About

The story behind picoDiffusion.

What is picoDiffusion?

picoDiffusion is a minimal, self-hosted AI image generator for Stable Diffusion 1.5. It runs in a Docker container on your own hardware and provides a clean web interface for generating images from text prompts.

It was built to be the tool we wished existed when we first started exploring AI image generation — something that just works, tells you what every control does, and does not overwhelm you on day one.

Why does it exist?

Most Stable Diffusion interfaces are powerful but overwhelming. They have dozens of tabs, hundreds of settings, plugin systems, and extension managers. That is great for power users, but it is a terrible first experience.

We wanted something different:

One page with all the controls you need and nothing you do not
A built-in help system that explains everything in plain language
Fast performance on modest hardware, not just on expensive GPUs
Docker-based so there is nothing to install except Docker itself

picoDiffusion is not trying to replace more powerful and more capable Stable Diffusion tools. It is trying to be the first tool you use — the one that teaches you what a checkpoint is, what CFG does, and why your hands look weird. Once you understand the basics, you can move to more complex tools if you need to. Or you might find that picoDiffusion does everything you need. Either way, the hands will still look weird. That is not us, it is sd1.5. Swear.

Platform

picoDiffusion is a Docker-based tool built for Linux with NVIDIA GPUs. That is the primary and supported platform.

PyTorch does work well on Apple Silicon via MPS, and the code already supports it. However, Docker on Mac cannot pass through the Metal GPU, so the Docker container falls back to CPU. There is a native install guide that covers running without Docker, but we have not tested it on a Mac. We do not have one. We do not know anyone who will lend us one. See the CUDA / NVIDIA Setup page for more details.

Security

Local use only. picoDiffusion is designed for personal, local use on your own machine. It has no user authentication, no access controls, and a completely open API. Under no circumstances should it be exposed to the internet, used on a public network, or used to host image generation for other people. If you need a multi-user or publicly accessible image generation service, this is not the right tool.

Who built it?

picoDiffusion is a lacquerlabs project.

lacquerlabs. Where we make stuff. And things. Go away.

Technology

Component	Technology
Backend	Python 3.12, FastAPI, Uvicorn
Inference	HuggingFace Diffusers, PyTorch (CUDA)
Frontend	Vanilla HTML, CSS, JavaScript — no frameworks
Container	Docker with NVIDIA GPU passthrough
Model format	Stable Diffusion 1.5 / 1.4 checkpoints (.safetensors)

Open source

picoDiffusion is open source. The code is available on GitHub.

Contact

Questions, feedback, or bug reports? Open an issue on GitHub or reach out to lacquerlabs directly.