Rembg

Warnings

• gotcha First-time usage requires downloading AI models (100-300 MB each). This causes an initial delay and requires an active internet connection. Models are cached locally in `~/.u2net` for subsequent offline use.
  Fix: Ensure an internet connection for the first run. Be aware of the initial delay for model download.
• gotcha For optimal performance when processing multiple images, explicitly create and reuse a session using `new_session()` instead of relying on the default behavior, which initializes a new session for each call.
  Fix: Initialize a session once with `session = new_session()` and pass it to subsequent `remove()` calls: `remove(image, session=session)`.
• gotcha GPU (NVIDIA/CUDA or AMD/ROCm) acceleration (`rembg[gpu]` or `rembg[rocm]`) requires specific system setups including `onnxruntime-gpu` or `onnxruntime-rocm`, potentially CUDA/cudnn-devel or ROCm libraries. Incorrect setup often leads to `ModuleNotFoundError` for `onnxruntime` or `DLL load failed` errors on Windows.
  Fix: Verify `onnxruntime` compatibility at `onnxruntime.ai`. On Windows, install the latest Visual C++ Redistributable. For NVIDIA, ensure CUDA and cudnn-devel are correctly installed. If issues persist, revert to `rembg[cpu]`.
• breaking Python version compatibility has changed. Current `rembg` versions require Python `>=3.11, <3.14`. Older versions might have supported broader ranges (e.g., `>=3.7, <3.11`). Using an unsupported Python version will lead to installation failures or `ModuleNotFoundError`.
  Fix: Use Python 3.11 or 3.12. Check the PyPI page for the exact `requires_python` range for your target `rembg` version.
• gotcha Depending on the model and image size, `rembg` can be memory-intensive, especially with GPU acceleration. Large images or batch processing can lead to Out Of Memory (OOM) errors.
  Fix: Consider processing images in smaller batches, resizing large images before processing, or using a less memory-intensive model (e.g., `u2netp`). Monitor system memory during operation.

Install

• pip install "rembg[cpu]" CPU support (recommended)
• pip install "rembg[gpu]" NVIDIA/CUDA GPU support
• pip install "rembg[rocm]" AMD/ROCm GPU support
• pip install "rembg[cpu,cli]" CPU with CLI (optional)

Imports

• remove

  from rembg import remove

  The primary function to remove backgrounds.
• new_session

  from rembg import new_session

  Creates an inference session for a specific model. Important for performance when processing multiple images.
• Image

  from PIL import Image

  Often used with `rembg` for opening and saving images.

Quickstart

This quickstart demonstrates how to use `rembg` to remove the background from an image. It highlights the use of `new_session` for efficiency and assumes `PIL` for image handling. For a real scenario, replace the dummy image creation with loading from a file (e.g., `Image.open('input.png')`).

from PIL import Image
from rembg import remove, new_session

# Create a session to improve performance for multiple images
session = new_session()

# Example: Load an image from a dummy source or local path
# For a real application, replace this with actual image loading
# For demonstration, we'll create a blank image
# In a real scenario, you'd use Image.open('path/to/your/image.png')

try:
    # Simulate loading an image (replace with actual image path)
    input_image = Image.new('RGBA', (200, 200), (255, 0, 0, 255)) # A red square
    # Or, if you have an actual image file:
    # input_image = Image.open('path/to/your/image.png')

    # Remove the background
    output_image = remove(input_image, session=session)

    # Save the result
    # output_image.save('output.png')
    print("Background removed successfully (output not saved in this example).")
    print(f"Original image mode: {input_image.mode}, size: {input_image.size}")
    print(f"Output image mode: {output_image.mode}, size: {output_image.size}")
except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure you have an actual image file or mock image data for processing.")