Photo-upload privacy statistics

Where your uploaded photo goes: the deletion window

5 of the 10 operators put a number on it. The spread is 168×: one hour versus seven days. The other 5 either publish no window at all (2 of them, after we read their pages line by line) or were not reviewed yet.

• PicWish · picwish.com

  “All images you upload will be permanently deleted automatically from the server 1 hour after you last edited or closed the image processing page.”

  https://picwish.com/app-privacy · checked 2026-11-01

• ObjectRemover · objectremover.com

  “Uploaded images and generated results are automatically removed within 2 hours, so files do not stay on our servers.”

  https://objectremover.com · checked 2026-11-01

  Stated on the homepage marketing band; the same figure repeats on /privacy.

• VanceAI · vanceai.com

  “Private by default. Across every model in our suite, uploads are encrypted in transit and permanently deleted within 24 hours.”

  https://vanceai.com · checked 2026-11-01

• Cutout.Pro · cutout.pro

  “Photos remain in the cloud for a very short period of 24-48 hours after you have last edited the photograph, so that you can access the image and make additional modifications if you choose to do so.”

  https://www.cutout.pro/privacy · checked 2026-11-01

  The published window is a range; we use its widest value (48 h = 2 days) in the ratio.

• DeWatermark · dewatermark.com

  “Repair uploads, masks, and outputs are stored privately for up to seven days for processing and recovery.”

  https://dewatermark.com/legal/privacy · checked 2026-11-01

Two publish no deletion window at all

This is a verified absence on the pages we read, not a claim that nothing is kept. The other tools in the sample are not in this list — for them we simply haven't finished the manual pass, and their silence says more about our access than about their policy.

• cleanup.pictures — homepage, pricing and privacy read in full (3/3 pages).
• SnapEdit — all 7/7 pages read.

Free-tier export limits

Only the number the vendor put inside a sentence about the free version counts here. The biggest resolution figure on a page is almost always an API ceiling or a paid tier — see what we rejected.

• cleanup.pictures · cleanup.pictures

  “You can import and edit picture of any size in Cleanup.pictures. Export will be limited to 720px.”

  https://cleanup.pictures/ · checked 2026-11-01

• SnapEdit · snapedit.app

  “Free for everyone $0 per month … Standard quality export (max 1600px).”

  https://snapedit.app/pricing · checked 2026-11-01

• jpgRM · jpgrm.com quotable, not in any ratio

  “Why is there a paid plan? The free plan limits downloaded image size to 720px.”

  https://jpgrm.com · checked 2026-11-01

  We could only read 1 of its 7 pages, so this quote is reportable but never inside a ratio.

• DeWatermark · dewatermark.com

  “Drop or choose JPG, PNG, WebP, or AVIF · up to 10 MB.”

  https://dewatermark.com · checked 2026-11-01

  The upload control itself. Two other figures we rejected: remove.bg's “12 MB” comes from a 2019 API changelog entry, and VanceAI's “up to 10 MB” sits in paid-tier copy.

Who says they never store it — and who says they'll learn from it

Marketing copy such as “built on models trained on millions of images” is not counted as a training clause: it describes how the product was made, not what it does with the file you just uploaded. Only wording addressed to your file qualifies.

The table (20 tools)

Rows are brands; ratios above are per operator, because two storefronts on one backend are one policy. “not published” means we found no such sentence; “none found” means we read the pages line by line and can attest to the absence. Rows marked · not reviewed are our own to-do list, not findings.

Excluded for thin coverage

Fewer than 3 of their own pages could be read, so their silence is about our access, not their policy. They are out of every denominator above.

• jpgrm.com — 1/7 pages read.
• photoroom.com — 2/7 pages read.
• hama.app — 2/7 pages read.
• magiceraser.io — 1/7 pages read.
• photofix.io — 0/7 pages read.

Numbers we rejected

Publishing the figures we threw away is the only way a reader can tell this was a method and not a scrape. Each of these matched a pattern a naive audit would have reported.

365 days “retention” (Watermarkly, jpgRM) · from pricing pages

Both come from a plan table reading “1-Year Access · 30-day money-back guarantee”. A subscription length is not a promise about your file.

15 days “retention” (Pixelcut) · from privacy policy

That is the deadline for responding to a data-subject request, not a deletion window.

90 days “retention” (Cutout.Pro) · from privacy policy

The sentence is about server logs. The same vendor's photos are gone in 24–48 hours.

12 MB upload cap (remove.bg) · from API changelog, 2019-09-25

A developer-portal figure about a different product surface than the web app your reader opens.

8,000 px / 5,600 px “max resolution” · from API format tables and paid tiers

The biggest pixel number on a page is almost never the free tier's cap; we publish only caps the vendor states about the free version.

Which host actually receives your file

None of the 10 operators disclose which host actually receives the file, and a static page could not reveal it either: the upload destination is only observable from the network request. We report this column as not observed rather than guessing, and we will not print a destination for any tool until a browser-capture pass has actually recorded it.

Methodology

Who we are, first. We make desktop software that processes your images without uploading them, and we collect no usage telemetry. So this audit has no conflict of interest — and no inside knowledge to lean on either. Everything below comes from the vendors' own public pages, and you can re-run it.

• Sample. 20 online image tools (19 operators) that require uploading a photo before you get a result — chosen as the tools most often listed in the “best online watermark remover / background remover / upscaler” roundups writers publish, not as a random draw.
• What was read. Each vendor's own homepage, pricing, FAQ/help, privacy policy and terms — 7 pages per site, fetched 2026-11-01, one request per page, no login, nothing behind a paywall, JavaScript left unrendered.
• Extraction. A standard-library Python script pulled the candidate sentences; a human then read every quoted sentence against the archived page text and signed off row by row. Only signed rows became figures — the script's raw output is not this page.
• Counting unit. The operator, not the brand. 20 storefronts run on 19 backends (PixelBin serves both WatermarkRemover.io and Upscale.media), and ratios over brands would quote one policy text twice.
• Denominator. The 10 operators that were both reviewed and readable on at least 3 of their own pages.
• What this cannot show. Where the file is POSTed; what happens after upload; whether a deletion promise is honoured; and any figure that only ever appears inside a JavaScript-rendered app.
• Reproduction. The fetcher, the per-site quote extractor and the archived text of every page behind a figure here are kept with the data and sent on request — email us for the bundle. A correction is welcome the same way; we re-check quarterly and note the change date.

Download the data

• photo-privacy-stats-2026-11.csv — one row per tool, the same columns as the table above.
• photo-privacy-stats-deletion-window-2026-11.svg — the chart, with the attribution line baked in.
• Press kit — the same figures on one page, plus the embed code for the chart, the logo files and who to ask about a number.

Licensed CC BY 4.0. Attribution that works: “Only 5 of the 10 audited operators publish a concrete auto-deletion window for the photo you upload. — PowerSoftware photo-upload audit, 2026-11-01, https://www.powersoftware.app/en-US/photo-privacy-stats#stat-deletion-window”

If this is the reason you're here

• Remove a watermark from a photo without uploading it →
• Upscale an image locally, free of export-resolution caps →
• AI photo editor comparison: what each tool does with your file →