Exploring Times: Multiple Tools One Destination

Excise.

How this works: Excise runs an on-device subject-detection model in your browser — no photo is ever uploaded. The model itself (a few MB) is fetched from a public library the first time you use the tool, so that first run needs an internet connection and takes a bit longer. It's tuned for photos with one clear person or foreground subject, not arbitrary objects.

Drop a photo here, or tap to choose one

Works best with one person or subject clearly in frame

—

Original

Cut out

Tap "Remove background" to preview

Loading the on-device model…

Replace background with

Edge softness

2px

Output format PNG JPEG WebP

About Excise

Excise cuts the background out of a photo using an on-device subject-detection model that runs in your browser. Your image is never uploaded — only the model itself is fetched from a library over the network, once, the first time you use the tool.

Features

• Automatic subject detection — no manual masking
• Transparent, solid color, or blurred backgrounds
• Adjustable edge softness for a cleaner cutout
• Side-by-side original and result preview
• Export as PNG, JPEG, or WebP
• Photo itself is never uploaded to a server

How to use it

1. Drop in a photo with one clear person or subject in frame.
2. Tap "Remove background." The first run fetches the detection model, so it takes a little longer.
3. Choose what to put behind the subject: transparent, a solid color, or a blurred version of the original.
4. Adjust edge softness if the cutout looks too hard or too fuzzy, then tap "Remove background" again to refresh the preview.
5. Pick an output format and download the result.

Frequently asked questions

Is my photo uploaded anywhere?

No. Detection runs entirely on your device. The only network request is a one-time download of the detection model itself, not your image.

Why didn't it find my subject?

The model is tuned for photos with one clear person or foreground subject well-lit and mostly unobstructed. Busy scenes, small or partial subjects, or objects instead of people are harder for it to isolate cleanly.

Why does the first run take longer?

The detection model has to be downloaded the first time you use the tool on a given page load. After that, it stays in memory and later runs are much faster.

Why do the edges look a bit rough?

Automatic detection won't be pixel-perfect, especially around hair or fine detail. Raising the edge softness slider blends the boundary more gently.

Can I use this on objects instead of people?

It's built around a person-detection model, so it works best with people. Results on other objects will be inconsistent.