InvokeAI
A self-hosted creative studio for AI image generation, with a layered canvas, node-based workflows and a model manager.
Invoke, also called InvokeAI, is a free and open-source creative engine for image generation under the Apache 2.0 license. It runs a local web server and interface on your own machine. Its Unified Canvas offers layer-based editing, with drawing, painting, inpainting and outpainting, and its node-based workflows let you build reusable generation pipelines and expose the settings you want as a simple form. A model manager handles checkpoints, LoRAs, textual inversions and ControlNets, and the gallery keeps prompts and settings with each image.
The README lists support for Stable Diffusion 1.5, 2.0, SDXL and 3.5, the FLUX.1 and FLUX.2 families including FLUX.2 Klein, Z-Image, Krea 2, Anima, Qwen Image and Qwen Image Edit, Ideogram 4 and ERNIE-Image. It also lists a few models that can only be used through an API, such as Nano Banana, GPT Image and Wan, and the documentation covers external providers and video generation. Low-VRAM mode, FP8 storage and quantization help larger models fit on smaller cards.
You install it with the Invoke Launcher, which sets up Python for you. The documentation says it runs on Windows 10 or later, macOS 14 or later and Linux, with Nvidia GPUs on Windows and Linux, AMD GPUs on Linux only, Intel Arc GPUs on Windows and Linux, and Apple silicon Macs with 16 GB or more of memory recommended. Its table of minimum requirements runs from 4 GB of VRAM for Stable Diffusion 1.5 to 24 GB for FLUX.2 Klein 9B and Krea 2, with smaller quantized builds possible.
Good for
- Artists and designers who want canvas-based editing and inpainting around a local image model
- Building repeatable generation pipelines with node-based workflows
- Trying new open image models such as FLUX.2 Klein, Z-Image and Krea 2 in one interface
- Keeping prompts, images and settings private on your own hardware
Watch out for
- Hardware needs depend on the model, and the best-supported setup is an Nvidia GPU
- Models marked API-only in the README run on the provider's servers, so those images are not generated locally
- The Apache 2.0 license covers the app, so check each model's own license before commercial use