intel/openvino-ai-plugins-gimp

OpenVINO™ AI Plugins for GIMP

What it is – A collection of GIMP 3 plugins that bring on‑device AI inference (via Intel OpenVINO) into the image‑editing workflow. The plugins let you run popular generative and vision models—Super‑Resolution, Semantic Segmentation, and a wide range of Stable‑Diffusion variants—directly inside GIMP, using CPUs, GPUs or Intel NPUs.

Key capabilities

Plugin	Main function
Super‑Resolution	Upscales images with AI‑enhanced detail.
Semantic Segmentation	Generates pixel‑wise class masks for easy object selection.
Stable‑Diffusion family	Text‑to‑image, image‑to‑image, inpainting, and ControlNet (OpenPose, Canny, Scribble) for 1.5, SDXL, SD 3.0/3.5, and Turbo/LCM variants.
FastSD	Faster diffusion generation using Latent Consistency Models, with CPU/GPU/NPU support.

All models are compiled and run through OpenVINO, so inference runs locally on Intel hardware without needing a cloud service.

How to get started

1. Install – Follow the OS‑specific guides in Docs/ (Windows user guide or Linux installation guide). On Linux the repo recommends building GIMP 3 from source; Flatpak is not supported.
2. Load a model – In GIMP open the Layers → OpenVINO‑AI‑Plugins menu, pick a model (e.g., “Stable Diffusion 1.5”) and click Load Models. The first load compiles the model and may take a few minutes; subsequent runs are fast because the compiled model is cached.
3. Run inference –
   • Prompt‑to‑image – Choose a layer or blank canvas, enter a text prompt and other parameters, then click Generate.
   • Image‑to‑image – Provide an initial image (the current canvas or a file) and generate a variation.
   • Inpainting – Add a layer mask to define the region to replace, then run the inpainting model.
   • ControlNet – Select a ControlNet variant (OpenPose, Canny, Scribble), enable Use Initial Image, and generate guided results.
4. FastSD – Manage FastSD models via the built‑in Model Manager; the first use downloads the model at runtime.

Supported hardware – Any Intel device supported by OpenVINO (CPU, integrated GPU, or dedicated NPU). Some models have an “int8 Power Mode” that accelerates inference on NPU‑only systems.

Development & contribution – The repository serves as reference code for integrating OpenVINO inference into GIMP plugins. Contributions are welcomed via pull requests.

License & legal notes

• Code – Apache 2.0.
• Models – Most Stable Diffusion models are under the Creative ML Open Rail M license; users must ensure their own usage complies with that license.

---

This summary is based solely on the repository’s README.