Hard to Distinguish between AI-Generated Fake Faces and Real Headshots

AI art-generating programs are becoming increasingly adept at creating hyper-realistic images. The author of the article has been using AI to create headshot images of completely fake people, mimicking the lighting style of famous photographers. They conducted a social experiment to determine if the average person can distinguish between images of real people photographed by real photographers and those created entirely with AI. The author invites readers to take a quiz and view the results comparing “Real vs AI Headshots.”

There is growing concern among whistleblowers about how artificial intelligence will shape the future for humanity. AI programs like OpenAI’s Chat GPT, Google’s DeepMind, and IBM’s Watson are seen as only the beginning of a race between computer science and human development. The author, however, is more interested in how AI is impacting the creative world, specifically in photography and videography.

AI art-generating programs were initially impressive but lacked the realism to be perceived as a threat to traditional photography. However, over the past year, these programs have significantly improved in reproducing photorealistic images. The author reflects on the rapid progress of AI technology, surprising even those who predicted its development.

The author conducted a social experiment to gauge how far AI-generated imagery has come. They wanted to see if average people, who are not professional photographers, could discern the difference between AI-created images of people and those captured by real photographers. The author used AI software called Stable Diffusion to generate realistic headshots based on the styles of famous photographers Peter Hurley and Martin Schoeller. The process involved using different AI-training models and refining the renderings of human eyes, hair, and skin.

The goal was to create over 100 usable AI-generated headshots. Initially, the author used AI-trained models that perfectly replicated the likeness of celebrities, models, actors, and even porn stars. However, they later scaled back the AI models to create more generic-looking people. For the test, a variety of real photographs taken by professional photographers were mixed with the AI-generated images. The participants were briefed on the photographic styles of Peter Hurley and Martin Schoeller.

To enhance the experience, the author decided to print all the headshots as 4x6 photographs on museum-grade FujiFilm Crystal Archive DP II Matte Paper. This not only brought the colors and details to life but also prevented fingerprints from affecting the viewing experience. The author acknowledges Saal Digital for sponsoring the prints and emphasizes the value of physical prints compared to digital images.

The author conducted the experiment with a small test sample of about 20 people. The preliminary results showed that the average person scored around 40-60% in distinguishing between real and AI-generated headshots. The author plans to gather more participants to obtain a more comprehensive understanding of how AI art-generating programs are impacting people’s perception of reality.

The article concludes by reflecting on the potential implications of AI-generated art blurring the lines between truth and falsehood. If AI can manipulate reality to such an extent, it raises concerns about living in a world where determining truth and honesty becomes increasingly challenging.