EasyDiffusion

EasyDiffusion is a beginner-oriented Stable Diffusion desktop client open-sourced by cmdr2 in August 2022. Its pitch is "no technical background required, just download and run." With 10k+ stars on GitHub, it sits alongside A1111 WebUI, Fooocus, and InvokeAI as one of the four mainstream local SD WebUIs. Its differentiation is one-tap installation and an all-in-one desktop bundle.

Core capabilities revolve around zero friction and just-enough functionality. First, one-click install: Windows ships an Easy-Diffusion-Windows.exe installer; Linux and macOS ship zip archives that start with ./start.sh. The bundle embeds Python and dependencies, so no environment setup is required. Second, a Task Queue lets users line up multiple prompts and walk away — the UI processes them in order. Third, automatic model recognition: a built-in YAML configuration database means dropping in a new model does not require manually editing metadata. Fourth, optional advanced features include ControlNet, Textual Inversion, LoRA training, model merging, and custom VAE/GFPGAN/UI plugins. Fifth, the v4 engine adds support for Flux 1/2, Z-Image, Lumina, Anima, and quantized models — as the README headline puts it: "Support for Ernie, Z-Image, Flux 1 and 2, Lumina, Anima and quantized models have been added."

Use cases include newcomers who want to skip command-line setup, creators in home or office environments who need a one-click install, users who want a consistent experience across Windows, Mac, and Linux, and beginners exploring Stable Diffusion XL. The onboarding bar is extremely low: 2GB of VRAM on NVIDIA or AMD cards, or an M1/M2/M3/M4 Mac, is enough; Intel Macs run as well. Uninstallation is just deleting the EasyDiffusion folder — no registry leftovers.

License note: the project uses a custom MIT-based license. The LICENSE file is titled "License" rather than a standard SPDX identifier; Section I is MIT terms and Section II introduces the CreativeML Open RAIL-M use-based restrictions, which is why SPDX returns NOASSERTION. The "MIT" label in queue.md reflects an earlier version; users redistributing models should review the use-based restrictions.