Generate images and videos locally with a pure C/C++ engine and no external dependencies
You can process text-to-image or image-to-image tasks on your local machine using a single executable file. This tool is a pure C/C++ diffusion model inference engine that follows the same structure as llama.cpp. It simplifies the process of running a model and obtaining results with a single command, without complex environment setup.
New model support and web UI introduction
Recent updates added Day-0 support for Qwen-Image-2.1 and quickly incorporated the latest models such as LTX-2.5 and MiniMax-H3. In April 2026, a built-in web UI was introduced to improve usability. These changes indicate that the project is under active development, so note that the API and command-line options may change frequently.
Supported models and feature scope
It supports various image models, including Stable Diffusion 1.x through 3.5, FLUX, Qwen Image, and Z-Image.
- Model formats It supports Pytorch checkpoints (.ckpt, .pth, .pt), Safetensors (.safetensors), and GGUF (.gguf) formats. It also provides a mode to convert model weights to .gguf or .safetensors.
- Inference backends It supports CPU (with AVX, AVX2, and AVX512 support), CUDA, Vulkan, Metal, OpenCL, and SYCL. It uses flash attention to optimize memory usage.
- Platforms It runs on Linux, Mac OS, Windows, and Android (Termux, Local Diffusion).
Execution methods and basic usage
You can download pre-built binaries from the release page or build directly from source. After preparing the model weight file, generate an image using the command ./bin/sd-cli -m [model_path] -p "prompt". For example, you can generate an image of "a lovely cat" using the Stable Diffusion v1.5 model. Refer to the CLI documentation for detailed command-line arguments.
Pre-use considerations
Since this project is in an active development stage and APIs and options change frequently, it is important to check the latest release. If you need performance optimization or want to reduce VRAM/RAM usage, you must review the performance guide and backend selection guide. If you intend to use it in other languages such as Golang, Python, or Rust, it is recommended to check the corresponding language binding projects. It is suitable for developers who want to generate images independently in a local environment or use it as a backend for existing tools like GIMP plugins or LocalAI.
