Skip to content
Get packs

rembg

Removes the background from images, usable as a command line tool, Python library, HTTP server, or Docker cont

GitHub Download v2.0.86
Stars
25.3k
Forks
2.5k
Licence
MIT
Language
Python
Last update
4 days ago
Type
App

What it is

Rembg takes an image and removes its background, leaving the main subject with a transparent background. You can use it from the terminal, from Python code, as a local web server, or in a Docker container. It is handy for preparing product photos, cutouts, or frames from video.

  • Remove the background from a single image from the command line with rembg i
  • Batch-process a whole folder of images with rembg p, optionally in watch mode
  • Run a local HTTP server with rembg s and call its /api/remove endpoint
  • Call remove() from a Python script on bytes, PIL images, or NumPy arrays
  • Process video frames by piping FFmpeg RGB24 output into rembg b

Great for

  • Developers who want to add background removal to a Python program
  • Designers or creators batch-processing a folder of photos
  • People who want a local tool rather than uploading images to a website
  • Video users piping FFmpeg frames through the batch mode

Not ideal for

  • Users on Python versions outside 3.11 to 3.13, since the README lists that requirement
  • People who need a point-and-click graphical app; the README describes CLI, library, server and Docker usage, with a Gradio UI only as part of the server
  • Users with NVIDIA or AMD GPUs who cannot install the matching ONNX runtime and CUDA or ROCm pieces, since the README says GPU setup may be difficult and points to a CPU fallback
Step-by-step guide: how to use rembgInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Install with a package manager

    You'll need: Python >=3.11, <3.14 · pip

    Install CPU version with CLI

    pip install "rembg[cpu,cli]"
  2. 2

    Run it

    • Install with pip install "rembg[cpu,cli]"
    • Run rembg --help to see the subcommands
    • Remove a background with rembg i path/to/input.png path/to/output.png
    • Optionally start the web server with rembg s and open http://localhost:7000/api
  3. 3

    Get the most out of it

    • Use -m to pick a different model, such as u2netp for a lighter option
    • Use -a for alpha matting and -dc for color decontamination on soft edges
    • Use -om to output only the mask
    • Use rembg p -w to watch a folder and process new files automatically
    • Use rembg s --no-ui to reduce idle CPU use when you do not need the Gradio UI

Downloads

Source code (.zip)main branch · straight from GitHubRelease v2.0.862 files incl. rembg-cli-cpu-installer.exe

Strengths

  • Offers several ways to use it: command line, Python library, HTTP server, and Docker
  • Supports multiple models and optional alpha matting and color decontamination, per the README
  • Released under the MIT license and has a large following on GitHub

Trade-offs

  • GPU setup depends on your hardware and can require extra packages such as onnxruntime-gpu, CUDA, and cudnn-devel
  • Some options call external cloud APIs (withoutBG, Atlas Cloud) that need API keys and may cost money
  • The README is truncated, so the full list of models and Docker steps is not confirmed here

How it compares

ProjectStarsTypeLevelUpdatedBest for
rembg (this)25.3kAppintermediate4 days agoDevelopers who want to add background removal to a Python program
supervision51.5kLibraryintermediate4 days agoPython developers building apps on top of object detection or segmentation models
filepond16.4kLibraryintermediate9 days agoWeb developers who need a polished upload widget in a site or app
blind_watermark14.8kLibraryintermediate8 months agoPython developers who want to add a hidden ownership or tracing mark to images
segmentation_models.pytorch11.8kLibraryintermediate4 days agoPython developers building image segmentation models in PyTorch
thumbor10.5kAppintermediate4 days agoWeb teams that need many image sizes generated on demand

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.