What is this?
A pure C/C++ inference engine for diffusion models like Stable Diffusion, Flux, Wan, Qwen Image, and Z-Image, built the same way as llama.cpp so you can generate images and videos without heavy Python dependencies.
A pure C/C++ inference engine for diffusion models like Stable Diffusion, Flux, Wan, Qwen Image, and Z-Image, built the same way as llama.cpp so you can generate images and videos without heavy Python dependencies.Is this a library or a ready-to-use app?It is primarily a command-line tool and library (same philosophy as llama.cpp), but it also ships a brand-new embedded web UI and has third-party GUIs and language bindings (Go, Python, Rust, C#, Dart) built on top of it.
It is primarily a command-line tool and library (same philosophy as llama.cpp), but it also ships a brand-new embedded web UI and has third-party GUIs and language bindings (Go, Python, Rust, C#, Dart) built on top of it.How do I get it running?Download a pre-built 'sd' executable from the releases page (or build from source via the build guide), download model weights (.ckpt/.safetensors/.gguf), then run the generate command with your prompt.
Download a pre-built 'sd' executable from the releases page (or build from source via the build guide), download model weights (.ckpt/.safetensors/.gguf), then run the generate command with your prompt.Do I need Python or any ML framework to use it?No, it's a plain C/C++ implementation based on ggml with no external dependencies — you just need model weight files in checkpoint, safetensors, or GGUF format.
No, it's a plain C/C++ implementation based on ggml with no external dependencies — you just need model weight files in checkpoint, safetensors, or GGUF format.Does it support GPU acceleration, or is it CPU-only?Both: it runs on CPU (with AVX/AVX2/AVX512 on x86) and also supports CUDA, Vulkan, Metal, OpenCL, and SYCL backends, so performance varies a lot depending on which backend you pick.
Both: it runs on CPU (with AVX/AVX2/AVX512 on x86) and also supports CUDA, Vulkan, Metal, OpenCL, and SYCL backends, so performance varies a lot depending on which backend you pick.Which platforms and model types are supported?It runs on Linux, macOS, Windows, and Android (via Termux/Local Diffusion), and supports a huge range of image and video models including SD1.x/SDXL/SD3.5, Flux, Qwen Image, Wan, LTX video, and more.
It runs on Linux, macOS, Windows, and Android (via Termux/Local Diffusion), and supports a huge range of image and video models including SD1.x/SDXL/SD3.5, Flux, Qwen Image, Wan, LTX video, and more.What extra features does it support beyond plain text-to-image?It supports LoRA, ControlNet, IP-Adapter, PhotoMaker, LCM/LCM-LoRA, VAE tiling for lower memory use, ESRGAN upscaling, ADetailer, and embeds generation parameters into PNG output compatible with webui.
It supports LoRA, ControlNet, IP-Adapter, PhotoMaker, LCM/LCM-LoRA, VAE tiling for lower memory use, ESRGAN upscaling, ADetailer, and embeds generation parameters into PNG output compatible with webui.What weight formats can I load?It can load PyTorch checkpoints (.ckpt/.pth/.pt), Safetensors, and GGUF files, and the convert mode can turn weights into .gguf or .safetensors.
It can load PyTorch checkpoints (.ckpt/.pth/.pt), Safetensors, and GGUF files, and the convert mode can turn weights into .gguf or .safetensors.Can I get reproducible results across different machines?Yes, it offers cross-platform reproducibility with --rng cuda (matching stable-diffusion-webui's GPU RNG) or --rng cpu (matching ComfyUI's RNG).
Yes, it offers cross-platform reproducibility with --rng cuda (matching stable-diffusion-webui's GPU RNG) or --rng cpu (matching ComfyUI's RNG).Are there bindings for other languages?Yes — community bindings exist for Go, C#, Python, Rust, and Flutter/Dart, letting you use stable-diffusion.cpp as a backend from other ecosystems.
Yes — community bindings exist for Go, C#, Python, Rust, and Flutter/Dart, letting you use stable-diffusion.cpp as a backend from other ecosystems.What are similar projects?WanGP, diffusers, stable-diffusion-webui, stable-diffusion-webui-colab, InvokeAI, stable-diffusion-webui-forge, LTX-2, generative-models
WanGP, diffusers, stable-diffusion-webui, stable-diffusion-webui-colab, InvokeAI, stable-diffusion-webui-forge, LTX-2, generative-modelsWhich topics is it in?esrgan, text2image, qwen-image, wan, flux, img2img, cplusplus, opencl, controlnet, llama-cpp, upscaling, diffusion
esrgan, text2image, qwen-image, wan, flux, img2img, cplusplus, opencl, controlnet, llama-cpp, upscaling, diffusion