This is one of the oddest and most curious cameras I’ve ever seen. It has a lens and a sensor and can actually record the real scene in front of you, but it doesn’t do that. Instead, it detects what’s in your scene using AI and turns it into a text description. This text description is then fed into an image-generating AI whereby it makes an image based on that description. And that’s the “photo” you end up with.
As I said, it’s a curious camera and a very interesting concept. The camera itself is very unassuming. Essentially it’s a twin-lens camera, much like a twin-lens reflex (without the reflex bit), in a big black box. It has a waist-level viewfinder on top and the actual camera lens below, along with a big red shutter button on the front. Shortly after you hit that button, it spits out an Instax print of the AI-generated scene.
Created by Jasper van Loenen, it’s called the Black Box Camera for fairly obvious reasons. Inside the big black 3D printed box is a Raspberry Pi 3 to capture the scene and do all the AI to produce your final image. Also housed in the ominous black monolith is a Fuji Instax printer that presents you with the print of your image after it’s been generated by Dall-E 2. Jasper faced a number of challenges during the creation of the camera, including reverse engineering the Instax printer’s Bluetooth protocol, which he has made open source and available to download on GitHub.
Despite releasing the reverse-engineered Instax Bluetooth code, the complete camera doesn’t appear to have been made open source. At least, not yet. But there are enough clues in the description on Jasper’s website to give you an idea of the process and how you might be able to create your own.
The Black Box Camera has an analog viewfinder and a button for the user, everything else is handled by the computer. This new camera interprets the subject you want to capture, generates a unique photo using AI and outputs it as a physical photo print.
This is very much a version 1.0, as I already have some plans for upgrading both the physical device and its internals.
So happy all the different parts needed to make this work finally came together. From AI models and APIs to (a first for me) reverse engineering the printer’s Bluetooth protocol.
– Jasper Van Loenen
It’s interesting to see a camera that sort of mimics the way humans see and remember a scene vs the typically objective way that cameras see them. Or at least, as objective as a traditional camera can be given the limitations of a single viewpoint perspective and field of view. How often do we think back to a place we’ve visited in the past and remember things differently from how they were in the real world?
The Black Box Camera takes a more subjective interpretation of the scene via the use of multiple AIs and captures it in a similar way to how humans record memories. We pick out the bits that are important to us, remember them as best we can, and then our brain fills in the gaps.
At the moment, the camera is “very much a version 1.0”, and Jasper says he plans to make some changes to upgrade the camera and its internal electronics. I’m looking forward to seeing how the project develops in the future and if other similar projects also start to pop up. For now, if you want to learn more about the Black Box Camera and see Jasper’s other projects, be sure to visit Jasper’s website.
Would you use a camera that has a mind of its own?
[via Hackaday]
FIND THIS INTERESTING? SHARE IT WITH YOUR FRIENDS!