The truth about the photo '4 sisters, we are 90 years old'

AI-generated images attract thousands of interactions on Facebook.

"We are four sisters, today we turn 90 years old, we are waiting for your congratulations." The caption was written along with a collage of photos, with four little girls on top and four old ladies on the bottom, all with bright smiles on their faces, attracting more than 45,000 likes, nearly 9,000 comments and more than 500 shares.

Below, there are many posts congratulating the four "old people" on their health and praising their beauty. However, this is an AI drawing. The account that shared it did not even create this image itself, but took it from a foreign source.

Thanks to the development of AI tools such as Copilot, Midjourney or DALL-E, users can create highly realistic images with just one command.

It is not difficult to find AI-generated images on social media. The images often have themes of family and life, appealing to emotions to attract more interactions.

Despite their widespread presence, it is not easy for everyone to distinguish between AI-generated images and regular ones. In addition to simply “like-baiting”, this content can be abused for malicious purposes, such as impersonation and spreading fake news.

Social media "like" tool

On Facebook, Instagram or X, users can easily come across images/videos created by AI. Content such as “We are 4 sisters” or “This girl drew a portrait of her late father but no one supported her”... attracts tens of thousands of interactions and thousands of comments.

The images are mainly related to family, love, sadness... Most of the comments expressed sympathy for the article, most did not recognize the image created by AI. Many accounts posting the article also did not clearly note it.

Some AI-generated images are mistaken for real photos on Facebook.

According to expert Le Cong Minh Khoi, AI content is increasingly appearing on Facebook. Many posts are recommended by the platform with high interaction, but most of them are not clearly noted as being created by AI.

“People with little exposure to technology can hardly tell the difference. Thousands of comments on the post are just sympathetic, like ‘so touching’.

AI has advanced to the point where some new models can correct common errors like hands and teeth. I myself have to re-check the images. Without AI notification labels, users would be almost unrecognizable,” Mr. Khoi shared.

According to a survey published in March 2024 by the University of Waterloo (Canada), only 61% of 260 survey participants could distinguish between AI-generated photos of people and photos of real people, lower than the researchers' expected rate (85%).

Some AI-generated content attracts tens of thousands of interactions.

As a person who is often active on social networks, Mr. Khoi commented that AI images/videos are being abused to increase views, increase interactions or even scam.

“AI content can easily cause viewers to be psychologically manipulated, believe in things that are not true, even be lured into buying poor quality products or believing in edited advertisements, conspiracy theories...”, Mr. Khoi added.

In fact, a study published in May 2024 by Google and Duke University (USA) showed that among the forms of spreading misinformation, the popularity of AI images is almost equal to that of text and regular edited photos.

How to distinguish AI-generated images?

The quality of AI-generated images is getting better and better, making it difficult to distinguish them from real photos. However, users can still rely on some details if they look closely.

AI often generates images based on data from other images, so the tools can struggle with less trained details.

According to security expert Ngo Minh Hieu, hands and fingers are notable details because AI often has difficulty recreating hands, leading to images with extra or deformed fingers.

AI-generated image of actor Ryan Reynolds. Photo: @joinrealmai/X .

Take the photo of actor Ryan Reynolds, for example. While most of the image is realistic, the fingers on Reynolds’ right hand are shaped unnaturally. The text on his jacket is also completely meaningless. This is a clear sign that the image is generated by AI, as it still cannot produce accurate text or logos.

Next, it is possible to track the lighting and shadow effects in the image. If the lighting or shadow angle is unusual, it is likely that the image is generated by AI.

For content that shares the same context, users should check for duplicate details because AI often has difficulty generating completely different elements.

According to the BBC , this is an example of “too perfect”. The image of singer Ariana Grande below looks very real, but if you look closely, you will see that her face is too smooth, making the photo very sharp. The arrangement of the flamingos in the background is also not good.

AI-generated image of singer Ariana Grande. Photo: @agswaffle/Instagram .

If the content is in video format, details to check include lip movements that may not match speech, eyes blinking less or unnaturally, skin texture that is blurry...

Finally, if you are still unsure about a face, hand, or lighting, you should do a reverse search on Google and similar tools to determine the source.

“If you have doubts about the authenticity of an image, you can use a reverse search tool to check if the image has appeared elsewhere before,” Mr. Hieu added.