InvokeAI

InvokeAI is an open-source creative studio for running diffusion models entirely on your own hardware. Its centerpiece is the Unified Canvas — a layer-based, non-destructive editing workspace with inpainting, outpainting, brush tools, regional prompting and ControlNet, so you iterate on regions of an image instead of regenerating from scratch. Around it sit a visual model manager, boards/gallery for organizing outputs, and a node-based workflow builder for repeatable pipelines. It serves a local web server with a React UI plus a FastAPI backend (Swagger at /docs) and supports multi-user deployments with JWT/Bearer auth. SD1.5, SDXL and FLUX families are supported. Distributed under Apache-2.0, it is the most commercially-friendly license of the three big local image UIs — no copyleft obligations on derivative use.

Automatic installer (recommended)

1. Download InvokeAI-installer-vX.Y.Z.zip from the GitHub releases page (Windows, macOS, Linux)
2. Unzip; on Windows first double-click WinLongPathsEnabled.reg to avoid long-path failures
3. Run install.bat (Windows) or install.sh (Linux/macOS); choose the default (latest) version, pick an install location with several GB free, and select your GPU device
4. Run invoke.bat / invoke.sh → choose "run the UI" → open http://127.0.0.1:9090
5. Install at least one generation model via the built-in Model Manager before generating

pip install (advanced)

1. pip install InvokeAI==6.14.2 (check PyPI for the current version)
2. Run the invokeai launcher and follow the first-time prompts

Updates: re-run the same installer and select the new version — existing models and configs are preserved.