KDE Dolphin integration for quick background removal using rembg.
Right-click any image in Dolphin and select "Quick RMBG" to remove its background instantly.
- Dolphin Context Menu: Right-click integration for seamless workflow
- Two-Pass Mode: Run background removal twice for cleaner results on difficult images
- Infinite Hop Mode: Interactive mode that keeps running passes until you're satisfied
- GPU Acceleration: Supports AMD (ROCm), NVIDIA (CUDA), and CPU fallback
- Preserves Originals: Saves output as
filename_noBG.pngalongside the original - Desktop Notifications: Shows success/error status via KDE notifications
- OS: Ubuntu Linux (tested on 24.04/25.04)
- Desktop: KDE Plasma with Dolphin file manager
- Python: 3.10+
- Package Manager:
uvorpipx - rembg: Must be installed separately (see below)
Choose the installation method based on your hardware:
# Install PyTorch with ROCm support first
pip install torch torchvision --index-url https://download.pytorch.org/whl/rocm6.0
# Then install rembg
pip install "rembg[cli]"# Install PyTorch with CUDA support
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121
# Then install rembg
pip install "rembg[cli]"# Install rembg with CPU-only PyTorch
pip install "rembg[cli]"Verify installation:
rembg --helpgit clone https://github.com/danielrosehill/Quick-RMBG.git
cd Quick-RMBG
./scripts/install.shThis will:
- Install the
quick-rmbgCLI wrapper (via uv or pipx) - Add the Dolphin context menu entry
- Rebuild the KDE service cache
- Navigate to any folder with images
- Right-click on an image (PNG, JPG, JPEG, WEBP, BMP, TIFF)
- Select one of:
- "Quick RMBG" - Single-pass removal
- "Quick RMBG (Two-Pass)" - Two passes for better results
- "Quick RMBG (Infinite Hop)" - Keep running until you're satisfied
- A notification will appear when complete
- Find the result saved as
originalname_noBG.png
Some images leave residual background artifacts after a single pass. Two-pass mode runs rembg twice—first on the original, then on the result—for cleaner output.
Output files:
- First pass:
originalname_noBG-first-pass.png - Final result:
originalname_noBG-second-pass.png
For images that need an unpredictable number of passes, Infinite Hop Mode lets you decide when the result is good enough. After each pass, a dialog asks if you're happy with the result:
- Click "Yes, I'm done!" to stop and keep the current result
- Click "No, run another pass" to process again
Output files:
- Each pass:
originalname_noBG-pass-1.png,originalname_noBG-pass-2.png, etc. - All intermediate passes are preserved so you can compare
Note: This mode uses kdialog (KDE) or zenity (GNOME) for the interactive dialogs.
# Basic usage
quick-rmbg photo.jpg
# Two-pass mode for cleaner results
quick-rmbg --two-pass photo.jpg
# Infinite hop mode - keep running until satisfied
quick-rmbg --infinite-hop photo.jpg
# Specify output path
quick-rmbg photo.jpg -o transparent.png
# Quiet mode (no output)
quick-rmbg photo.jpg --quiet
# No notification (print to stdout instead)
quick-rmbg photo.jpg --no-notifyConfiguration is stored in ~/.config/quick-rmbg/config.json:
{
"output_suffix": "_noBG",
"model": "u2net",
"rocm_gfx_version": "11.0.1"
}| Option | Default | Description |
|---|---|---|
output_suffix |
_noBG |
Suffix added to output filename |
model |
u2net |
rembg model to use (u2net, u2netp, u2net_human_seg, silueta, isnet-general-use, isnet-anime) |
rocm_gfx_version |
11.0.1 |
GFX version for AMD ROCm (set to null to disable) |
AMD ROCm users: The default rocm_gfx_version of 11.0.1 is for Navi 32 GPUs (RX 7700/7800 XT). Adjust this value for your GPU or set to null if not using ROCm.
NVIDIA users: Set rocm_gfx_version to null in the config file. CUDA will be used automatically if available.
CPU users: Set rocm_gfx_version to null. Processing will be slower but works without GPU.
./scripts/uninstall.shThis removes:
- The quick-rmbg CLI tool
- The Dolphin context menu entry
- Configuration files
Note: This does not remove rembg itself, which you installed separately.
Ensure rembg is installed and in your PATH:
pip install "rembg[cli]"
which rembgRebuild the KDE service cache and restart Dolphin:
kbuildsycoca6
# Then close and reopen Dolphin- AMD GPU: Ensure ROCm is installed and
rocm_gfx_versionmatches your GPU - NVIDIA GPU: Ensure CUDA-enabled PyTorch is installed
- First run is slower due to model download
Check your ROCm installation and GFX version:
rocminfo | grep gfx
# Update config.json with the correct versionMIT


