Here’s how to do it:

Not actually the running process, but the last command line. This does unfortunately add a couple of milliseconds of lag but it’s fairly imperceptible. If you can make this happen faster, please comment on the gist.

Mole with hard hat - Stable Diffusion 2024

A quick note on the mullvad exit node integration feature offered by tailscale.

Tailscale doesn’t allow you to select a different UDP port for the connection to the mullvad node! It’s hardcoded to port 51820.

This is in contrast to the official mullvad client, which lets you use port 53 or a custom port of your choice.

If you’re using this feature in the mullvad client to tunnel out of a network which blocks common VPN ports like 51820, the tailscale integration won’t work! And there will be nothing you can do about it. Tailscale’s relay network doesn’t come into play here, the client just seems content to leave your machine unable to route any packets at all. Presumably there is debug logging to be found somewhere.

There’s an issue open on the tailscale client github repo - go and vote for it so I can tunnel out via mullvad’s servers when I’m in the library rather than a spare VPS.

Cargo crates - Stable Diffusion 2024

I’ve released a rust crate called cargo-open. It provides a handy cargo subcommand that opens an installed crate’s directory in your $EDITOR. It’s modelled on bundle open from Ruby’s Bundler.

To install it, run:

cargo install cargo-open

Then:

cargo open crate-name

Note that this is meant for inspecting a crate’s contents, making changes is not a good idea.

Have fun! 🦀

The two halves of the ZigStar OliZig POE

My zigbee-based IoT empire (mostly turning lights on and off) has historically involved a raspi running zigbee2mqtt, talking to the zigbee network via a CC2652R-based USB dongle called a zzh. Zigbee devices are presented on the mqtt message bus, and can thus be automated however one chooses.

Latterly the zzh has been repurposed as a bluetooth sniffer, so I needed a new TI chip to take over zigbee duties. Enter the OliZig POE. It’s built on an Olimex ESP32-POE, an ESP32 dev board with a PoE ethernet port. On top of this sits a daughter board which is basically the same thing as the zzh. The ESP32 presents the zigbee radio on a TCP socket, and zigbee2mqtt on the pi can connect to it over the LAN. This allows us to place the zigbee coordinator where it can get the best signal, while keeping the pi tucked out of the way.

It works great out of the box using the shipped firmware, with no need to change any settings or re-pair anything, two thumbs up. However we’re now in a situation where most of the work is still happening on the pi, and we’re using the powerful ESP32 SoC just to send the raw zigbee packets over the LAN to be processed by a bunch of javascript code in the form of zigbee2mqtt. Not very tidy. Surely the ESP32 could handle zigbee2mqtt’s job and process the packets locally? Enter tasmota.

Before christmas I finally installed the Stable Diffusion text-to-image model on my local machine. This post features my favourite images from a couple of afternoons spent playing around with it.

Click through to see them all, and a few thoughts on the AI art debate at the bottom - the page might load slowly if you’re not on decent wifi.