Last reviewed: November 11, 2026
text remover
All three free watermark removers returned a downloadable file from the same 12 MP photo without a login or a card: watermarkremover.io gave back a 1,348,316-byte 12 MP JPEG, phototune.ai a 5,078,655-byte 12 MP PNG, and unwatermark.ai a 259,218-byte 1 MP JPEG; none left the white SAMPLE WATERMARK lettering visible.
| Free tool | What you get | Output size | Watermark left | Free allowance | Login needed |
|---|---|---|---|---|---|
| watermarkremover.io | 1,348,316-byte JPEG | 4000 x 3000 (12 MP) | No lettering; background behind the text visibly smoothed | Not measured: the free path stated no login and no card, but gave no job count | No |
| unwatermark.ai | 259,218-byte JPEG | 1154 x 866 (1 MP) | No lettering; whole photo resampled to 1 MP and softer | 3 automatic removals per day on the free plan (pricing page, 2026-11-08) | No |
| phototune.ai | 5,078,655-byte PNG | 4000 x 3000 (12 MP) | No lettering; sharpest background detail of the three | 6 free credits; one 12 MP job cost 4 (measured 6 -> 2) | No (dismissible sign-up) |
If you need the full 12 MP back, phototune.ai keeps the sharpest background detail but one job spends 4 of its 6 free credits, while watermarkremover.io is the lightest full-size download at 1,348,316 bytes with no login or card; unwatermark.ai is the smallest file at 259,218 bytes but resamples the whole photo to 1 MP, so it is the wrong pick for print or a tight crop.
Per-block results
Measured on 2026-11-11 with a different 2560 x 1440 input: each of the five opaque text blocks was compared crop by crop against that input, and the drop below is the fall in dominant ink coverage in the download.
- top-centre banner: unwatermark.ai 3.4% drop, STILL PRESENT; phototune.ai 100.0% drop, REMOVED.
- centre wordmark: unwatermark.ai 4.8% drop, STILL PRESENT; phototune.ai 99.9% drop, REMOVED.
- bottom-left box: unwatermark.ai 2.4% drop, STILL PRESENT; phototune.ai 98.4% drop, REMOVED.
- mid-right pill: unwatermark.ai 1.2% drop, STILL PRESENT; phototune.ai 4.0% drop, STILL PRESENT.
- top-left tag: unwatermark.ai 4.4% drop, STILL PRESENT; phototune.ai 100.0% drop, REMOVED.
The separate 2026-11-11 text-removal run
On 2026-11-11 a different 2560 x 1440 JPEG (3.69 MP, 835,341 bytes) carrying 5 opaque text blocks was pushed through the same tools. phototune.ai erased 4 of 5 and returned a 3,582,639-byte 2560 x 1440 PNG; unwatermark.ai erased 0 of 5 and returned a 153,672-byte 1332 x 749 JPEG; watermarkremover.io returned HTTP 403 with a Cloudflare security verification interstitial, 0 upload controls and no downloadable output file, so no output numbers exist for that tool on that date. The 2026-11-08 run in the table above did return a full-size 12 MP file from watermarkremover.io. These are two separate measurements on two different input images; neither record has been changed, and the difference is itself the finding. The original input, both real downloads and the machine-readable record of this run are published at the 2026-11-11 evidence files.
Measurement dates and the day before
The three-row table above was measured on 2026-11-08 with one 4000 x 3000 JPEG. The text-removal run below was measured on 2026-11-11 with one 2560 x 1440 JPEG. On 2026-11-09, a separate measurement with a different 2000 x 1400 JPEG carrying 4 opaque text blocks found phototune.ai erased 0 of 4 and unwatermark.ai erased 0 of 4, recorded in our remove text page. These are not the same run, neither record has been changed, and the difference is itself the finding.
What each column in the three-row table means for you
- Free tool. The service whose real downloaded file is measured in that row; no paid plan was used.
- What you get. The exact byte size and format of the file you can download.
- Output size. The pixel dimensions of that file, which decides whether it holds up for print or a tight crop.
- Watermark left. Whether the white lettering is still visible, checked crop by crop against the input.
- Free allowance. How many jobs the free path gives you: measured credit spend where we could measure it, the pricing page figure where we could not, and "not measured" where neither was available.
- Login needed. Whether an account or a card was required to reach the download.
How the downloads were checked
For the 2026-11-08 three-row table, the same 12 MP input was uploaded to each free path, the real downloaded files were measured, and the central watermark band of each output was resampled to 4000 x 3000 and compared with the same band of the input; two independent retrievals of each file returned the same byte count and MD5. For the 2026-11-11 text-removal run, five opaque text blocks were compared crop by crop against that input at the same coordinates, and a block counts as erased when its dominant ink coverage falls by more than 60%; a block marked STILL PRESENT is one whose lettering remains in the download.