Facebook seems like a scary company sometimes but then they go and do something like this and totally confirm it. TextStyleBrush is a new AI function the social network has been working on, in between attempting to build a mind-reading interface and
hollowing out Mark’s volcano lair teaching robots to teach themselves to navigate.
TextStyleBrush is rather clever
This invention is a little less world-ending, though, unless it manages to convincingly forge a letter from Kim Jong Un to an American president — but that would never happen, would it? No, Facebook’s TextStyleBrush is just an AI that uses just a snippet of text — either as a font or handwritten — and is able to replace all of the text in a given image with the extrapolated visuals.
Not totally accurately, of course, but it’s an impressive skill for a computer to learn nonetheless. It works in much the same way as style brush tools do in word processors (hence the punny name), but with images, and Facebook reckons they’ve taught it to be more accurate in a broader range of situations than previous implementations of the tech.
But why are these skills being learned? Well, that faked letter example? It’s not quite as hilarious as you might think. Facebook AI said in a blog post that “[i]f AI researchers and practitioners can get ahead of adversaries in building this technology, we can learn to better detect this new style of deepfakes and build robust systems to combat them.” So… we need to build the thing people will use to screw others over… in order to stop them from screwing others over. Right. Got it.
There are some less horrifying aspects of TextStyleBrush, those being “…translating text in images to different languages, creating personalized messaging and captions, and maybe one day facilitating real-world translation of street signs using AR.” Which are absolutely fantastic, non-threatening ways to use Facebook AI’s newest technology, at least.