Stable Diffusion on Your Own Machine

Generate images on your own computer. No metering. No monthly bill. No prompt leaving your machine. Stable Diffusion is a family of open-weights models you download and run yourself. That one property changes everything: images cost nothing per generation, the same seed reproduces the same picture exactly, and a large ecosystem of community models puts almost any aesthetic within reach. It also asks something in return. Setup takes an afternoon. The settings panel is intimidating. And nobody hands you a manual. This is the manual, for someone who is comfortable installing software and is not a programmer. Inside: - The front-end landscape, by category, and which to start with - Hardware, explained without pretending your machine is the reference machine - The controls that actually matter: checkpoint, guidance, sampler, steps, seed - Why you fix the seed before you change anything else - Prompts and negative prompts, and why pasting a stranger's negative prompt is cargo cult - Checkpoints and LoRAs: where the look really comes from - img2img and inpainting for the good image with the broken hand - ControlNet, for telling it where things go - Reproducibility, and the settings notebook that outlives every version bump - Licenses, rights, and the responsibility of running a model with no filter in front of it The licensing chapter promises nothing. Open weights do not mean free to use commercially. Different checkpoints carry different licenses, the law on owning machine-generated work is unsettled, and this book tells you to read the license on the file you downloaded, every time. You will also learn what it is bad at: hands, legible text inside the image, exact object counts, and consistent characters across a set. Plain language. No jargon, no hype, no motivational filler. Control, reproducibility, and no per-image cost. It still will not give you taste.