This Neural Network Instant Camera Turns Everything You Shoot Into Cartoon Doodles

Photographers who still shoot on film claim they like the lo-fi, grainy aesthetics of the medium. But by that logic, they should be throwing hundreds of dollars at Dan Macnish who created a custom instant camera that converts what the camera’s lens sees into crude, stick figure doodles.

Recent Video

This browser does not support the video element.

Remember Google’s Quick, Draw! online game that used a neural network to attempt to guess what people were trying to draw? That game ended up producing a database of over 50 million identified stick figure drawings that Macnish was able to harness to help power his camera, which runs on a Raspberry Pi, with a thermal printer and some other electronics that help it function not unlike an iconic Polaroid does.


But unlike a Polaroid, Macnish’s camera doesn’t have a viewfinder, or a preview screen. You blindly point it at a scene, push the trigger, and out pops a small doodle, not a photo, from the thermal printer.

The doodles the camera creates also don’t necessarily mirror what was being photographed in the first place. The resulting images instead reflect how the background neural network powering the camera interpreted the scene coming through the lens. A selfie could result in a doodle of a bicycle wheel, and your dog might become a rocket ship. But at least the photographer doesn’t have to fiddle with focus, shutter speed, or depth of field.


[Dan Macnish via Hackaday]


Read more!
Want Gizmodo’s email newsletter?