What has long been true, because Photoshop has been able to manipulate images for some time, becomes even more obvious with AI: Don't trust any image you didn't fake yourself. Therefore, let me say it right at the beginning: All images in this news story were created using artificial intelligence.

AI (Artificial Intelligence) is on everyone's lips. My curiosity piqued, and I experimented with Midjourney , one of the leading tools for translating an image description into a photo. You enter the most detailed image description possible in English, and the system generates four versions of an image. The source for this is apparently five billion photos circulating on the internet. The whole thing currently operates in a legal vacuum, because with the exception of one provider, the tools access images for which they have no right to use. This isn't (yet) a problem for experiments, but what happens if a company or brand uses AI-generated images in which one or more of the original images are still clearly recognizable? It's only a matter of time before courts have to address the question of whether an AI-generated image is a completely new work or whether the authors of the photos used must be paid a license fee.

Depending on the subject, the images, or at least parts of them, still look very artificial. Sometimes things aren't quite right—take a closer look at the hands in the last image, for example (click on the image for a larger, full view). I'm pretty sure this will improve in the future, and my still relatively limited knowledge of AI image generation probably also influences the image quality.

Why the Pope and the Dalai Lama?

I took the liberty—and please forgive me—of using the Pope and the Dalai Lama as a starting point for this article. I wanted to create an image that could exist. This earned me criticism from a dear friend, whom I greatly respect. He rightly said that this image could spread widely on the internet, and at some point the image would be so taken out of context that people would no longer know it was an AI-generated image. He suggested labeling the image as an AI photo, but unfortunately, AI is very capable of removing any annotations from an image. The latest versions of Adobe Photoshop demonstrate how frighteningly and fascinatingly advanced image editing programs have become in this area. Another acquaintance liked the idea of the two spiritual leaders and suggested generating a similar image with Putin and Zelensky—as a contribution to the peace process... This feedback demonstrates the great danger of this technology. No one knows anymore what is real and what isn't. Currently, AI-generated images still look somewhat artificial, and as mentioned above, they occasionally contain minor errors. Therefore, upon closer inspection, it's often possible to determine whether the image is a real photograph or an AI image. But in my opinion, it's only a matter of time before AI-generated images are no longer distinguishable from original photographs.

What does this do to us as a society?

When I talk to friends about AI-generated images or show them how to generate photos with AI, I feel two emotions: fascination and frustration. Fascination because it's exciting that you can describe a scene in words and have the AI generate an image from it. The sky (or rather, your own imagination) is the limit – a hackneyed saying that holds true here too. We take the wings of technology to soar ever higher. But after some reflection, my conversation partners often experience the Icarus effect: a crash. They then formulate sentences like, "Now I can no longer trust a picture." Image editing programs have already greatly reduced belief in photos as a reflection of reality. And it seems to me that AI is completely destroying what little faith remained. In a society already unsettled by the events of recent years and the tense global situation, this new technology is further increasing uncertainty. Due to the abuse of extremist groups, absolute values are frowned upon, but now that everything is becoming relative, we feel a sense of loss because there seems to be no stability left. Perhaps that's a bit dramatic, and AI-based images are perhaps just one piece of the puzzle in this whole web of uncertainty, but it seems to me that this artificial generation of images suddenly makes the uncertainty more tangible and real. We are facing major challenges as a society, and technology—which also brings many positive aspects, let me explicitly mention this here—certainly cannot be stopped. We must not lose our authenticity as humans and as a society. Much has been and continues to be written about this, and it challenges each and every one of us to confront the question of authenticity. Interestingly, the photography industry has seen this development coming and launched an exciting initiative:

Content Credentials

Content Credentials is an organization that, as far as I understand, encodes the RAW file in the camera, which identifies the original file as an unedited original. This way, you can prove at any time whether an image has been manipulated or not. In the future, photographers will have the option of having the Content Credentials code added to their images as they are taken. The first company to incorporate Content Credentials into a camera is Leica, in the Leica M11P. You can find out how it all works here.

The results of my experiment

As a curious and open-minded person who works in advertising and photography, this technology is definitely important enough for me to pursue it. The trick lies in formulating the so-called "prompt." "Prompt" is the name of the image description you enter. My first experiments with AI-based image generation showed me that even small changes in the text can have a major impact on the image. "Prompting" is a kind of programming language in which you describe the images. The trick is that sometimes you have an image in your head, and then the AI generates something completely different. While you can upload your own images as style templates to Midjourney, my first test with an uploaded image showed that while the style of the image was somewhat retained, I otherwise didn't recognize my photo in the AI-generated version (which is very good for users regarding copyright issues...).

Here are the first results of my experiments with Midjourney: