“You don’t know anything specific” is the High German version of a Bavarian idiom. Even the double negation indicates a very concrete fact, and this is also the case with supposed photos that are taken of our earth satellite with newer Samsung smartphones. Supposedly because the device’s software still makes respectable images from the fuzziest pixel mass.
A simple experiment with a real, blurred image of the moon, which was photographed from a monitor with a cell phone, shows what is apparently behind it: details appear that were not there in the original. It is unclear whether Samsung is using an AI in the phone, retrieving existing images from its cloud, or both. That’s why you don’t know exactly, because nobody has really looked into the software. This is also difficult because of the many firmwares from camera modules to radio modems with proprietary code and encryption.
The question of which picture is really “real” is becoming more and more urgent. For press photos, the Content Authenticity Initiative (CAI), which is only slowly getting off the ground, should be remembered. With a kind of authenticity seal, all changes to an image should be traceable, right up to the retrieval of the original recording. As the latest heavyweight, the CAI also joined Canon at the beginning of 2023, but again without naming specific dates or cameras that should support the functions.
Beauty filters change self-perception
In the most widespread cameras, those in smartphones, nothing can be expected from such mechanisms, manufacturers are not interested in them. In any case, as with CAI, one would also have to include the online platforms. And on them a completely unregulated technical race is raging, among other things, for the most blatant beauty filters. The latest example is “Bold Glamour”, with which anyone on TikTok can transform into a fully styled supermodel. This now works deceptively real in real time for videos. Tate Ryan Mosley aptly described it for Technology Review as “a mass experiment on girls and young women”. A recommended long read for the weekend.
Of course, behind such filters there is always machine learning, aka AI. And this is now linked and integrated across all media types. While GPT was originally just a large language model, i.e. intended for written language, GPT-4 is now also able to understand the content of images or to analyze it as a template for a text. This means that the AI works in both directions, or that the developers have now also activated these paths for the users.
Midjourney calculates almost perfect portraits
The technology, also known as “Generative AI”, is also making great progress with artistic images and those that look like a designed photo. Here, too, user input may have played a decisive role, because Version 4 of Midjourney, which was only presented in November 2022, generated render images that were chic but recognizable as such. Human facial features and movements were recognizable, but everything appeared with a certain plastic look. Version 5, released this week, creates, among other things, more realistic textures, such as human skin and hair. It’s still a bit too perfect for the trained eye of the photographer, but compare that to the social media filters: many people have long since become accustomed to the artificial look of images.
Unfortunately, the brutal violence of the First World War is not artificial at all, but is presented in frightening realism, as shown in the remake of “Nothing New in the West”. As the first German film – well, with a lot of money from Netflix – the epic won four Oscars, including those for camera and production design. If you compare the depressing scenes with the opening sequence of “Saving Private Ryan”, which is 25 years older and filmed in analogue mode, you will notice a digital look, which skilfully supports the statements of the anti-war film. As in social media, the distribution medium apparently has an influence on viewing habits: a Netflix production has to look exactly like this today.
Canon supports M series and announces EOS R1
The film says “Cut!” so now we’re going to make a hard cut to what’s been on this week’s new photo hardware. The focus here is Canon, more precisely what came up in the background discussions at the CP+ trade fair a few weeks ago. On the one hand, Takeshi Tokura, vice of the imaging department, announced that new lenses for the RF mount should appear continuously. With around 30 optics for mirrorless Canons, there is currently only half the selection of the older EF bayonet. Seven to eight lentils are to be released each year, so that a tie would be reached in 2028.
On the other hand, the statements made by Tetsuji Kiyomi, Canon’s imaging product manager, are not quite as clear. He was, along with other Canon executives, from the French photo site Phototrend interviewed. The conversation appeared two weeks ago, but he is now making one statement from it: The EOS M series in the APS-C form factor should therefore live on for a little longer. We also assumed with the EOS R50 that the M series could have come to an end. However, if you put Kiyomi’s sentences through both human and machine translators, he only said that the M-series is still in high demand and will therefore remain on offer. There is no talk of new devices. As with the gradual discontinuation of a drug with ever lower doses, Canon will probably gradually phase out the Ms according to the current status.
Canon manager Go Tokura then became refreshingly specific at the same point: There will be an EOS R1. That should then be the professional block, sorry, the flagship of mirrorless cameras. Although – with two movements and over 6,000 euros for the body, the previous R3 is clearly a tool for professional photographers. In the fall of 2023, however, the camera will be two years old, so perhaps there will not be a Mark II of the R3, but an R1 as its successor. And then with more than 24 megapixels, because Nikon with the Z9 with 45 megapixels and Sony with the Alpha 7R V with 61 megapixels have now clearly overtaken Canon.
(huh)
