# Frame
A small e-ink photo frame for our home. It pulls from a self-hosted [Immich](https://immich.app/) library, checks a self-hosted [Home Assistant](https://www.home-assistant.io/) to see if anyone is home, and shows a photo on the [PhotoPainter](https://www.waveshare.com/wiki/PhotoPainter) (Waveshare 7.3" 6-colour panel hooked up to a Raspberry Pi Zero 2W) for everyone to enjoy.
The bottom corners show the photo's capture age and EXIF location.
## Why
Most digital frames either require you to pick and preprocess photos that you put on an SD card or they want to talk to a cloud service like Google Photos. Realistically, you're not going to update the photos more than once a year on the SD card. As for cloud providers, I'd rather not give them access to my cherished memories or let them be held hostage.
Coming home to a photo from earlier that day, or one from five years ago, hits differently when it's just sitting on the wall instead of buried in your phone.
It was a fun afternoon project with Claude Code, a bit of experimenting with different dithering and post-processing, and then fine tuning the photo picking algorithm. Besides powering my frame, I share this repo as a reference and as inspiration for self-hosting, to show that the combination of these services can produce something that otherwise wouldn't have been possible to hack together so easily.
## How it works
`src/display.py` runs every 15 minutes triggered by cron. Each run:
1. Quits if it's between midnight and 7am.
2. Asks Home Assistant whether anyone in `HA_PRESENCE` is home. If not, quits to preserve power and not strain the e-ink unnecessarily.
3. Picks a random photo from Immich. The pool is weighted: ~30% "on this day" memories (10% if only the ±3-day fallback fires), ~18% favourites, ~36% the last 30 days, and ~36% everything else. A 7-day rolling history avoids repeats; orientation match gets 4x the weight of mismatch. Before accepting a candidate, the picker verifies that every detected head fits inside the crop with a small safety margin; rejected candidates are skipped. See [immich.py](./src/lib/immich.py).
4. Crops around any detected faces, boosts contrast and saturation (both lacking on e-ink), dithers down to the 6-colour palette, and pushes it to the panel. The capture age and EXIF location are painted into the bottom corners as white-on-black-stroke text, so dithering can't smear the edges.
## Image pipeline
The two choices that matter most are `face_aware_crop` and Atkinson dithering.
### Cropping
The frame can only be in one orientation at a time, but I didn't want to limit it to only show portrait or landscape photos. So `face_aware_crop` resize-crops to fill the frame while biasing the crop around the faces returned by Immich. A landscape shot with room around the subject usually crops cleanly to portrait this way.
The important guardrail is `heads_fit_in_crop`: before the picker accepts a downloaded candidate, it checks the exact crop window against each face box extended upward to cover the head and padded by `HEAD_SAFETY_MARGIN`. If the crop would cut into any visible padded head area, the photo is rejected and another candidate is tried.
See the following examples from [crop_compare.ipynb](./notebooks/crop_compare.ipynb) that show how the head bounding boxes affect the final crop and which candidates would be accepted or rejected.
### Dithering
The panel can only show six colours: black, white, red, yellow, blue, and green. There's no intensity control like on an LCD, every pixel is one of those six. To get anything legible we have to dither, and there are many algorithms with wildly different running times. [dither_compare.ipynb](./notebooks/dither_compare.ipynb) shows a comparison between a few.
Ultimately, I chose Atkinson dithering which seems to keep the highest contrast without too much artifacting. Pure Python was unusably slow on the Pi Zero, so the inner loop runs through numba with perceptual-weighted (0.299/0.587/0.114) nearest-colour matching. Roughly 100x faster after the first warm cache.
## Setup
To run the project, copy [src/.env.example](./src/.env.example) to `src/.env` and fill it in:
| Variable | Purpose |
| ---------------- | -------------------------------------------------------------- |
| `IMMICH_URL` | Base URL of your Immich server |
| `IMMICH_API_KEY` | Immich API key (Account Settings, API Keys) |
| `HA_URL` | Base URL of your Home Assistant instance |
| `HA_TOKEN` | Home Assistant long-lived access token |
| `HA_PRESENCE` | Comma-separated entity IDs. Any `home` state triggers a render |
| `IMMICH_PEOPLE` | Default people for `--people` (must match Immich person names) |
| `SYNC_TARGET` | rsync target for `./sync.sh`, e.g. `pi@192.168.0.81:~/frame/` |
`.env` is gitignored.
Follow [setup](./setup.md) for the additional setup steps.
## Usage
The script takes the following options:
```sh
python3 src/display.py # uses IMMICH_PEOPLE from the .env file
python3 src/display.py --album "Holiday 2025"
python3 src/display.py --people "Alice,Bob"
python3 src/display.py -o 90 # portrait orientation
python3 src/display.py --saturation 1.5 --contrast 1.1 --gamma 0.85 # change preprocessing settings
```
`display.py` only runs on the Pi (it needs SPI). For off-device experiments see the Jupyter notebooks.
## Learnings
The Pi Zero 2W is overkill for this. It chews through battery if you try to run without cables, and most of the time it's just sitting idle waiting for the next cron tick. It could've been mitigated by connecting an RTC timer to an interrupt pin and using deep sleep in-between, however, the waveshare board didn't have this soldered already and I couldn't be bothered to hack it. Realistically, if I were doing this again for a battery-powered build I'd probably reach for an ESP32 with deep sleep as there would be plenty of time to do the image-processing and dithering in the 15 minute idle period so performance wouldn't be a bottleneck.
A few small reliability quirks worth knowing: the Pi Zero 2W's wifi drops if power-save kicks in, so a separate cron job runs `wifi-check.sh` every 5 minutes to reconnect. Swap is masked and journald set to volatile because SD card writes are the only thing likely to slowly kill this build. The render holds an `flock` so a slow refresh never overlaps the next 15-minute tick.
I would also give [Inky Impression](https://shop.pimoroni.com/products/inky-impression?variant=55186435244411) a try with a custom-made frame for a larger display and perhaps integrated lights, as the e-ink looks a bit muddled in the evening lights. I think a separate light source would be the greatest improvement by far.