About Glimbench
Glimbench is an independent AI tool workbench. Its first live tool removes image backgrounds and exports cutouts you can use in designs, product photos and social posts.
Who runs the service?
Glimbench is operated by Sentinel, an independent developer. Questions or problem reports can be sent to the support address below.
What is available today?
The Background Remover accepts JPG, PNG and WebP. It can create transparent PNGs at the original pixel dimensions, replace the background with a solid color and export to standard or custom canvas sizes. The free plan supports unlimited single-image use and batches of up to 5 images, with a 25 MB limit per file.
Audio and text tools shown as Coming soon are planned and cannot be used yet. The pricing page describes Pro features and whether purchases are open.
How image processing works
The standard model is ormbg, using ONNX and WebAssembly in a background worker. The browser downloads the model and processing software, predicts a foreground mask, resizes that mask to the original image dimensions and composites the output. Original-size export preserves pixel dimensions; it does not guarantee perfect fine detail.
The Pro high-quality option uses BiRefNet lite and requires a supported WebGPU device. Hardware support and available memory affect whether it can run.
Privacy and network access
Selected image pixels are processed on your device and are not uploaded to Glimbench. Downloading model files and runtime code requires network access. Website analytics and Pro license checks also use the network and are described in the Privacy Policy.
Known limitations
Fine hair, fur, transparent glass, shadows and colors close to the background can produce imperfect cutouts. Check edges on both light and dark backgrounds. There is no manual erase or restore brush.
The first use includes a model download of about 44 MB for Standard, or about 115 MB for High quality. Processing time varies by device and image. Large images may exceed a phone’s memory or canvas limits; successful use on one device is not a guarantee for every browser.
Cached models can be reused. Offline processing requires the page and processing software to remain available as well; this site is not a fully installed offline app.
Tell us what went wrong
Include your browser, device, image dimensions and the step that failed. Do not email private photos unless you want to share them. We cannot inspect your selected files because they stay on your device.
Open-source models and licenses
The standard model uses Apache-2.0 licensing and BiRefNet lite uses MIT licensing. The license page lists the models and processing libraries. You remain responsible for the rights to the images you process.