Remove Background for GIMP 3.2

AI-powered background removal using rembg + ONNX Runtime with GPU acceleration. Cross-platform: Linux, Windows, macOS.

Requirements

• GIMP 3.2+
• Python 3.10+ with python3-venv
• GPU (optional): see Hardware / Provider Mapping below

Dependencies

Direct (pip-installed by install.sh)

Package	Required	Size	Notes
onnxruntime-gpu	NVIDIA only	~300 MB	CUDA acceleration (Linux/Windows)
onnxruntime-directml	Windows GPU	~150 MB	AMD, Intel, NVIDIA via DirectML
onnxruntime-silicon	Apple only	~50 MB	CoreML acceleration (M-series)
onnxruntime (CPU)	Fallback	~30 MB	Universal CPU — much smaller
rembg	Required	~10 MB	AI background removal engine
rembg[gpu]	With CUDA	—	Meta-extra for GPU-enabled rembg
pillow	Required	~5 MB	Image I/O
numpy>=2.0,<2.5	Required	~20 MB	Array computation

Only one onnxruntime-* package is installed — install.sh auto-detects your hardware.

Sub-dependencies (pulled automatically by rembg)

Package	Size	Purpose
opencv-python-headless	~25 MB	Image processing, alpha matting
scikit-image	~20 MB	Image filtering
scipy	~25 MB	Scientific computation
pymatting	~5 MB	Alpha matting algorithm
pooch	~1 MB	Model download manager
onnx / protobuf	~5 MB	ONNX model parsing
huggingface_hub	~10 MB	Model hub access

Total install size: ~200–500 MB (package-dependent). First run downloads the AI model (~176 MB) and caches it.

Hardware / Provider Mapping

Hardware	Package Installed	Execution Provider
NVIDIA Linux	onnxruntime-gpu	CUDAExecutionProvider
NVIDIA Windows	onnxruntime-directml	DmlExecutionProvider
AMD Windows	onnxruntime-directml	DmlExecutionProvider
AMD Linux	onnxruntime (CPU)	CPUExecutionProvider
Intel Arc/iGPU Windows	onnxruntime-directml	DmlExecutionProvider
Intel iGPU Linux	onnxruntime (CPU)	CPUExecutionProvider
Apple Silicon	onnxruntime-silicon	CoreMLExecutionProvider
CPU-only	onnxruntime (CPU)	CPUExecutionProvider

Windows NVIDIA note: install.sh defaults to DirectML (zero-config) instead of CUDA to avoid requiring a separate CUDA Toolkit installation. Falling back to CPU is always automatic.

Install

Automatic (recommended)

install.sh detects your GPU and installs the correct ONNX Runtime package automatically.

Linux / macOS:

chmod +x install.sh && ./install.sh

Windows: Install Git Bash, right-click the plugin folder → Git Bash Here, then:

chmod +x install.sh && ./install.sh

---

Manual Install

Pick the onnxruntime-* package for your hardware (see Hardware / Provider Mapping).

Linux / macOS

# 1. Install system dependencies (if missing)
#    Debian/Ubuntu:
sudo apt install python3 python3-venv
#    macOS:
brew install python3

# 2. Create shared venv
python3 -m venv ~/.gimp-plugin-shared-venv/venv

# 3. Install packages (pick ONE onnxruntime package for your GPU)
source ~/.gimp-plugin-shared-venv/venv/bin/activate
pip install --upgrade pip

#    NVIDIA GPU:
pip install onnxruntime-gpu rembg[gpu] pillow "numpy>=2.0,<2.5"

#    Apple Silicon:
pip install onnxruntime-silicon rembg pillow "numpy>=2.0,<2.5"

#    CPU / AMD Linux / Intel Linux:
pip install onnxruntime rembg pillow "numpy>=2.0,<2.5"

deactivate

# 4. Copy plugin files
PLUGINS=~/.config/GIMP/3.2/plug-ins/remove-background
mkdir -p "$PLUGINS"
cp remove-background.py run_worker.sh bg_remove_worker.py "$PLUGINS/"
chmod +x "$PLUGINS/remove-background.py"
chmod +x "$PLUGINS/run_worker.sh"

# 5. Flatpak only — grant host access
flatpak override --user --talk-name=org.freedesktop.Flatpak org.gimp.GIMP

# 6. Restart GIMP → Filters → Enhance → Remove Background

Windows

Run all commands in Git Bash:

# 1. Install Python 3.10+ from https://python.org/downloads/
#    Check "Add Python to PATH" during install.
#    Verify:
py --version

# 2. Create shared venv
mkdir -p ~/.gimp-plugin-shared-venv
py -m venv ~/.gimp-plugin-shared-venv/venv

# 3. Install packages (pick ONE onnxruntime package for your GPU)
source ~/.gimp-plugin-shared-venv/venv/Scripts/activate
pip install --upgrade pip

#    NVIDIA / AMD / Intel GPU (DirectML):
pip install onnxruntime-directml rembg pillow "numpy>=2.0,<2.5"

#    CPU only:
pip install onnxruntime rembg pillow "numpy>=2.0,<2.5"

deactivate

# 4. Copy plugin files
PLUGINS="$APPDATA/GIMP/3.2/plug-ins/remove-background"
PLUGINS="$(echo "$PLUGINS" | sed 's|\\\\|/|g' | sed 's|C:|/c|')"
mkdir -p "$PLUGINS"
cp remove-background.py run_worker.sh bg_remove_worker.py "$PLUGINS/"

# 5. Restart GIMP → Filters → Enhance → Remove Background

Usage

1. Open an image in GIMP
2. Select a layer
3. Filters > Enhance > Remove Background
4. New layer [name] (bg removed) appears with transparent background

First run downloads the AI model (~176 MB). Subsequent runs use the cached model.

How it works

Layer → Export PNG → rembg (onnxruntime-gpu + CUDA) → Cutout PNG → New layer

On Flatpak GIMP, flatpak-spawn --host escapes the sandbox to reach the host GPU.

Troubleshooting

Plugin not in menu: chmod +x ~/.config/GIMP/3.2/plug-ins/remove-background/*.py

ModuleNotFoundError: onnxruntime: re-run install.sh

CUDA not detected: verify nvidia-smi, re-run install.sh

Related Plugins

• Smart Object Selection — AI-powered object selection tool
• AI Upscaler — Real-ESRGAN upscaling

License

GNU General Public License v3.0. See LICENSE.

Support

If this plugin helps your workflow, consider donating to support further development.

Support on Ko-fi