AI-generated “photos” and artwork are sweeping the web at eye-opening speeds. Often, you can tell when they’re obviously fake. Other times, however, they walk a tightrope between photorealism and reality, like with those pictures of the Pope in a puffer jacket, the faux Pentagon explosion, and the false depictions of satanic clothing being sold at Target.
If you find yourself doubting every photo these days, you can now get a second opinion from another artificially-intelligent tool (because it takes one to know one), AI or Not. Created by a startup called Optic, this free-to-use web app promises to detect if an image is AI-generated or human-made within seconds.
To use the program, you simply have to upload, or drag and drop, the suspected image into the website. AI or Not—versed on works produced on Stable Diffusion, Midjourney V4, DALL-E, and GAN—will then compare this image against others on the web to find its source, and attempt to determine its authenticity.
All told, Optic claims to be able to check media for AI intervention with 95% accuracy. That figure slightlyvaries between news articles, but this was the number given the last time we checked the website. This means the feature isn’t 100% bulletproof; according to Gizmodo, when it ran the bogus image depicting a blast near the Pentagon, the system reported that it was “generated by Human.”
In addition, AI or Not isn’t optimized to properly scan artwork churned out on the latest Midjourney 5 yet.
Skeptic users who wish to authenticate images at the drop of a hat may be glad to know that Optic is working out a Google Chrome extension to streamline their checks. There’s also a Telegram bot by the startup that fulfills the same task.
Below are a couple of works we created on Midjourney, along with one stock photo, that we ran against AI Or Not. You can give the tool a go yourself here.
Screenshot of Midjourney-generated artwork, processed on AI or Not
Screenshot of Midjourney-generated artwork, processed on AI or Not