Samsung Galaxy S23 Ultra: the little secrets of its stunning photos of the Moon

The most high-end smartphone models have for some time prided themselves on being able to photograph the stars and the Moon. An exercise that seems complicated at first sight, even with a traditional case and a long focal lens. When the Samsung Galaxy S23 Ultra was released, many customers shared very cute photos of the Moon captured with the Space Zoom feature on social media. While a diffused white ball appears on the screen, the final shot suggests a satellite teeming with details. But this rendering raises questions.

A member of the Reddit forum, ibreakphotos, looked into the matter, and delivered a detailed analysis of what he considers to be a hoax. According to him, when the Galaxy S23 Ultra’s camera captures an image of the Moon, an algorithm detects the star and replaces the captured image with another to create the illusion. It would therefore not be a specific treatment to improve the image, but a snapshot resulting from an algorithm developed using hundreds of images of the Moon.

To prove his point, he performed a simple but effective test. He first downloaded a high-resolution image of the Moon from the Internet, then reduced it to 170 x 170 pixels and then applied a Gaussian blur to make the details disappear. He displayed it on his full-screen monitor in his office, and turned off the light in the room. He then took a photo of this degraded Moon from the other end of his office with the mode in question… and was able to see that the Moon was displayed perfectly.

The proof that a specific artificial intelligence has been developed for this very particular exercise, and that it “draws” among models when it detects that our natural satellite is targeted. To drive the point home, he carried out the test by displaying the same star twice, but the algorithm was able to manage only one, leaving the second blurred.

An “enrichment” of the initial photo

It should however be specified that the algorithm does not stupidly stick an image of the Moon on the screen, but uses its network to enrich the initial texture. If the Moon of the original photo has a particular color (yellow, red etc…), it will be kept on the final shot. This is also the case here, the initial conditions displaying it in an orange hue. It is therefore not a feat of the sensor and its associated optics. The information provided by the latter is kept, and the rest comes from the machine learningwho trained on hundreds of lunar photos to recognize points of detail and enrich them thanks to his database.

The editor specifies that this type of process concerns only the Moon and that no modification of this type is carried out on the other photographs. If the scene optimizer is deactivated, the images in question are also much less convincing.

This controversy is reminiscent of that linked to Huawei and its P30 Pro. When the smartphone was released in 2019, the Chinese brand touted the prowess of its Moon mode. Later, she explained how her mode worked to Android Authority. “Moon mode recognizes and optimizes details in an image to help users take better photos. It does not replace the image in any way. This would require an unrealistic amount of storage space since AI mode recognizes more than 1300 scenarios. Based on the principles of machine learning, the camera recognizes a scenario and optimizes focus and exposure to improve details such as shapes, colors and highlights / lowlights.then declared the brand.

Samsung therefore seems to offer similar technology, but even more successful. So you have to take this mode for what it is: a cosmetic embellishment, which simulates or “augments” what current photographic equipment certainly cannot capture on its own.