One Open Source Project a Day (Part 59): Dream Recorder - Turn Dreams into AI Videos with a Physical Device

Introduction

"Record your dreams. Wake up. Speak. Watch them come to life."

This is Part 59 of the "One Open Source Project a Day" series. Today's project is Dream Recorder (GitHub).

Want to turn your dreams into replayable videos? Dream Recorder is Modem's open-source physical dream recording device: wake up, speak your dream into the microphone, and the system uses OpenAI for text/prompt generation and LumaLabs for AI video, displayed on a 7.9" ultrawide screen. Built with Raspberry Pi 5, capacitive touch, and a 3D-printed enclosure, ~€285 to build, ~$0.15 per dream.

What You'll Learn

Dream Recorder hardware and workflow
Single-tap and double-tap interaction
Hardware BOM and 3D-printed enclosure
Local development and dreamctl
Comparison with software-only solutions

Prerequisites

Basic familiarity with Raspberry Pi and Docker
3D printer (optional, for enclosure)

Project Background

Project Overview

Dream Recorder is Modem's open-source dream recording and AI video generation device. After waking, users speak their dreams into a microphone; the system transcribes speech, generates video prompts, and calls LumaLabs to produce 5-second 21:9 ultrawide videos for playback on the device. Hardware: Raspberry Pi 5, 7.9" 1280×400 ultrawide display, capacitive touch, USB microphone; enclosure is 3D-printable in translucent PLA.

Team

Team: Modem (modemworks.com)
Collaborators: Mark Hinch (software & hardware), Ben Levinas & Joe Tsao (industrial design), Alexis Jamet (illustrations)

Project Data

⭐ GitHub Stars: 1.5k+
🍴 Forks: 115+
📄 License: MIT
🌐 Website: modemworks.com/projects/dreamrecorder
💰 Hardware cost: ~€285 (May 2025)
💰 Per-dream cost: OpenAI < $0.01 + LumaLabs ~$ 0.14

Core Features

Workflow

Speak your dream: Double-tap to start recording → speak into mic → single-tap to finish
AI processing: OpenAI speech-to-text + video prompt → LumaLabs generates 5s 540p 21:9 video (ray-flash-2)
Playback: Single-tap plays latest dream; single-tap during playback = previous dream; double-tap = back to clock mode

Interaction Logic

Action	State	Result
Single tap	Idle	Play latest dream
Single tap	Playing	Play previous dream
Double tap	Playing	Return to clock mode
Double tap	Idle	Start recording
Single tap	Recording	End recording and generate dream

Use Cases

Personal dream journal: Speak right after waking, save as replayable video
Creative inspiration: Visualize ideas from sleep or waking
Hardware + AI: Full example of Raspberry Pi + touch + mic + cloud AI

Quick Start

Hardware BOM (~€285):

Raspberry Pi 5 8GB
Waveshare 7.9" ultrawide 1280×400 (touch)
USB microphone
TTP223B capacitive touch sensor
Power, cables, cooler, MicroSD, etc.

See README shopping list.

Install & configure:

Flash Raspberry Pi OS (64-bit) with Pi Imager, hostname dreamer, enable SSH
SSH in, run raspi-config to enable VNC and set timezone
Connect with RealVNC, set screen orientation
Get OpenAI and LumaLabs API keys
git clone → ./pi_installer.sh → enter keys → sudo reboot

Local development:

git clone https://github.com/modem-works/dream-recorder
cd dream-recorder
cp .env.example .env
cp config.example.json config.json
# Edit .env with API keys
docker compose -f docker-compose.dev.yml build
docker compose -f docker-compose.dev.yml up -d
# Visit http://localhost:5000

3D-Printed Enclosure

Files: 3DAssets/ — dream-recorder_front.stl, dream-recorder_back.stl, pre-sliced G-code
Material: Translucent PLA recommended
Settings: 0.2mm layer height, 0.4mm nozzle, 15% infill, tree supports
Assembly: Press-fit front and back, no glue or screws

Project Advantages

Comparison	Dream Recorder	Software-only	Text diary
Form	Physical device, bedside	Phone/PC app	Pen/notes
Output	AI video	Varies	Text
Interaction	Touch + voice	Touch + voice	Handwrite/type
Cost	~€285 build + API	Subscription/free	Minimal
Open source	MIT, full HW+SW	Varies	N/A

Deep Dive

Tech Stack

Languages: Python, JavaScript, Shell, HTML, CSS
Runtime: Docker Compose, Raspberry Pi OS
AI: OpenAI (speech-to-text, prompts), LumaLabs (video)
Hardware: Raspberry Pi 5, GPIO, TTP223B, USB mic

dreamctl Commands

Command	Description
`./dreamctl config`	Edit config
`./dreamctl test`	Run unit tests
`./dreamctl test-cov`	Test coverage
`./dreamctl gpio-logs`	View GPIO logs
`./dreamctl help`	Help

Management UI

Visit http://dreamer:5000/dreams to view and manage generated dreams.

Cost & APIs

OpenAI: Speech-to-text + video prompts, < $0.01/dream
LumaLabs: 540p 21:9 5s video (ray-flash-2), ~$0.14/dream
Suggested: ~ $5 OpenAI credits, ~$ 20 LumaLabs credits

Project Links and Resources

Official Resources

🌟 GitHub: https://github.com/modem-works/dream-recorder
🌐 Website: modemworks.com/projects/dreamrecorder
📄 Assembly guide: docs/manuals/assembly_guide.pdf
📈 Architecture: docs/diagrams/application_architecture.mmd
🐛 Issues: GitHub Issues

Target Audience

Creatives who want to visualize dreams
Raspberry Pi and hardware DIY enthusiasts
Developers learning AI video + hardware integration
Users with 3D printers for the enclosure

Visit my homepage for more useful knowledge and interesting products