stable-diffusion.cpp
A fork that ships Linux CUDA builds and locks down the server — Pinhole's engine.
- Linux builds upstream doesn't ship
- CUDA 12
- lines of server hardening
- ~60
Pinhole needed an inference engine it could trust on a user’s machine. Upstream stable-diffusion.cpp is excellent, but its server will answer anything that can reach it, and there were no prebuilt Linux CUDA binaries.
What the fork adds
--api-key— Bearer-token auth onsd-server.--reject-origin— refuses requests that come from web pages, so a malicious site can’t drive your GPU.- A release pipeline that builds Linux CPU, Vulkan and CUDA 12, plus Windows CPU, Vulkan and CUDA, and
publishes everything with
SHA256SUMS.
The engine itself is untouched; the fork exists so Pinhole can download a pinned, verified binary.