Pricing

“Free” is not one thing — it is a metered allowance, a resolution, and a file that leaves your machine

Three watermark removal sites publish a free tier, and none of the three means the same thing by it. One gives three credits a month. One gives ten credits a day, where a picture costs between one and three of them. One does not sell you an image count at all — it sells you three automatic removals a day and twenty manual ones. Below: the published numbers, read off each pricing page on September 30, 2026, the arithmetic that turns them into pictures, and a separate measurement of the cost nobody prints — what a capped output resolution takes out of your image.

Competitor figures are quotations from public pricing pages retrieved on September 30, 2026 (HTML sizes: 229,704 / 183,631 / 81,997 bytes) and may change. Anything we calculated from them is labelled as derived. Resolution and timing figures come from two headless Chrome 153 runs on Windows the same day, on synthetic 4000 × 3000 scenes built from a fixed seed. Scope and limits are stated at the bottom of the page.

The answer, first

1. A credit is not a picture, and the exchange rate is the whole comparison. One site states 1 Credit = 1 Image = 2 PDF Pages. Another says most tools use 1–3 credits per image — a threefold range on the same product. Without that rate, two prices cannot be compared at all.
2. The three free tiers differ by roughly two orders of magnitude. In pictures per month, using each site's own published exchange rate: 3 on one, 90 to 300 on the second, and 90 automatic or 600 manual on the third.
3. No free allowance carries over. Daily credits that do not accumulate, monthly credits that expire, and subscription credits valid 30 days. Paid one-time packs are the only credits that survive, and only on some plans.
4. The cost that is never printed is resolution. Measured on a 4000 × 3000 scene: capping the output at 1024 px removed 96.3% of the file's bytes and left 3.9% of the original fine detail, while the image still scored 30.5 dB — a number that sounds fine and is not the number you care about.
5. Two of the three state in writing that they do not refund. Both give the same reason — “the high cost of AI” — and both point you at the free plan first. That is an instruction worth taking literally.

None of this makes the paid tiers a bad deal. It means “free” is a unit of account you have to convert before you can tell whether you are being offered three pictures or three hundred.

Currency 1: money, as published

These are the prices on the page, not averages or estimates. The third site rendered its prices in yen on our connection, so its figures are quoted in yen with no conversion applied.

Two things fall out of that table. The first is that the cheapest rate always sits behind the largest commitment: ¥2.09 per credit requires buying 12,000 of them, and $0.011 per credit requires buying 30,000. The second is the gap between the two ends of one site's own list — unwatermark.ai sells 50 credits for $4.99 ($0.100 each) and 30,000 for $330 ($0.011 each), a ninefold spread on identical goods. Buying small is expensive; that is the actual shape of “pay as you go”.

One number we will not give you: a single “average price of watermark removal”. The sites use three different units, one of them does not publish its exchange rate, and one prices in a currency that depends on where you are reading from. Any average would be made up.

Currency 2: pictures, which is what you are actually spending

Here is the same information converted into the unit you care about, using each site's own words for the conversion.

The spread between the smallest and largest free tier here is about 200×, and none of it is visible in the word “free” on a landing page. If you have one photo, all three work. If you have forty photos from one shoot, the smallest tier is over before you finish, and the largest handles it in a morning.

Two details in that table are worth more than the headline numbers.

One site's own page contains two free allowances that do not reconcile. watermarkremover.io's FAQ says “You can try our service for up to 2 images daily without creating an account” and, a few lines later, “After you sign up, you get 3 free credits” with three free downloads a month. Read at face value, not signing up is the more generous of the two. We are not going to guess which one the software actually enforces — we did not test it — but if you are choosing whether to create an account, it is a question worth asking before you give them an email address.

unwatermark.ai meters quality separately from quantity. Its free plan allows 3 automatic removals a day but 20 manual ones. The automatic path — the one that finds the mark for you — is the scarce one by a factor of nearly seven. A free tier that charges you six times more for the convenient mode is a design decision, and it tells you which mode costs them money.

Currency 3: pixels — the cost nobody prints

The three pricing pages describe what you may do and how often. None of them says what comes out the other end. That matters because “HD quality” appears on phototune.ai's paid plans as a benefit, which is the standard way of saying the cheaper tier is not HD — and because a resolution cap is the one cost that is permanent. You can buy more credits tomorrow. You cannot buy back detail that was averaged away.

So we measured it. One synthetic 4000 × 3000 scene, built from a fixed seed, encoded as JPEG at quality 0.92 — the file you would hand to any of these tools — and then reduced to a series of capped longest edges and written out again. Each result was scaled back to 4000 × 3000 so the comparison is like for like, and scored two ways: how far it lands from the original, and how much fine detail survives, measured as the mean absolute Laplacian of the luma channel.

Read the error column and the detail column together, because they disagree, and the disagreement is the point. Between the 2048 cap and the 800 cap the average error moves from 4.737 to 5.498 — barely anything, and PSNR from 32.0 dB to 30.0 dB, which is 2 dB on a scale where 2 dB is not much. By those two measures, all five rows are roughly the same picture.

By the third measure they are not remotely the same picture. Detail kept falls from 10.1% to 3.0%. The file is 97.7% smaller and looks, in a thumbnail and in a whole-image score, almost identical. It has lost the fine structure — the edges, the texture, the small type — that a whole-frame average cannot see. This is the same effect documented from the other direction on our quality page: a whole-image score dilutes local damage, and here it dilutes the entire loss.

Because part of that drop is the grain being averaged out rather than structure being destroyed, we ran the same experiment on a version of the scene with no grain added, so the number isolates structural loss.

Even with nothing to average away but the scene itself, cutting to 2048 px keeps 13.6% of the fine detail and cutting to 1024 px keeps 7.7%, while the error stays under 1.4 of 255 and the PSNR stays above 35 dB. If you are judging an output by whether it looks like your photo, you will pass all five rows. If you are judging it by whether the small text in the corner is still readable, or whether the repaired patch can be blended without smearing, they are not close.

That is the practical reason a cap matters specifically for watermark removal: the fill has to borrow texture from the pixels around the hole. Shrink the picture and you shrink the evidence the repair depends on.

Currency 4: time, and the bytes that leave

Measured end to end in Chrome 153 on the same 12 MP file: decoding it, resizing it, and writing the result out took 657 ms at the 1800 px cap used on this site's homepage and text removal page, and 519 ms at the logo page's 2400 px cap. Output was 4,170,537 bytes and 7,563,408 bytes respectively. Nothing crossed the network, so nothing waited on the network.

For a tool that runs on a server, the same file has to go out and come back. We measured the file, not the transfer, so here is the arithmetic rather than a claim: our 12 MP JPEG at quality 0.92 is 3,748,049 bytes, and the lossless PNG of the same scene is 25,793,942 bytes. At 20 Mbit/s of usable upstream — a number we assumed, not measured — that is roughly 1.5 seconds out for the JPEG and about 10 seconds for the PNG, before the server does any work and before the result comes back. The measured part of that sentence is the byte count; the rest is division.

That gap is also the privacy line, and it is the one currency that cannot be bought back with credits. Every byte above is a byte that leaves your device on an upload-based tool and stays on it on a browser-based one. If you want to check which kind you are using, including on this site, the three methods on our verification page take about a minute and need no special software.

What the free tier never says out loud

Three clauses appear on the pages we read, in the vendors' own words, and they are the fine print of “free”:

• No refunds, twice, for the same stated reason. “Due to the high cost of AI, we cannot offer refunds. Please explore the free plan first.” And, on another site: “Due to the high cost of AI, we cannot provide refunds. If needed, please understand the free plan first.” Both vendors direct you to the free tier — take that as the instruction it is.
• Credits expire, and the free ones expire fastest. Daily credits that do not carry over; monthly credits that disappear; subscription credits valid 30 days and cleared at renewal. One site keeps paid one-time packs valid for a year; another says one-time credits never expire. That asymmetry is the point: the thing you paid for lasts, the thing you were given does not.
• Subscription credits and one-time credits are different products. One site spells it out: subscription credits are valid 30 days and are cleared at each renewal, while one-time credit packs “are permanently valid”. Two prices per credit on the same page for what looks like the same token is not a mistake.

None of these are hidden — all three are published. They are simply never in the headline, and the headline is the only part most people read.

How to read any free tier in about a minute

1. Find the unit before the number. Is the allowance in credits, in images, or in operations? If it is credits, find the exchange rate — “1 Credit = 1 Image” and “1–3 credits per image” are different products with the same price tag.
2. Find the reset and the carry-over. Daily-with-no-carry-over and monthly-with-no-carry-over behave completely differently if you work in bursts. Ten a day beats three a month by a factor of about 100.
3. Check what the paid tier lists as a benefit. If “No watermarks” and “HD quality” are listed under the paid plan, the free tier is the tier those benefits are being sold against. The page is telling you the shape of the difference even when it does not name it.
4. Compare output dimensions, not promises. Open the result's properties. If it is smaller than what you put in, that is a resolution cap, and the measurement above is what it costs.
5. Find out whether the picture leaves your machine. It is a separate axis from price, and no tier makes it optional. Three checks are on the verification page.
6. Treat the free tier as the trial before you spend anything, because that is the role the vendors themselves assign it when they decline to refund.

What this site charges, and what it costs instead

Nothing, and there is nothing to meter: no account, no credits, no daily or monthly allowance, and no limit on how many images you run. The image never leaves the browser — zero bytes of it cross the network, which is the same fact measured from three directions on the verification page and the no-signup page.

The cost is different in kind rather than absent. The engine on the homepage and the text page caps the longest edge at 1800 px; the engine on the logo page caps it at 2400 px. Those are resolution caps, not byte limits — the number above is the honest description of what they do to a large file, and it is the same class of cost as any other cap in this category. It is also one image at a time: there is no batch, no queue, and no history, which is covered on the no-signup page.

Where that trade is a good one: sensitive images, one-offs, and anything where “how many do I have left” is a question you would rather not be asking. Where it is not: a hundred product photos on a deadline, where a metered cloud tool with a batch upload is the right answer even at $0.026 an image.

Scope of these figures. Competitor prices and free-tier wording are quotations from public pricing pages retrieved on September 30, 2026 (watermarkremover.io/pricing 229,704 bytes; phototune.ai/pricing 183,631 bytes; unwatermark.ai/pricing 81,997 bytes). They may change at any time. One site rendered prices in yen on our connection; no exchange rate has been applied and no currency conversion has been performed. Every figure we calculated rather than read — price per credit at the best rate, pictures per month, cost per image — is labelled as derived and uses that site's own published exchange rate. Where a site does not publish an exchange rate, we did not invent one. Measurements come from two headless Chrome 153 runs on Windows the same day on synthetic 4000 × 3000 scenes generated from a fixed seed (gradient, 90 soft radial blobs, 120 thin strokes, and, in the first run only, uniform per-pixel noise of ±11 levels per channel). Caps were applied by drawing the source into a smaller canvas with high-quality smoothing, then drawn back to 4000 × 3000 before comparison. Detail kept is the mean absolute Laplacian of the luma channel after scaling back up, expressed as a share of the uncapped value; it counts any local variation, which is why the grain-free run is reported separately. Timing covers decode, resize and encode only. We did not create accounts, upload anything, or test any competitor's actual output; we did not measure network throughput, and the transfer estimates are explicitly arithmetic from measured byte counts against an assumed 20 Mbit/s. We do not publish, and cannot know, any site's traffic, ranking or index counts.