leejet/stable-diffusion.cpp — reverse-engineered prompt
Reverse engineered prompt
Build me a lightweight local image generation tool in plain C or C++ that feels like a native version of the usual stable diffusion apps, with almost no outside dependencies. I want it to run from the command line with a really simple prompt like giving it a model file and a text prompt, but also include a small built in web UI so it is easy to use without memorizing commands.
It should work for normal text to image generation, image editing, and if possible the supported video models too. Please make it handle common model files people already have, including ckpt, safetensors, and gguf, and support popular model families like Stable Diffusion, SDXL, FLUX, Wan, and Qwen Image where this repo already supports them. I also want useful quality of life stuff like negative prompts, seeds for repeatable results, LoRA support, ControlNet for SD 1.5, image upscaling, and saving the generation settings into the output PNG.
Please make it work on CPU first and use GPU backends when available. Look up current docs online if you need to.
Want more depth? Deep Reverse