Technology

How to spot AI-generated images

Quoc Duong DNUM_BJZACZCACF 15:15

Images generated by artificial intelligence (AI) are becoming increasingly popular on social media. However, they always leave a distinctive mark. Nghe An Newspaper guides 7 ways to detect images generated by AI

AI-Generated Images – Benefits and Risks

Images generated by artificial intelligence (AI) are becoming increasingly common on social media. But they always leave behind a telltale trail, from odd details like extra fingers to nonsensical text. As fake content proliferates, detecting it becomes more important than ever.

Fake images generated by AI are spreading at a dizzying rate, creating a new form of ‘tech junk’ to attract clicks. Although GenAI technology has only been around since 2022, there are already more than 15 billion AI-generated images circulating on the internet.

According to the latest data from OpenAI, more than 2 million images are generated using DALL-E 2 every day. In 15 months, the tool alone has generated 916 million images. Most of them are entertaining and unrealistic, such as images of Jesus made of shrimp, animals with human-like features, etc.

While many AI images are still humorous or surreal, the technology is advancing rapidly and is producing increasingly realistic images. This has led to many controversial situations.

In 2024, Republican political activist Amy Kramer shared an AI image of a girl holding a puppy after Hurricane Helene, using it as evidence to criticize President Joe Biden. Also recently, a French woman was scammed out of 830,000 euros with a sophisticated AI image of actor Brad Pitt.

Even though AI image quality is getting better, there are still ways to identify artificial content. Nghe An Newspaper guides 7 ways to detect images created by AI

1. The sign lies in the human form

One of the most obvious signs is in the human form. AI often has trouble replicating hands, resulting in fingers that are too long, too short, or the wrong number (sometimes just four, sometimes six or seven).

Additionally, AI can also produce unnatural body proportions, with disproportionate limbs or asymmetrical faces. Eyes are another common weak point, as they can appear misaligned, reflect light incorrectly or appear lifeless despite the overall image appearing realistic.

2. Nonsense text

One of the telltale signs of AI-generated images is gibberish or unreadable text. AI models often struggle to transcribe text accurately, resulting in gibberish.

For example, signs, billboards, and product labels often contain jumbled or nonsensical characters in AI-generated images. A case in point is the infamous Willy Wonka Experience event in Glasgow, where AI-generated posters featured odd phrases like “encherining" (could be "enchanting") or "Cartchy Tuns" (probably "catchy tunes").

3. Unnatural shadows and reflections

Light and shadow are the elements that distinguish real images from AI-generated images. In natural photos, light always follows the laws of physics, but AI sometimes creates shadows and reflections that do not match the actual light source. This makes the photo look strange and unreal.

Some AI-generated images are too evenly lit, making them look plasticky, or too glossy, making them feel more Disney-style than real. Additionally, AI images often lack the natural imperfections inherent in real-life photography.

4. Unusual Perfection

Another characteristic of AI photos is that they are too perfect in an unnatural way. In real-life photography, especially in low-light situations, digital noise often appears. However, AI either completely removes this noise or evenly creates fake noise, making the photo look unrealistic.

AI also tends to over-smooth textures, making human skin, object surfaces, or fabrics look unnaturally smooth. In some cases, AI portraits look like they’ve been over-airbrushed, obliterating details like pores and wrinkles that are present in real faces.

5. Unnatural repeating patterns

Another sign of AI images is repetitive details. AI often has difficulty creating completely different elements, resulting in subtle duplication or repetition of details.

For example, in images containing crowds, AI can create people with nearly identical faces. Or in the background of a photo, a pattern can appear multiple times in an unusual way, instead of having the natural variation that would be present in a real photo.

6. Check the context of the image

Even if an AI photo looks convincing, its context may still contain incongruities.

AI-generated images often include details that don’t match, like road signs that are the wrong color, architecture that doesn’t match the real place, or light reflections that don’t match reality. These can be “red flags” for fake photos.

7. Verify authenticity using image search tools

If an image looks suspicious, users can use AI image detection tools. While these tools are not perfect, they can help check whether an image is AI-generated.

Another way to verify is to performimage searchreverse engineer the image on the internet. This helps determine whether the image has appeared elsewhere before, thereby checking its authenticity and origin.

AI-generated images are getting more sophisticated, but there are still features that make them recognizable. By paying attention to text details, lighting, texture, repeating patterns, and context, we can avoid being fooled by fake content on social media.

Featured Nghe An Newspaper

Latest

x
How to spot AI-generated images
POWERED BYONECMS- A PRODUCT OFNEKO