A fresh focus
The Poetry Camera began life as an ‘AI classifier’ and was partly inspired by Kelin’s friend Susi Fu’s Artist and Machine performances examining how artists and computers can learn from each other. “Susie would draw sketches of the person standing in front of her, while a machine – using Raspberry Pi – printed out AI-drawn sketches of the same subject.”
An MIT computer science graduate turned digital product designer, Kelin took charge of the Raspberry Pi prototyping, learning how to solder, plus some basic electronics, while designing her first PCB (a HAT for Raspberry Pi). The initial cardboard design took only a few days to complete.
Industrial designer Ryan, meanwhile, has worked as a toy designer and in a creative technologist-type design role at Google where he learned “Javascript and a bit of Python,” and was introduced to Raspberry Pi for prototyping, which he’d “definitely recommend” for anyone who wants to build hardware prototypes that need a logic layer.
He was “ecstatic to create something from scratch, and prototype it on his home 3D printer”. Having started out as pen and paper sketches, Poetry Camera’s form was created and iterated on in Rhino.
Power play
As “the brain of the whole device,” Raspberry Pi Zero 2 W connects to a Camera Module 3 and a thermal printer via UART. It calls on remote AI models via an API for pointers on poem writing. The idea was that they’d get faster responses this way, but this approach necessitates connecting to Wi-Fi hotspots, which could be challenging depending on where they took Poetry Camera out and about. “In ideal conditions, it works like a well-oiled machine but Wi-Fi networks can be very spotty, especially at crowded events where there’s a lot of signal interference”. Six AA batteries keep everything powered (the thermal printer can drain batteries fast) with a buck converter to step down the voltage for Raspberry Pi. Kelin and Ryan chose Raspberry Pi for its wireless connectivity and the volume of tutorials on how to interface with cameras and thermal printers, starting with Raspberry Pi 3B+ before switching to Zero 2 W “since it hits a sweet spot in terms of small size and fast processing power”. They made use of Adafruit’s Python Thermal Printer library and found ChatGPT “very well versed” – pun hopefully intended – for creating code. “We were able to ask a question in our own naive way and get a custom tailored response that often works right out of the box, instantly.”
The pair are constantly tweaking and updating Poetry Camera. It’s already on version 4, and its creators have been delighted by how well it’s been received. “In the future, we’re looking forward to letting people customise their cameras’ outputs – by updating the poem prompts, or adding images, or using their own servers.”
