It is increasingly difficult to distinguish the images generated by AI: this website promises to tell the truth (but it has a trick)

We are witnessing live the evolution of generative artificial intelligence. About a year ago we were surprised by DALL·E’s ability to create so picturesque pictures like that of a shiba inu breed dog with beret and turtleneck or a Salvador Dalí painting with a robotic-looking part of the face.

But they were just that. Picturesque images.

It was extremely unlikely that any of these artificial intelligence-driven creations would end up masquerading as a photograph. Although, sometimes, photorealistic results could be obtained, certain details in the face or in the hands of the individuals evidenced the presence of generative algorithms, but this has just changed.

A possible solution to the great problem of AI

Recently, images of Pope Francis in a white coat began to circulate on social networks, Donald Trump being arrested by the New York Police and even by Emmanuel Macron in the middle of a protest in some French city. These images had in common that they were all completely false, but also plausible.

As you may imagine, it is increasingly difficult to distinguish whether content has been created using AI tools or is actually real images. This problem, which is emerging like the visible tip of an iceberg, is causing concern because it is an open door to fake news and misinformation.

Along with this phenomenon, however, there are those who are already trying to provide solutions. On the one hand, we have the European Union showing itself in favor of mark the content generated by AI, and on the other hand private actors such as Optic, who bet on the idea of ​​a detector of images generated by AI publicly available on the web.

The company has just introduced AI or Not, an analysis system that, as its name indicates, allows us to identify if the visual content that we have loaded has been created using an AI. We just have to drag a JPEG or PNG image and wait for the analysis result that will be presented to us in percentage values.

In our tests we have uploaded a photograph of a Red Panda captured with a Nikon D7000 camera and Optic’s algorithms have indicated that they are 87% sure that the image it was not generated by AI. We have also tested with one generated by AI, in this case, the system believes in 77% that it has been generated with DALL·E (which has been the case).

The creators of AI or Not explain that the system is more accurate with Stable Diffusion, Midjourney and DALL E, although their algorithms can also detect “signs of AI generation” in images from other models, although the accuracy may decrease. The good news is that, for now, using the detector is totally free.

Images: Javier Marquez with DALL·E | Screenshots | Mathias Appell

In Xataka: Adobe Firefly Tried To Compete With Midjourney – Backfired (But For Good Reason)

(function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)[0]; if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript); } })();

–
The news

It is increasingly difficult to distinguish the images generated by AI: this website promises to tell the truth (but it has a trick)

was originally posted on

xataka

by Javier Marquez.

Source link