In the era of artificial intelligence (AI), the line between work done by humans and work generated by computers has become increasingly blurred. AI’s ability to create photorealistic images has advanced considerably. Without going any further, Boris Eldagsen won the Sony World Photography Awards with a photograph he generated with AI, although he ended up giving up the award.
In recent months we have witnessed the controversies generated by surprising creations such as that of Pope Francis dressed in an elegant white jacket from the Balenciaga brand or the Great Cascadia Earthquake of 2001, which was a hoax generated by AI. The issue is making headlines this week in our country due to much more compromised situations, after several episodes in which minors have used AI tools to create child pornography without the consent of girls their age – friends, classmates and neighbors – and distribute them through WhatsApp.
In the weekly edition of her newsletter in La Vanguardia, Delia Rodríguez analyzes the consequences for our society of this dangerous drift. As things stand, she poses an important question: How can we determine if an image is genuine or created by AI?
An initial way to detect if an image or video has been created using AI is to review the description and comments related to the image. Authors often mention whether their work was generated using this technology. In order to identify it, you must locate keywords such as “IA”, “DALL-E”, “Midjourney” or other names of generative artificial intelligence tools.
Some AI generators, like DALL-E, add watermarks to the images they create. In this case it is a set of five squares in yellow, turquoise, green, red and blue. They are usually subtle and can be located in the corner of the image. However, there are also ways to remove this watermark, so other factors must be taken into account when verifying the images.
It is common for AI-generated images to present strange visual distortions when viewed in detail. Although at first glance they appear to be perfect creations, there are anomalies that help certify that the photograph has been generated with some AI. An example is an excessive number of teeth on human faces, deformed hands or even hands with more than five fingers.
These imperfections can be revealed by zooming in on the image and can also reveal pixelated areas or strange contours or out-of-place shapes.
If an image is newsworthy or appears extraordinary, it is a good idea to perform a reverse image search to determine its origin. If you can’t find it in reputable news sources and it seems too amazing to be true, it may have been generated by AI. As in the case of the Great Cascadia Earthquake of 2001 that we mentioned at the beginning of the article and that circulated on social networks such as Reddit, but did not appear in any reputable media.
There are online tools that can help you detect if an image was created by AI. For example, SynthID is a tool Google created that detects AI-generated images. Another option would be AI or Not, a website that allows you to upload an image and check if it was generated using AI. However, its results are not infallible and it is important that it be used in combination with other methods.
