Installation

Diffuse has two parts: the diffuse binary (Rust) and a Python worker that runs the model on PyTorch. The installer sets up both.

Prebuilt binaries are published for Linux (x86_64 and ARM64), macOS (Apple Silicon and Intel) and Windows (x86_64). None of them need a Rust toolchain. Anything else builds from source, which is a single cargo build plus the worker environment; all paths are covered below.

Platform	Binary in the release
Linux x86_64	diffuse-linux-x86_64
Linux aarch64	diffuse-linux-aarch64
macOS Apple Silicon	diffuse-macos-aarch64
macOS Intel	diffuse-macos-x86_64
Windows x86_64	diffuse-windows-x86_64.exe

Each is published with a .sha256 next to it, and the installers check it.

Prerequisites

Requirement	Why
python3 3.10 or newer	runs the model worker
git	fetches the worker source
curl	downloads the binary (prebuilt path)
Rust toolchain	only on platforms with no prebuilt binary
~4 GB free disk	PyTorch and a small model

The worker downloads PyTorch on first setup, which can take a few minutes.

Quick install (Linux and macOS)

One command. It picks the binary for your platform, verifies its checksum, installs the worker, and sets up its Python environment.

curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

Read before you run

The script is install.sh. Read it, then pipe it to bash.

It installs to:

• ~/.local/bin/diffuse — the binary
• ~/.diffuse/worker — the Python worker and its virtual environment

If ~/.local/bin is not on your PATH, add it:

export PATH="$HOME/.local/bin:$PATH"

Linux distributions

Install the prerequisites, then run the quick installer above.

Debian / Ubuntu

sudo apt update
sudo apt install -y python3 python3-venv python3-pip git curl build-essential
curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

Fedora / RHEL

sudo dnf install -y python3 python3-pip git curl gcc gcc-c++ make
curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

Arch

sudo pacman -S --needed python git curl base-devel
curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

macOS

The same command as Linux. It picks the Apple Silicon or the Intel binary from uname -m, so there is nothing to choose.

brew install python git
curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

Gatekeeper

The binary is not notarised, so macOS quarantines it on download. The installer clears that flag for you. If you fetch the binary by hand instead, run xattr -d com.apple.quarantine ./diffuse once.

Windows (PowerShell)

Install Python 3 and Git, then run one line. No Rust toolchain: the .exe is prebuilt.

irm https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.ps1 | iex

It installs to %USERPROFILE%\.local\bin\diffuse.exe, puts the worker in %USERPROFILE%\.diffuse\worker, sets DIFFUSE_WORKER_DIR, and adds the binary directory to your user PATH. Open a new terminal afterwards so the shell picks those up.

ARM and Raspberry Pi

ARM64 has a prebuilt binary, so the quick installer works on Ampere, Graviton, a Raspberry Pi 4/5 running a 64-bit OS, and anything else aarch64. Two notes:

• PyTorch on ARM. The standard pip install torch provides ARM64 CPU wheels. On a Raspberry Pi running a 64-bit OS this works out of the box; a 32-bit OS is not supported.
• Memory. A Raspberry Pi holds only a few layers. That is fine, that is the point. Run it as a host contributing a small slice rather than trying to serve a model alone.

sudo apt install -y python3 python3-venv python3-pip git curl
curl -fsSL https://raw.githubusercontent.com/UnlikedOne/diffuse/main/install.sh | bash

Build from source (any platform)

The universal path, summarized:

git clone https://github.com/UnlikedOne/diffuse.git
cd diffuse

cargo build --release

python3 -m venv worker/.venv
source worker/.venv/bin/activate
pip install --upgrade pip
pip install torch transformers safetensors grpcio grpcio-tools protobuf numpy psutil huggingface_hub
pip install -e "worker[multimodal,diffusion]"
( cd worker && bash scripts/gen_proto.sh )
deactivate

export DIFFUSE_WORKER_DIR="$PWD/worker"
export PATH="$PWD/target/release:$PATH"

DIFFUSE_WORKER_DIR tells the binary where the worker lives. The installer sets this up for you; from a source build you set it yourself.

Verify

diffuse --version
diffuse models

diffuse --version prints the build version derived from the Git tag. diffuse models lists what the network is serving right now.

Next

• Quickstart to send your first prompt.
• Troubleshooting if the worker will not start.