ESP32 e-paper board turned into a Gemini gadget for speech, pictures and songs
A palm-sized Waveshare ESP32-S3 e-paper board that transcribes speech live and turns it into images, voices or music using Gemini models.


A Waveshare ESP32-S3 e-paper board now listens to you talk, prints a live transcript on its screen, and turns what you said into a picture, a voice or a bit of music. The builder reports it’s gone down well enough that they’re now the favourite uncle.
It’s a thin client. The board streams audio to the Gemini API, and different Gemini models do each job: one for realtime transcription, Flash-Lite to rewrite and route the prompt, another for voice design and speech, Nano Banana 2 Lite for images and Lyria 3 Clip for music. Nothing runs on the chip itself, so it needs wifi and an API key.
The builder works at Google DeepMind, so these are their own models. The code is on Thor’s GitHub, a fork of an earlier Gemini Live ESP32 project. Its README grants no licence for the application code, so check before reusing it.
All credit to Thor 雷神 ⚡️ (@thorwebdev). The full story is at the original source.


