Find this useful and want to support my work? You can buy me a coffee (or whatever) here. :)

Interactive Background Remover is a user-friendly tool designed to remove backgrounds from images using a combination of interactive models (Segment Anything) and automatic whole-image models (such as u2net, disnet, rmbg, and BiRefNet). This allows you to refine and fine tune your background removal, similarly to apps like Photoroom, instead of only running models on the entire image.

• Interactive Model Support: Utilise Segment Anything for detailed object selection by clicking or drawing boxes on the image.
• Automatic Whole-Image Model Support: Apply models like u2net, disnet, rmbg, and BiRefNet for quick background removal across the entire image, which can be further refined.
• Zoom and Pan: Zoom in on specific parts of the image and pan around for detailed editing. The models are only run on the viewed area so you can incrementally build up an image from higher resolution patches.
• Manual Paintbrush Mode: Manually refine the background removal with a paintbrush tool for areas not covered by the models.
• Mask Refinement: Smooth edges, filter anomalous areas and soften the mask for a natural look. You can directly edit the mask for high fidelity background removal.
• Alpha Matting and Trimap Editor: Refine mask edges for improved accuracy with hair and similar features. You can use different matting models and also edit the trimap for higher quality results.
• Smart Refine Paintbrush: Clean up the tricky parts of the output image such as hair
• Batch Editing: Utilising all the features of the app
• Drop Shadow, Glow, and more effects: Basic drop shadow effect for the cutout object.
• Background Options: Choose from various background colours or apply a blurred background effect.
• Image Editing: Includes a built-in image editor and cropper to preprocess images before background removal.
• Object Removal/Inpainting: Using open weight models LaMa and DeepFillV2
• Undo/Redo
• Save Options: Save the processed image in various formats (PNG, JPEG, WebP) with customisable quality settings and auto-trim.
• Clipboard Support: Load images directly from the clipboard for quick editing.
• Windows, Linux and Mac Builds: In the Github releases. Mac is currently untested, please let me know
• Model Downloader: Easily download compatible models.

• Python 3.x
• Required Python packages: Pillow, numpy, onnxruntime, opencv-python, PyQt6, requests, pymatting

pip install pyqt6 Pillow numpy onnxruntime opencv-python requests pymatting

Or download prebuilt executables for Windows, Linux and Mac from the Github releases

When running from the script, you will need to install the appropriate onnxruntime package for your hardware. Only install one of these packages, and remove any onnxruntime packages before installing a new one. Executable releases are built with the appropriate runtime.

pip install onnxruntime # CPU only
pip install onnxruntime-gpu # Nvidia CUDA, TensorRT
pip install onnxruntime-openvino # Intel iGPU and GPU
pip install onnxruntime-directml # Microsoft Windows general machine learning (Nvidia, AMD, Intel)

This application requires pre-trained background removal/segmentation models in onnx format to function correctly.

The application includes a model downloader to make getting started very easy.

Interactive Models

• Segment Anything 1 and 2 + mobile-sam:

I recommend using mobile-sam as it has near instant inference results, and you can zoom into the image for higher resolution masks. I haven't found much benefit to using the larger models.

Whole Image Models

• rembg 1.4 and 2: https://huggingface.co/briaai/RMBG-1.4/tree/main/onnx
• u2net, u2netp
• disnet
• BiRefNet
• MVANet
• MODnet
• BEN2

Alpha Matting Models

• ViTMatte
• AlphaMatte
• PyMatting

Inpainting Object Removal

• LaMa
• DeepFillV2

Run the prebuilt executables: Github releases

Or run the script from the command line:

python interactive_background_remover.py

You can also provide image file paths as command-line arguments to load them directly:

python interactive_background_remover.py image1.jpg image2.png

• Input Canvas: Displays the original image.
• Output Canvas: Displays the processed image with the background removed.
• Controls Panel: Contains various tools and options for interacting with the image and models.

1. Load Image: Click "Open Image" to load an image file or "Open Clipboard" to use an image from the clipboard.
2. Select Model: Choose a model from the "Mask" section.
3. Zoom and Pan: Use the full image or zoom into region of interest (arrow keys or middle mouse to pan, mouse scroll to zoom)
4. Generate Background Removal Mask:
   • Interactively: Segment Anything: Left-click to add a positive point, right-click to add a negative (exclusion) point, or drag to draw a box around the object.
   • Automatic: Whole-Image Models: Click "Run whole-image model" to apply the selected model to the entire image (or zoomed area).
5. Refine Background Removal Mask:
   • Use the "Add mask" and "Subtract mask" buttons to add the model output to the background removal mask.
   • Toggle "Manual Paintbrush" mode to manually edit the mask. Right click to draw a mask, left to erase.
   • Check "Post Process Model Mask" (binarise the mask and smooth edges) and "Soften Model Mask/Paintbrush" to adjust the model output
   • Use Show Full Mask to edit the mask directly instead of the output image
6. Adjust Background: Select a background color or enable the "Blurred" option for a blurred background.
7. Save Image: Click "Save Image As...." to save the processed image.

• a: Add current mask to the output image.
• s: Subtract current mask from the output image.
• Ctrl+z: Undo the last action.
• p: Toggle manual paintbrush mode.
• c: Clear current mask and coordinate points.
• w: Reset the current working image.
• r: Reset everything (image, masks, coordinates).
• v: Clear the visible area on the working image.
• Ctrl+s: Save as....
• Ctrl+Shift+S: Quick save JPEG with a white background.
• u: Run u2net model.
• i: Run disnet model.
• o: Run rmbg1.4 model.
• b: Run BiRefNet model.
• Middle Mouse, Left, Right, Up, Down: Pan the image.
• Ctrl + trackpad: Zoom

Click "Edit Image" to open the built-in image editor, where you can crop, rotate, and adjust various image parameters like brightness, contrast, saturation, and white balance. You can also inpaint and remove objects using downloadable models (LaMa and DeepFillV2).

• No models found: Ensure that the required models are downloaded and placed in the Models/ directory. Use the model downloader to ensure the filenames and locations are correct.
• Performance issues: Adding multiple effects can cause slowdown, such as blurred background and drop shadow. The app is currently single-threaded, so running large models can freeze the app temporarily during processing.

Find this useful and want to support my work? You can buy me a coffee (or whatever) here. :)

I'm fairly new to python so any improvements to the code, features, and suggestions are welcome. There are likely bugs.

This was originally inspired by the command line program RemBG by Daniel Gatis, and some of the original inference code was adapted from this.

Huge thanks to Meta for Segment Anything and all the other model authors for releasing their models.