AI image generators have become a source of huge controversy in the creative world, but until now those suspicious of, or directly opposed to, the technology could take solace in one thing. Despite impressive advances, AI was rubbish at accurately rendering human hands.
Hands with an unconventional number of fingers became one of the main giveaways that could tell us that an otherwise realistic-looking image was generated by a text-to-image AI model (see how to use DALL-E 2 for more on the technology). Alas, it seems that AI may now have overcome that hurdle.
Based on the Discord social platform, Midjourney has been one of the most popular AI image generators since it was released just last March. It's recently released version 5 of its model, and of all the enhancements, one in particular is causing celebration – and sowing terror – online. Its improved ability to tackle human hands.
Just a heads-up - Midjourney's AI can now do hands correctly. Be extra critical of any political imagery (especially photography) you see online that is trying to incite a reaction. pic.twitter.com/ebEagrQAQqMarch 16, 2023
"Be extra critical of any political imagery (especially photography) you see online that is trying to incite a reaction,” the 3D character designer @TheCartelDel warned on Twitter. "Eventually it’ll be flawless - scary times," another observer said. And it's certainly getting there.
Midjourney Version 5 is still by no means reliably successful in its rendering of hands. It still produces many, many anatomically bizarre arrangements of limbs and, particularly, fingers. This seems especially so if there is more than one person in the image. While the main subject might now have the right number of fingers, people in the background still tend to have alien anatomies.
Even with the right number of fingers and thumbs, many of the images that Midjourney produces still have that 'uncanny valley' feel. But the fact that it can now generate hands that look natural, at least part of the time, is another milestone that shows how fast AI image generation is advancing.
While DALL-E 2, Stable Diffusion and Midjourney have been the main players in text-to-image AI over the past year or so, Adobe is now getting in on the action too. It's just released Adobe Firefly, an AI image generator that's "designed to be safe for commercial" by correcting the ethical concerns that dog other models.
Meanwhile, as AI image generators continue to advance, we're expecting to see the release of the first tool that aims to protect artists' work from AI. It seems to work – but only for now.