unbagrnd

unbagrnd removes photo backgrounds with on-device AI. Its desktop app processes images locally through ONNX Runtime; the maker says images, filenames, and metadata are not sent anywhere. The app makes a one-time model download and has no analytics, update checks, or telemetry. Users can choose among several models with different speed and accuracy tradeoffs; the default is described as 43 MB. Editing tools include brush erase and restore with undo and redo, edge shift and smoothing, background color and shadow styling, and alpha matte export. The app is intended for single photos or folders of images, including cutout refinement and background styling. A Docker-based, self-hosted REST API processes images inside the user's container. Its documented defaults allow images up to 20 MB and 8192 pixels wide or high, batches of up to 10 images, and 60 requests per API key per minute. The software is free and MIT-licensed. The API uses plain HTTP, so the maker recommends a reverse proxy for TLS when exposing it publicly.

unbagrnd combines local desktop background removal with editing controls and a self-hosted API. It is free and MIT-licensed, with installer caveats on macOS, Windows, and Android; public API deployments need a reverse proxy for TLS.

unbagrnd may suit people who want to remove backgrounds from individual photos or image folders, then refine cutouts or style backgrounds. Its self-hosted API is an option for users who want image processing inside their own container.

Overview

unbagrnd is a free background-removal tool for desktop photo work and self-hosted image workflows. It is best for people who want local processing, batch jobs, or an API they control. Its privacy-minded design is compelling, but unsigned or unnotarized installers and extra API deployment work narrow its appeal.

Key features

The desktop app uses ONNX Runtime to remove backgrounds on the device. The maker says images, filenames, and metadata are not sent anywhere; after a one-time model download, the app makes no further network requests, analytics calls, or update checks. That suits private image work, though users should not mistake the initial download for a fully offline installation.

Several models let users trade speed for accuracy, with a 43 MB default. Folder processing is useful for repeated jobs, and brush erase and restore, undo and redo, edge shift, and smoothing provide meaningful control when an automatic cutout needs correction. Background color and shadow styling and alpha matte export help finish the result without switching tools.

The Docker-based REST API processes images in the user’s container. Its documented defaults allow up to 20 MB per image, dimensions up to 8192 pixels in either direction, batches of 10, and 60 requests per API key per minute. Separate API and admin key headers, SQLite storage limited to API-key metadata, and in-memory image processing that discards uploads and does not log them make the API suitable for controlled deployments. The API speaks plain HTTP, however, so a public deployment needs a reverse proxy for TLS.

Pricing

The Free plan costs nothing and is MIT-licensed. There are no paid tiers or usage limits stated, and there is no free trial because the project is free. That makes it a strong option for individuals and developers who can install the app or operate a container; the trade-off is that support is through GitHub issues and pull requests rather than a stated commercial support plan.

Platforms

unbagrnd supports Android, Linux, macOS, Windows, and a self-hosted API. Installation friction varies: the macOS app is not notarized, the Windows installer is not code-signed, and the Android APK is not on the Play Store. Those caveats may deter users who expect conventional store distribution or signed installers.

Who it's for

It is less suitable for people who want a browser-first service, store-installed Android app, or a publicly exposed API without managing TLS and deployment security.

Pros and cons

• Pros: Local desktop processing and a one-time model download keep routine image work off a remote service.
• Pros: Batch processing, model choice, and edge and styling controls support both repeated jobs and detailed cutout cleanup.
• Pros: The free MIT-licensed project includes a self-hosted API with defined upload, batch, dimension, and request limits.
• Cons: macOS and Windows installation caveats, plus Android APK distribution outside the Play Store, add setup friction.
• Cons: Public API use needs a reverse proxy for TLS; container deployment is not as turnkey as a hosted service.

Alternatives

AI Background Removers is the broader category to browse if you want to compare more tools. For a hosted multi-platform option with a free tier and paid monthly credit plans, consider Cutout.Pro. Pixelcut may suit users who want a mobile or web workflow with limited free background removal and upscale. EraseBG is another free choice with full-resolution downloads and no account or watermark. RemoveBackgrounds.ai offers a free browser tool with batches of up to 30 images. Zawa AI Background Remover (formerly X-Design) may fit users who prefer a daily-credit model across mobile, API, and web. Erase.bg has a free starter plan with three credits and downloads for personal use. For browser-based batch work, Don’s Tools Background Remover supports batches of up to 150 images and JPG, PNG, and WebP. Deback is another free browser option, with batches up to 50 images and a 12 MB per-image limit.

Verdict

unbagrnd is a strong choice for privacy-conscious users and developers who want free local background removal, practical edit controls, or an API they can host themselves. Its main advantage is control over where images are processed; look elsewhere if signed, store-distributed installs or a managed public API matter more than self-hosting.