Arduboy's $149 AI Candle Recreates Real Flames With Just 144 LEDs
Creator Kevin Bates has built an open-source electronic candle that uses a tiny neural network to generate constantly changing flames instead of playing a looping animation.
Technology news
A digital candle might seem like an unlikely place to find artificial intelligence, but Arduboy creator Kevin Bates has turned a simple lighting accessory into an unusual demonstration of on-device machine learning. His new $149 Digital Candle uses 144 amber LEDs and a compact neural network to reproduce the shifting appearance of a real flame. Unlike conventional electronic candles that rely on repeating lighting patterns, this device continuously generates its own animation, creating a different sequence of movements every time it runs.
The hardware is deliberately minimal. The candle measures approximately 10 inches tall and features a 9-by-16 matrix of individually addressable LEDs mounted near the top of a narrow circuit board. An Espressif ESP32-S3 microcontroller handles the simulation, while a USB-C connection supplies power through an included dock. There is no integrated rechargeable battery, so the candle needs an external power source. Buyers can choose between black and white circuit-board versions, both listed at $149 on Arduboy's official store.
The underlying AI model is particularly small. According to Arduboy's published specifications, it uses an eight-dimensional latent representation and a convolutional decoder containing just 3,545 parameters. The system was trained using recorded footage of an actual candle flame and produces its animation at 30 frames per second. That approach allows the device to create continuously varying lighting patterns without relying on a predetermined video loop. The result demonstrates how compact machine-learning models can generate dynamic visual effects directly on inexpensive microcontroller hardware.
Bates is best known for creating the Arduboy, a credit-card-sized gaming handheld built around open hardware and software. The Digital Candle follows a similar philosophy. Arduboy provides open-source development tools, and the ESP32-S3 includes Wi-Fi and Bluetooth Low Energy capabilities. Owners interested in programming can use the hardware for applications beyond simulated fire, including clocks, notification displays, and experimental visualizations. This flexibility gives the device a second identity as a small development platform rather than merely an electronic decoration.
The main limitation is its price. At $149, the Digital Candle is an expensive way to imitate an object that normally requires little more than wax and a wick. Its coarse LED matrix also cannot reproduce the full visual detail or warmth of an actual flame. Nevertheless, the absence of combustion makes it an alternative for environments where open flames are undesirable. Its more interesting contribution may be technical: a demonstration that neural-network-generated visuals can run continuously on a compact embedded device without requiring a cloud AI service or a powerful graphics processor.