Example of using library?

[hammer-ai]

Hi there! I was just going to start using stable-diffusion.cpp in my Electron app directly, but I saw this and it looked like a nice wrapper. Do you have plans to add examples of using it? Would be happy to be an early user if so - thank you!

[lmagder]

There is not much in the way of documentation since this library is basically a 1-1 wrapper of the API in stable-diffusion.h, but there is a simple command line tool here: cli.ts that I mostly wrote to test that the library works.

To be totally honest, I'm not sure how much time I'll have to spend on this since I was kind of just messing around as part of a hobby project, but feel free to use (or fork) it if it helps you! There's actually not much code here, so much of its build script and other packaging trivia. The entire extension is in src/NodeModule.cpp

Just as a heads up, there are a couple of elements here that are challenging which you may want to address, or at least be aware of if you use it:

• The package is constantly hitting up against npmjs package size limits, not a problem if you're distributing it yourself but that's the reason why it downloads the CUDA DLLs from nvidia.com on npm install vs including them. Also that's why it only builds for a single GPU arch (RTX 3000-series) since otherwise the .node file is too large. You can adjust both of these in CMakeLists.txt
• This is more an issue with GGML, which stable-diffusion.cpp uses but they don't support multiple backends (eg AMD, CUDA, Vulkan, etc.) in one binary beyond the CPU fallback. Some similar projects like the node llama.cpp bindings ship multiple extension binaries to work around and this and dynamically pick which one to load. Maybe something you would want to consider doing in your app. This wrapper doesn't care about which backend you use if stable-diffusion.cpp supports it so you could build multiple copies of it.