THE’artificial intelligence it’s everywhere and we often use it without knowing it. One of the most “visible” is the generative one, that is, the one that allows you to create content with a simple command. The tools are multiplying in the wake of the success of ChatGpt and allow you to create texts, articles but also editorial plans and simulations of professional figures with the speed of a click on the keyboard. And example: “ChatGpt can you create me a 200 word text that tells the poetics of Leoparti?“. But generative AI is becoming more refined and in a short time it has begun to create “believable” images that are precise and at the same time disturbing.
The photo in this article, to give an example, was created through Grok the AI tool linked to Twitter and wanted by Musk. With a very simple prompt (this is what the instructions to be given to the ai are called) we asked: “Generates an image with a hyper-realistic woman’s face. Must be young with brown hair and dark eyes with Caucasian features and light complexion. The shot must reach behind“. In a few seconds the result is quite satisfactory. If we had asked for a few more details it would have been more precise and defined, but even so a distracted eye who sees it while quickly scrolling on social media would not “register” that it is a fake.
So how can we understand if we have something artificial and created by an AI in front of us? Let’s say right away that i portraits they are the most difficult to recognize. For example, they could be photos manipulated with some special filter, cleaned of imperfections. But in general a good point to start from is that ofextreme perfection. All major ai tend to produce “perfect” images. Ours also looks like a model without any kind of imperfection, a mole, a small vein in one eye or even just hair that is out of place or disproportionate to the head. So the first piece of advice is to be wary of perfection, that is, if the photo seems perfect perhaps it could be artificially generated. If the portrait is extended to other subjects, observe carefully how they interact between them. At the moment the AI is unable, for example, to create the hands correctly and often ends up creating images in which the bodies merge seamlessly.
Another element to consider is the context in which the image is generated. Even if the prompt were precise, for example, “generates a photo of Piazza Duomo in Milan“, there would be errors, elements that don’t add up. At present the AI is unable to exactly reproduce an environment, even if it is famous. That is, it lacks datasets of information that complete its knowledge to generate the image So if we see a little girl in a flooded area, looking at the background can help us understand how much real be the image.
A free and very useful tool for having a real-time check of an image is the one linked to Google Lensa Google tool with which it is possible to carry out reverse searches on images and which return, for example, all the web and social pages where that image was published, but not only. It also gives the possibility to know which is the first page on which the photo appeared. Thus, for example, if the image talks about a given event, perhaps one that occurred in real time, it is possible to discover that the photo had instead been used for another event a few years ago.
We can also use the ai against itself. ChatGpt has a function that allows you to accompany the prompt with an image. Ideally we could ask to generate a similar one or with the same style, but also to try to understand if the image we have was generated by an AI. If we do the test with the disputed image of the little girl and the hurricane Helene ChatGpt lets us know that it is probably from an AI because the little girl’s expressions are too strong especially on the forehead and that the eyes and nose are too perfect. Then he tells us that the dog’s fur is too perfect and too bright in some places.
The ai writes again: “Overall, it seems that the image could be generated by AI, mainly due to the drama and visual rendering of the facial expression, which could be too strong and perfectly processed compared to an authentic photograph“.