Is Your "Free" Background Remover Uploading Your Images to a Server?

Here's a question most people never ask when they drop a photo into a "free background remover": where does this image actually go?

The answer, for 7 out of the 8 tools I tested: it gets uploaded to a remote server. Sometimes to Amazon AWS. Sometimes to a Cloudflare-backed origin. Sometimes to an IP address owned by a company you've never heard of. The homepage might say "free," "easy," "instant." It rarely says "we're sending your image to our servers where it will be stored for an unspecified period, processed by third-party infrastructure, and potentially used for purposes beyond the background removal you requested."

So I tested them. Here's what I found.

How I tested

The methodology is simple and you can replicate it yourself in about 30 seconds per tool:

1. Open the tool in Chrome or Firefox
2. Open Developer Tools (F12 or Ctrl+Shift+I)
3. Go to the Network tab
4. Check "Preserve log" so requests don't disappear after navigation
5. Upload an image and process it
6. Look at every network request that fires during processing. Filter by "Fetch/XHR" to see API calls, or "Img" to see image transfers
7. Any request sending a payload that matches your image size (in KB/MB) to a remote domain is an upload

I tested with a 2.4MB JPEG photo of a coffee mug on a wooden table. A standard product photo. Nothing sensitive. The kind of image someone removes the background from a thousand times a day.

What "stored for a limited period" actually means

remove.bg's privacy policy is more detailed than most, and it still doesn't answer the basic questions. "A limited period" — is that 5 minutes? 24 hours? 30 days? Until their next database cleanup job runs? The policy doesn't say.

Clipping Magic says images are "deleted shortly after processing." "Shortly" is doing Olympic-level heavy lifting in that sentence.

Adobe Express falls under Adobe's general privacy policy, which is a 12,000-word document that covers everything from Creative Cloud to PDF reader telemetry. Buried somewhere in there is language about how they process "your content" to "provide the services." Adobe also had a widely publicized incident in 2023 where users discovered their Creative Cloud content was being used for AI training by default. They updated the policy after backlash. The current policy still requires you to opt out of content analysis separately from the main privacy settings.

The pattern across all of these: the company is probably not doing anything malicious with your images. But the policies are written to give them maximum flexibility with minimum accountability. If they decide to use anonymized processing data for model training next year, their policy probably already allows it. If a junior engineer misconfigures an S3 bucket and your images are publicly accessible for 4 hours before someone notices — do you think you'd be notified? The policy doesn't require it.

The one that doesn't upload: BackgroundDelete

This isn't a policy difference. It's an architectural difference. BackgroundDelete doesn't have a server to upload images to. The AI model (ISNet, the same class of segmentation neural network that powers the server-side tools) is compiled to WebAssembly and runs entirely inside your browser. The image data stays in your computer's memory, gets processed by the model, and the result gets written to a canvas element that you then download as a PNG. At no point does any image data traverse the network.

You can verify this yourself in 30 seconds:

1. Open BackgroundDelete
2. Press F12, go to Network tab
3. Check "Preserve log"
4. Drop an image and process it
5. Filter the network log for any POST request with a payload >100KB

There won't be one. The only network activity you'll see is the AI model downloading from jsDelivr's CDN (first use only — it's cached after that). After the model is cached, you can turn off your WiFi and the tool keeps working. Try that with remove.bg.

Why browser-based AI matters beyond privacy

The privacy advantage is obvious. But there are other reasons browser-based AI processing is better for certain use cases:

No upload/download time. A 5MB product photo takes 2-4 seconds to upload on a typical home connection, then another 2-4 seconds to download the result. With browser-based processing, you skip both round trips. On slow connections, this is huge.

No server queue. Free cloud tools deprioritize free users when servers are busy. You've seen the "You are 14th in queue — estimated wait: 3 minutes" messages. Browser processing has no queue. It runs on your device, at your device's speed, immediately.

No rate limits or credit systems. Cloud tools meter your usage because every image costs them GPU time. Browser processing costs them nothing after the initial model download — it's your CPU/GPU doing the work. That means no artificial limits on how many images you can process.

The tradeoffs

Browser-based AI isn't magic. Here's what you give up compared to server-side tools:

• Model size is limited. Server-side tools can run 500MB+ models on NVIDIA A100 GPUs. Browser models are 4-80MB, quantized to fit in WebAssembly. The quality gap is real, especially for hair and transparent objects. It's not huge — probably 5-10% worse on the hardest cases — but it's measurable.
• First-use download. The AI model downloads on first visit (4MB for Fast mode, 40MB for Balanced, 80MB for Best). On a slow connection, that's a one-time wait of 10-60 seconds. After that, it's cached and runs offline.
• Device performance matters. Processing is CPU/GPU-bound. A 2015 laptop will be noticeably slower than a 2023 MacBook. Server-side processing offloads this to data center GPUs that are faster than any consumer device.

For the vast majority of use cases — product photos, portraits, logos, document scans — the quality difference is barely noticeable and the privacy/ convenience tradeoff strongly favors browser-based processing.

How to check any background removal tool yourself

You don't have to take my word for any of this. Here's the 3-step verification that works on any web-based tool:

1. Open DevTools Network tab before you upload anything. Check "Preserve log."
2. Upload and process your image. Watch the network activity.
3. Look for uploads. Any POST or PUT request with a payload approximating your image size means your image left your device. The destination domain tells you where it went.

If you see nothing but CDN downloads (jsDelivr, cdnjs, unpkg), the tool is processing locally. If you see requests to api.[toolname].com with multi-megabyte payloads, your image is on their server.

Try BackgroundDelete — open your Network tab and verify nothing gets uploaded →