Smartphones dream of stochastic moons
Smartphones dream of stochastic moons
A close-up of the moon reveals the evolution of photography at a crossroads
In March 2023, the old conversation on smartphone photography getting better and better took an interesting turn when a Reddit post proved that pictures of the moon taken with latest Samsung smartphones were actually too good to be real.1 Brilliant close-ups of the moon, impossible by conventional physics, were only achieved through re-texturing the image with the help of machine learning. The Verge took notice and illuminated the implications of the finding beyond being a marketing fail for Samsung: AI enhancements provoke unease because they are challenging our concept of authentic photography.2 Any digital camera leverages quite a bit of embedded post-processing, but painting on blurry images with a neural network's recollection of the moon takes it from modifying pixels on the micro level towards generating entire macro sections of an image.
Samsung crossed a line that might be less controversial in the near future. While AI tools creep into the everyday world, our understanding of authenticity is evolving. Enhanced low-light shots don’t bother anyone, deep fakes are scaring everybody. AI-filled TikTok backgrounds are controversial, marking the frontier of a development towards individualist expression with hyper customized artifacts and virtual personas on social media.
Meanwhile dedicated stills cameras, once deemed obsolete, enjoy an unexpected renaissance. Fujifilm introduced a retro-style digital stills camera with huge success3, Pentax developed the first analog stills camera since decades4 and Kodak is offering an all new movie camera for analog filmmaking5. The rising prices of analog film stock seem to be the main factor holding back a trend of revived interest in capturing impressions of reality without additional layers of big data processing.
The evolution of photography currently seems to differentiate into separate strains. Classic photography stays focused on the capturing of light and is flourishing in its niche, meanwhile generative photography is just getting started. AI performance will likely become a more important driver for Smartphone cameras than optical characteristics, when tools to declutter your living room for a selfie turn out to be more convenient than concerning for many. I assume future smartphones will move further and further away from being “as good as a Leica”, towards being considered for what they actually are: powerful processing units, constantly communicating with even more powerful processing units in the cloud.
Already today, classic photography and generative photography are co-existing tools, influencing each other in more nuanced ways than the over-used frameworks of disruption and obsolescence could imagine. Projects like Quantum Mirror and Paragraphica camera are exploring the boundaries between these two flavors of creating images in exciting ways. After all, capturing images and creating images may not be that different. Both are manufacturing an impression of perceived reality by putting things into a frame and keeping other things out of it. Both are a way of answering the world “I was here” without being asked. Both have the potential to be amazing or annoying. And nothing will stop some company from selling a crappy retro-style stills camera with embedded AI eventually.
Many, including me, are exploring the possibilities and pitfalls of generative AI right now. Doing so I wonder if having machines create endless versions of reality for us will free our minds more than prejudgements hidden in the datasets will trap us in an eternal prison of inconsequential variation.
“The moon pictures from Samsung are fake. Samsung's marketing is deceptive. It is adding detail where there is none (...) Samsung is using AI/ML (neural network trained on 100s of images of the moon) to recover/add the texture of the moon on your moon pictures, and while some think that's your camera's capability, it's actually not.”Reddit: Samsung "space zoom" moon shots are fake, and here is the proof
“For photography, the standard of “realness” is usually defined by the information received by an optical sensor: the light captured when you take the photo.(...) In this particular case, though, the Moon images captured by Samsung’s phone seem less the result of optical data and more the product of a computational process. In other words: it’s a generated image more than a photo. (..) In time, we’ll probably forget we ever called such images fake.” The Verge: Samsung caught faking zoom photos of the Moon
"Despite the latest X100VI only being released a couple of months ago, the backlog has already grown to immense proportions. (...) Georghiades went on to discuss how they have increased production to 15,000 units. However, rumors suggest that the current backlog is more than 500,000 units. If this is true, it could take up to 3 years before the current backlog is cleared, and by then we might even have the X100VII launched." Fujifilm Explains Why The X100VI Is Still Out of Stock | Fstoppers