On-device AI
Background removal runs through a real neural network via ONNX Runtime, inferred locally. No image, filename or metadata is ever sent anywhere.
unbagrnd erases the background from any photo, instantly, privately and entirely on your machine. No cloud API, no account, no upload, ever.
why.md
Every feature is designed around one rule: your images never leave your device.
Background removal runs through a real neural network via ONNX Runtime, inferred locally. No image, filename or metadata is ever sent anywhere.
MIT-licensed, built on a free, open-source stack: no API keys, no paid tiers, no usage limits.
Choose from several background-removal models in Settings, trading speed for accuracy, from a lightweight 43 MB default up.
The only network request unbagrnd ever makes is the one-time model download. No analytics, no update checks, no telemetry.
Process one image, or drop an entire folder, with live previews as each file finishes. Batches share one background and shadow, name files from a template, can be cancelled or retried, and export as a ZIP.
Native installers for macOS (Apple Silicon), Windows and Linux, plus an Android build. One small app, no Electron.
screenshots/
Single-image mode, batch mode and the model picker in Settings.
editor/
On-device editors for the spots the model gets almost right: brush, edge controls and background styling. No AI and no network, just the result you already have.
Brush over parts of the subject to erase or restore, with full undo/redo history.
Shift the outline in or out, smooth jagged edges or feather them. A live preview makes hair and fur easy to dial in.
Fill in a solid color (basic, pastel, neutral or lifted straight from the photo) and add a soft drop shadow computed on-device, with angle and opacity presets.
Save the cutout’s transparency as a black-and-white alpha matte, ready to use as a layer mask in your design tool.
batch/
Set a background, a shadow and a naming scheme once, then drop in a whole folder. It all runs on your device, one image at a time.
Apply one background color and drop shadow to every image in the batch.
Name outputs with a template such as {index}_{name}-studio; collisions are renamed instead of overwritten.
Stop a run at any time, then retry only the files that failed or were not processed.
Bundle the finished images into a single archive in one click (desktop).
videos/
A quick showcase of the app, and a full walkthrough of how background removal works end to end.
A quick tour of the interface, single and batch modes, and the on-device model picker.
A hands-on demo of background removal from start to finish, entirely offline.
how-it-works.md
unbagrnd runs IS-Net “general use” as an ONNX model through the Rust ort bindings for ONNX Runtime, the same model family used by the popular rembg tool.
Your image is loaded and resized to the model’s input size, fully in memory.
The neural network predicts a foreground mask, entirely on your CPU.
The mask becomes an alpha channel, composited back onto the original pixels.
The result is written straight to disk. The frontend never touches the network.
releases/latest
Pick your platform. Every build is free, and the source is right there on GitHub.
v0.6.0 · Apple Silicon
Not notarized yet, so macOS will say the app is “damaged”. Move it to Applications, then run this once in Terminal:
xattr -cr /Applications/unbagrnd.app v0.6.0 · Windows 10 / 11 · x64
Not code-signed yet, so SmartScreen may flag it. Click “More info”, then “Run anyway”.
v0.6.0 · x86_64
v0.6.0 · Native photo picker · on-device inference
Not on the Play Store, so Android warns about installing from an unknown source. Allow it once in the install prompt.
unbagrnd is MIT-licensed. Read the code, file an issue or send a pull request: it’s all on GitHub.
Built by Zidni Ridwan Nulmuarif. Not affiliated with Apple, Microsoft, Google or any Linux distribution.