People are asking AI to recreate the same image over and over again, each iteration getting further and further away from the original image.
The results are sometimes interesting and sometimes disturbing. In some cases, these images completely shape shifts into crazy abstract forms. In other cases, facial features are exaggerated.
One of the most viral images of actor Dwayne “The Rock” Johnson has been repeated 101 times.
Reddit’s people told chatgpt Copy the image of the rock without changing anything 100 times pic.twitter.com/icrgmwhcwk
– Kristi Yamaguccimane (@thewapplehouse) May 3, 2025
While the first few iterations are very similar to the original photo, subsequent versions see Johnson’s feature deformation and distortion, which eventually becomes completely abstract.
chatgpt Tip “Create an exact copy of this image, don’t change things” 74 times pic.twitter.com/u6e8avthy2
– Internet Hall of Fame (@internetth0f) May 1, 2025
I tried “create the exact copy of this image, don’t change things”. trend. pic.twitter.com/xcacsvbrjp
– rish agarwal (@rish404) April 30, 2025
So what happened under the hood? This is mainly the result of training of AI models and how they encode and reconstruct images.
When the AI is asked to recreate the image, it not only copies and pastes the original pixels. Instead, it breaks the image into a complex set of features and patterns and then reassembles based on the understanding of the image should look.
However, this process is imperfect in nature and introduces small errors or biases every time. As the images are repeatedly fed into the AI, these biased compounds lead to increasingly distorted or unexpected results.
It’s kind of like playing a visual game of “phone” or “whispering”, where you whisper to the next person will introduce new features.
However, distortions of AI may also reveal information about the biases and assumptions of baking in these models. For example, some images seem to exaggerate facial features or create a warmer, orange palette.
Users also noticed that eyebrows became highly exaggerated – almost drawn in the form of social media filters. As for the orange tones, some speculate that warm tones are preferred in photography, so they are more common in training data.
In fact, though, we don’t know what’s happening in the huge “black box” in today’s largest border model.
But at the same time, social media users seem to have a lot of fun with the surreal, disturbing results generated by repeated recursive AI images.
Trends involving AI, like we recently used AI action characters, have recently stood out on social media, with thousands of people participating in X, Reddit, Tiktok and Instagram.
One joked, “I drained the ocean and copied my image 100 times.” Not to be a buzz, but it’s a good point.
