dai-hongtao/InkTime

What it solves

InkTime is an AI-powered electronic photo frame that transforms a static photo album into a curated memory gallery. Instead of random playback, it uses AI to understand the content of photos and selectively display the most meaningful memories from "today in history" on an e-ink display.

How it works

The project consists of three main components:

1. Photo Analysis: A Python script uses a Vision Language Model (VLM) to scan a photo album, categorize photos, score them based on "memory value" and aesthetics, and generate a descriptive caption for each.
2. Image Rendering: A second Python script selects the highest-scoring photo from the same date in previous years and renders it into a binary format compatible with the ESP32.
3. Hardware Display: An ESP32-S3 microcontroller wakes up daily, downloads the rendered image from a server, refreshes a 7.3-inch e-ink screen, and returns to deep sleep to conserve battery.

Who it’s for

Users who want a low-power, automated way to relive their digital photo memories through a physical device, and hobbyists interested in combining VLMs, Python servers, and ESP32 hardware.

Highlights

• AI-Curated Selection: Uses VLMs to score and caption photos rather than simple chronological order.
• - Energy Efficient: Designed for long battery life (up to six months on 18650 batteries) via ESP32 deep sleep.
• - Hardware Ready: Includes full PCB designs, BOM lists, and firmware for a 7.3-inch four-color e-ink screen.
• - Flexible Deployment: Supports local VLM services (like LM Studio) or cloud APIs via OpenAI-compatible interfaces.