stable-diffusion.cpp

Why we included this project

stable-diffusion.cpp brings the llama.cpp approach to image and video generation: a pure C/C++ implementation on ggml with no Python runtime and no external dependencies beyond the weights themselves. It runs Stable Diffusion, FLUX, Qwen Image, Wan, and LTX checkpoints, and new models tend to land here quickly, sometimes with day-one support. That makes it a strong fit for on-device or embedded generation, and for anyone who wants to run open diffusion models from a single native binary on CPU or GPU (CUDA, Vulkan, Metal, OpenCL, SYCL). A CLI and an embedded web UI cover straightforward text-to-image and image-editing workflows, and a small server mode can expose generation over HTTP for integration into other tools. Because most checkpoints convert to GGUF and quantized weights are supported, it is also a practical route when memory or VRAM is tight.