ShareX 21.0.0 landed on July 3, 2026, and it brings a feature many users have been waiting for: one-click background removal that runs entirely on your own PC. The open-source screenshot and media sharing tool now includes an AI-powered Background Remover tool, but it isn’t plug-and-play. To use it, you must manually download a compatible ONNX model file and point ShareX to it.
It’s a deliberate choice—and one that will split users between those who value privacy and those who just want it to work.
What’s New in ShareX 21.0.0
The headline addition is the Background Remover. It lives under the Tools menu alongside the existing image editor, color picker, and OCR engine. Once you feed it a properly formatted ONNX model, it can strip backgrounds from screenshots, photos, and other images without sending a single pixel to the internet.
But there’s a catch. ShareX does not bundle a model. It doesn’t download one automatically. The release notes are explicit: you must provide your own ONNX model—specifically a segmentation model that outputs an alpha mask. Several open-source models work, including variants of U²-Net, MODNet, and RMBG-1.4, all of which are available from model hubs like Hugging Face. The tool expects the model to accept an RGB image and return a single-channel mask, but it does not restrict you to a single architecture.
Version 21.0.0 also ships with the usual round of bug fixes and minor enhancements. The image editor gets a few new hotkeys, and region capture performance has been improved on multi-monitor setups. But the background remover is the reason this release will grab attention.
What It Means for You
For Everyday Users
If you regularly share screenshots and want to quickly isolate objects, this is a genuine step forward. Before ShareX 21, Windows users had to rely on web-based tools like remove.bg, built-in Paint 3D (which Microsoft has since deprecated), or heavyweight editors like GIMP. Each came with friction: uploads, watermarks, subscriptions, or steep learning curves.
Now, you can bind background removal to a hotkey inside a tool you already use. After selecting a region, capturing a window, or opening an existing file, you can fire off the Background Remover and get a transparent PNG in seconds. No internet lag, no file size limits, no monthly credits.
The trade-off is setup. Downloading a 200 MB model file and pointing an app to a folder isn’t difficult, but it’s a speed bump. ShareX’s core audience is tech-savvy—this won’t deter most—but casual users may be confused the first time they click the tool and see a blank window asking for a model path.
For Power Users and IT Professionals
If you manage fleets of Windows machines or just obsess over your toolchain, local AI processing is the right call. Background removal typically involves sending images to a cloud API. That’s a non-starter for screenshots that may contain proprietary information, passwords, or PII. By keeping computation local, ShareX eliminates data egress and privacy risks.
Performance is model-dependent. A lightweight ONNX model like MODNet runs comfortably on a CPU, cutting out a subject in under a second on modern hardware. Larger U²-Net variants benefit from a GPU but still work without one. Because ONNX runtime leverages DirectML on Windows, any compatible GPU (NVIDIA, AMD, Intel Arc) will accelerate inference automatically.
Admins can standardize a specific model across a team and deploy it via script or GPO. ShareX stores the model path in its JSON configuration file, so you can pre-populate it during installation. No registry hacks required.
For Developers
The Background Remover is built on ONNX Runtime, which means it’s essentially a wrapper around a portable AI model. Developers who build custom ShareX workflows (e.g., automated screenshot pipelines for CI/CD dashboards) can now add a mask generation step without leaving the ecosystem. Combined with the existing custom uploader system, you could build a fully local image preprocessing pipeline.
The code is open-source (GPLv3), so the community is already experimenting with model-swapping and extended post-processing. Expect forks that bundle everything into a single download, despite Microsoft Store and winget restrictions on package size.
How We Got Here
ShareX has been the Swiss Army knife of Windows screenshot tools since 2012 (as ZScreen before that revision). It evolved from a simple Grab-and-send utility into a capture powerhouse with over 80 destinations, a built-in image editor, screen recording, OCR, and a workflow system that rivals commercial products.
AI features have been creeping in for years. The Optical Character Recognition tool uses Tesseract; a basic QR code reader shipped years ago. But heavy machine learning tasks were always deferred to cloud services. In 2024, the ShareX team began discussing local AI features on GitHub after seeing demand from privacy-conscious enterprises. The challenge: shipping a multi-hundred-megabyte model would bloat the installer and run afoul of distribution guidelines.
The answer, arrived at through community debate, was to ship the tool without the model. This mirrors what other open-source projects have done. Audacity offers AI plugins but requires you to supply models. OBS Studio’s background removal plugin works the same way. The approach keeps installers lean and sidesteps licensing complications—many ONNX models are derived from datasets with non-commercial clauses.
Microsoft’s own efforts in this space inadvertently paved the way. Windows 11’s 24H2 update expanded ONNX Runtime integration and DirectML support, making it easier for apps like ShareX to tap into hardware acceleration without writing device-specific code. Meanwhile, the rise of Hugging Face and the ONNX Model Zoo made quality segmentation models trivially available.
By mid-2026, the pieces were in place. ShareX 21.0.0 ships exactly what was promised: a local, privacy-respecting background remover that puts the user in control.
What to Do Now
If you already use ShareX, check for updates. Version 21.0.0 is available from the official website, GitHub releases, and winget (winget install ShareX.ShareX). The Microsoft Store version typically lags a few days; as of this writing, it still shows 20.x.
To enable background removal:
- Download a compatible ONNX model. Start with RMBG-1.4 (onnx version) or U²-Net from Hugging Face. The RMBG-1.4 model is well-tested: search for “RMBG-1.4 ONNX” and download the model.onnx file (roughly 200 MB).
- Place the .onnx file in a permanent location, such as
%APPDATA%\ShareX\Models. - Open ShareX, go to Tools → Background Remover.
- In the tool window, click … next to “Model path” and select your .onnx file.
- Close and reopen the tool—it should now show a preview area.
After that, you can access the remover from the right-click menu on any captured image, from the Tools menu, or by assigning a hotkey under Hotkey Settings → Tools → Background Remover.
If results look poor, experiment with different models. U²-Net excels at portraits; MODNet handles real-time video streams well and may produce cleaner edges for screenshots with solid backgrounds. The tool does not currently offer refinement brushes or edge smoothing, but the output is a standard PNG with alpha, so you can tweak it in any editor.
For command-line automation, ShareX accepts the -backgroundremover argument (undocumented in the UI but present in the CLI), which can be scripted.
Outlook
ShareX 21.0.0’s model-bring-your-own approach marks a turning point for open-source Windows tools. As local AI hardware becomes ubiquitous—NPUs are now standard in laptops and even some desktops—we’ll see more projects adopt this pattern. The next logical step is a curated model downloader inside ShareX itself, something the maintainers have hinted at in issue discussions. For now, users who value privacy have a fast, free, and fully local way to knock out backgrounds. The burden of one extra download seems a small price to pay.