stable-diffusion.cpp

In the spirit of llama.cpp, stable-diffusion.cpp by leejet implements diffusion model inference in pure C/C++ on the ggml tensor library, with no Python runtime required. Coverage has grown far beyond Stable Diffusion 1.x, 2.x, and SDXL to include SD3 and 3.5, the FLUX family with Kontext editing, Qwen-Image, Z-Image, and video models such as Wan 2.1 and 2.2 and HunyuanVideo, plus LoRA and ControlNet support. Models load in GGUF quantized formats to shrink memory footprints, and compute backends span plain CPU with AVX, AVX2, and AVX512, CUDA, Vulkan, Metal, OpenCL, and SYCL, so the same approach works across NVIDIA, AMD, and Apple hardware or on CPU alone. Prebuilt binaries are published on the releases page, or the project builds from source following docs that cover backend selection, performance tuning, quantization, and per-model setup; a small embedded web UI arrived in 2026 alongside the command line interface. MIT licensed with about 6,600 stars and over 800 commits, it is the standard choice for diffusion inference in constrained or Python-free environments.