 2023
Convincing AI-generated video seems to be still some way off. It's a lot more complex to generate moving images than stills since by definition video requires multiple frames, which need to be consistently composed and lit. For now, AI video tends to look glitchy and contain a lot of flickering, and some of the experiments are more than a little disturbing. "The thing I'm enjoying most about working in AI is the sheer unpredictability of it all," Sam Asante said about this AI video of Will Smith eating spaghetti. It far exceeds any fear we have about how we might eat spaghetti if we ordered it on a first date.
05. The AI live-action South Park
Here's another stab at AI video, this time bringing cartoon characters to life in a live-action South Park. The characters look frighteningly realistic... almost. Created by YouTuber demonflyingfox using Midjourney plus animation software, the video has colours and an image quality that make it look like it could actually have been made around the time South Park appeared back in 1997. But like with many pieces of AI art, there's that uncanny valley feeling that creates a profound sense of unease. All that silent blinking is seriously unnerving.
Want to see more? See our pick of weird AI art for more bizarre gems.
