BeWe11/ai_image_frame

Installation

Raspberry Pi

Turn on ssh functionality on the pi.
For audio support, install the following packages:

sudo apt-get install libasound-dev portaudio19-dev

Run the following command to install all required Pimoroni libraries as well as the inky Python Package

curl https://get.pimoroni.com/inky | bash

Copy a Rhino context file to ~/ai_image_frame onto the pi
Copy .env.dist as ~/ai_image_frame/.env onto the pi, fill out all values
Follow this tutorial to setup audio on the pi. Make sure to increase mic input volume in alsamixer to 100%.

To install the systemd that autostarts the main script:

scp ai_image_frame.service {user}@{pi_location}:

On the pi:

sudo mv ai_image_frame.service /etc/systemd/user
sudo systemctl --user daemon-reload
sudo systemctl --user enable ai_image_frame.service
sudo reboot

You can check script output with

journalctl --user-unit ai_image_frame

After changes, reload the service with

systemctl --user daemon-reload && systemctl --user restart ai_image_frame

Mac

brew install libjpeg
poetry env use 3.9.2
run poetry install

Assets

Most used assets have a free license and are checked into this repository. The following assets are not checked and have to be placed into the approprate paths by the user:

assets/sounds/waiting.wav: An audio file containing music that will be played when the user has to wait, e.g. when images are generated or send to the inky display.

Test microphone

arecord --format=S16_LE --rate=16000 | aplay --format=S16_LE --rate=16000

Deploy changes

Run deploy.sh

Run the program

A run_image_frame_loop script is installed.

TODO

~~Use Stable Diffusion instead of Dall-E~~ -> using the official OpenAI API now
Implement state machine with possibility to go back to main state at any point
Refactor run.py, it's way too imperative
Allow button and voice choices simultaniously
Decrease image frame width, it takes away too many valuable pixels
Put Text-to-label-fitting algorithm into its own function
Maybe: use picovoice porcupine for hotword instead of button press for initial action
Use script arguments to run.py instead of global and environment variables to define behaviour
Remove all sensitive data (pi username etc)
Don't check in assets, only reference in readme that these have to be downloaded

BeWe11 / ai_image_frame