AI images have been taking over the world lately with their ultra-realism. While generating AI images can prove to be a useful feature for several individuals, some try to use this tool by creating deepfakes and forged images. People have previously been tricked into thinking that former US President Donald Trump had been arrested or the Pope had developed a new taste in fashion using AI images.
However, there are some countermeasures that individuals can use when it comes to segregating these images from the real ones. Today let’s take a look at five such measures.
Here are five ways to tell if an image is AI-generated
– Check the image title, description and tags
This is probably the easiest way to separate AI images from real ones. With so much paranoia surrounding AI-made visuals, companies that generate such material have been urged to be transparent about their product’s origin. A big part of it involves the image’s metadata. This means that stock agencies who admit AI images in their libraries will label their files as AI-generated. It could be mentioned in either the image title, description or tags. You can easily spot an AI image if you keep a lookout for these specific things.
– Examine for physical distortions
While AI has come a long way in the last few years, it still isn’t perfect. You can also point them out by finding visual physical errors in the image. These anomalies are usually caused by the improper functioning of machine learning algorithms which further affect the creation of the image. These imperfections may include uneven eyes, fake-looking teeth, misshaped fingers and many more.
– Look for garbled text
The algorithms that have been programmed to generate images using AI are also designed to come up with new combinations of existing content. However, for a written language a new combination of letters will likely turn a meaningful sentence into gibberish. This leads to AI images creating storefronts, bedroom posters and street signs with text that looks like alien language. Hence, checking for hidden text in the image’s background can help you detect these images.
– Strange lighting and shadows
A similar theory applies to weird shadows and strange lighting in an AI image. AI will often try to experiment with lighting and shadows when creating an image that could make the scene look unnatural. These imperfections aren’t as obvious as the other ones but can be spotted if you look for them.
– Use AI-detection software
It is one of only a few automated solutions for detecting AI-generated images. Applications such as the Microsoft Video Authenticator and Mayachitra’s GAN detector are a few software that can detect AI-generated visuals. Not all of these software are perfect in real-world usage but it’s safe to say that their performance is only going to improve further in the future.
Comments
0 comment