DALL-E 2 Test: Can you tell the difference between human and AI art?

DALL-E 2 Test: Can you tell the difference between human and AI art?

Image: thisimagedoesnotexist.com

The article can only be displayed with activated JavaScript. Please enable JavaScript in your browser and reload the page.

DALL-E 2 and similar AI systems generate authentic images. Can you distinguish between human and machine art?

AI research and industry have made great strides in developing multimodal AI models over the last year. At the beginning of 2021, OpenAIs DALL-E 1 showed where the path would lead: The AI ​​generates fascinating, surreal, photorealistic and all other desired motifs on text input.

About a year later, these advances in multimodal models enabled OpenAI’s DALL-E 2. The artificial intelligence easily outperforms the previous model and continuously generates impressive images in many styles and for many subjects. At the end of May, Google showed Imagen, a generative image AI that even overtakes DALL-E 2 in some areas. In both cases, a brief text entry is enough – and two minutes later the picture is there.

AI Art: Can humans still tell the difference?

To draw attention to these advances, Sahar Mor, Stripe product manager and AI enthusiast, launched thisimagedoesnotexist.com. Visitors can test whether they can still distinguish between man-made and machine-made images.

For the website, inspired by the classic AI fake website thispersondoesnotexist.com, he collected several hundred DALL-E 2 generated as well as human generated images and created a simple voting process: Does the image come from DALL-E 2 or does it originate in a human brain?

A robot with a hat. Could be from DALL-E 2. Or?

Machine fried chicken? Who knows.

Sahar describes his website as a visual Turing test: It shows 30 images one after the other that are either AI-generated or human-made. For each picture, visitors vote for one of the two options and then see if they are correct. For AI-generated images, the website still shows the text input used for generation.

DALL-E 2: It looks (not) good

In the first week after the website was launched, more than 40,000 visitors from more than 100 countries voted more than 400,000 times. The average score is 18 out of 30. On average, visitors were only able to correctly assign the pictures in just over half of the cases. This is little better than coincidence.


This is good news for OpenAI. For some people who earn their living with graphics, illustrations and photos, probably not.

Visitors were particularly often wrong about the two images shown above: 74 percent thought the robot with the hat (image source: Flat Wave) was AI-generated, 82 percent thought the portion of fried chicken was a real photo. It’s the other way around. The photo was generated by DALL-E 2 using the input “KFC Original Recipe Chicken, fresh, hot and juicy, 8K high resolution, studio lighting”.

Here you can test your own AI image detection talent.

#DALLE #Test #difference #human #art

Leave a Comment

Your email address will not be published.