Black Box Camera – The ‘dark chamber’ of AI

Created by Jasper van Loenen, the Black Box Camera interprets the subject you want to capture in your photo and uses artificial intelligence to generate a unique physical print based on its own interpretation of that subject.

The device has an analog viewfinder to find your subject, and a single button to take the photo; everything else is handled by the internal system. Using machine learning the photo you took is analyzed and a caption is created, which in turn is used to generate a ‘photo’. This final image is presented to the user as a physical photo print, straight from the camera.

Machine learning models creating images from textual prompts have sprouted many ‘prompt masters’; people who know exactly how to create prompts that nudge the system in the right direction. The Black Box Camera takes a different approach by removing the tweaking and changing, and only allowing the user a single input: the subject the camera is pointed towards.

After pressing the button, the entire internal process is (as the name of the device suggests) hidden from the user until the final result emerges: a physical photo print with the generated image.

Internally the Black Box Camera uses a Raspberry Pi with camera module to take a photo when the user pressed the button. This photo is then analyzed and a caption is generated, which is used as the input prompt for Dall-e. The resulting image then gets printed by the internal Instax portable photo printer, for which the bluetooth protocol was reverse engineered to control it from custom software.

Project Page | Jasper van Loenen

See /?s=camera for more ‘camera’ related projects on CAN




Leave a Reply