Free · No Signup

Redacting for Someone Who Already Knows Half the Answer

Almost every redaction checklist tests your image against a stranger. Most images are not sent to strangers — and against a reader who already holds half the picture, the same black box can be worth almost nothing.

🔒 No upload · Runs in your browser · Instant download

Nearly all redaction advice, including most of what is on this site, quietly assumes one kind of reader: a stranger. Someone who arrives at your image holding nothing, reads whatever is legible, and leaves with what they could extract. It is a sensible default for anything posted publicly, and it gives a clean test — cover whatever a person who does not know you could work out from the frame.

But most images are not posted publicly. They go to a landlord, a support queue, a claims adjuster, a group chat, a manager, a parent, an opposing party. Those readers do not arrive holding nothing. They arrive holding the lease, the ticket history, the policy number, your name, the other half of the conversation, the seating plan of the office, the room the photo was taken in. Against a reader like that, a black box can remove nothing whatsoever — not because it was drawn badly, but because it is sitting on something they already have.

Which leads to the uncomfortable part: there is no such thing as a correctly redacted image, only an image redacted correctly for a particular audience. Redaction is subtraction, and what it subtracts depends entirely on what the reader brought with them. The same file can be generous protection for one person and decorative for the next, and nothing in the file itself records which case you are in.

This page is about establishing that before you draw anything: who is actually going to open this, what they already hold, how far the file travels after they have read it, and which of your instincts about covering things stop working when the reader is not a stranger. It is the question upstream of technique, and it is the one almost nobody writes down.

Related: finding everything that needs redacting before you start.

Mode
Shape

Drop your image here

Or click to browse · Paste with Ctrl+V also works

PNG · JPG · WebP · GIF
How It Works
1

Open

Drop your image in or paste from clipboard.

2

Pick Mode

Black Box, Blur, or Pixelate.

3

Select Areas

Rectangle, oval, or freehand lasso — then hide what you selected.

4

Download

Hit Download PNG. Done.

SquooshNeed to shrink your image after editing? Squoosh is a free browser-based image compressor with no upload required.

Visit Squoosh →
Guide

Everything below follows from one sentence: a cover removes information from a reader only in proportion to what that reader did not already have. This sounds obvious written down, and it is almost never applied, because the act of drawing a box feels absolute. The rectangle is opaque, the pixels underneath are gone, the subtraction looks total. But the subtraction happens in the reader, not in the file, and different readers are standing in very different places when your image arrives.

Two consequences pull in opposite directions. A redaction can be far weaker than it looks, because the reader already holds what you covered and is learning from what you left. And it can be pointlessly strict, covering a value the reader has in front of them while costing you the entire point of sending the image. Both are common. Neither is visible in the exported file.

The same image, two readers

Take a screenshot of a work chat with the participant names blacked out. To an outsider that is a real redaction: the names were the identifying content, they are gone, what remains is text with no owner. To a colleague in the same workspace it is barely a redaction at all. They have the channel layout, the message ordering, the avatar colours, the time of day, the length of each contribution, the way each person writes. Names are the label; everything else is the handle, and the handle was never covered. Reattributing the messages takes them seconds, and now they also know that you found this conversation worth screenshotting.

Or a bank app screenshot sent to a family member to show a charge. Your name is covered, the balance is not. The name was never the secret to that reader. The balance is the thing they did not have, and it is the one thing left legible. The cover went where the habit said, not where the information was.

Or a photograph with a face blurred, sent to a group who know the person by their coat, their dog, the car in the driveway, the kitchen. Faces are what redaction advice talks about because faces are what identifies people to strangers. Within a circle that already knows everyone, identification is not the risk — the risk is whatever the picture proves about where that person was, when, and with whom.

The pattern in all three is the same. You covered the thing that identifies the subject to someone with no prior knowledge, and the reader you actually have was never short of that. What leaked was the other axis: the amount, the date, the association, the fact that the image exists.

Why an insider gets more out of a weak cover than a stranger does

There is a second, sharper reason a knowledgeable reader is harder to defend against, and it is about the task they are performing. A stranger has to reconstruct. They must turn whatever survives your blur into a value they did not previously possess, and degraded detail genuinely costs them. Someone who already holds part of the picture only has to confirm. They arrive with a candidate answer and need enough residue to decide whether it is right.

Confirming is enormously cheaper than reconstructing. A word of about the right length in about the right place, a covered field whose shape matches a form they have handled before, a timestamp pattern, a number of lines — none of that is legible, and all of it can move a guess from probably to certainly. This is why the stranger test is not the conservative test people take it for. It is a test against a reader with no prior, and priors are exactly what your real audience has.

It also changes which tool you should reach for. Blur and pixelate are derived from the pixels underneath, so they preserve coarse structure — extent, contrast, rough shape, word and line divisions. A solid fill is not derived from anything. Against a stranger, the difference can be marginal. Against someone testing a hypothesis, coarse structure is frequently the whole of what they needed. How much of a field you can leave visible works through the same arithmetic for partial masks of structured values, and lands in the same place: a partial value is a join key for whoever already holds the list.

When the reader has the original, the boxes are the only new information

The extreme version of a knowledgeable reader is one who holds the unredacted source. This is routine and almost never planned for: the other party in a dispute has the same email thread; the coworker was in the same meeting; the company whose portal you screenshotted can see their own records; the person you are complaining about received the same message you did.

For that reader, none of the covered content is new. The covers are. The set of boxes tells them what you consider sensitive, what you consider relevant, which passages you thought worth hiding, and — from the geometry — roughly how much of each you covered. In an adversarial setting that is a meaningful disclosure of your own reasoning, and it can be read as an admission about what the covered text says.

If that applies to you, a better box is not the answer. The answers are structural: cover a uniform region rather than surgically boxing individual phrases, so the pattern carries less; or send the part you need to send and nothing around it; or drop the image and put the point in writing, where you are not simultaneously publishing a map of your own concerns. The tool cannot help with this one, because the problem is not in the pixels.

The audience is a union, not a recipient

Everything above argues for reading your audience accurately. The complication is that the audience is not a person, it is the set of everyone who will hold the file over its life, and you know only the first element of that set.

The realistic membership, in rough order of how often it is forgotten: the person you sent it to; anyone they can forward it to without consulting you; anyone with access to the place it lands, which for a support ticket, a shared inbox, a claims file or a group chat is a larger and more changeable group than the sender pictures; whatever system stores it, for as long as that system stores things; and readers in the future, including after your relationship with the original recipient has changed. A private channel bounds who you are talking to. It does not bound who ends up holding the file.

So the two directions resolve into one rule. Calibrate coverage to the widest reader the file plausibly reaches, and make deliberate, named exceptions for values the immediate reader genuinely needs. Not the other way round. The default should be the stranger standard, because the stranger is the one most of your eventual readers resemble; the knowledgeable-reader analysis then tells you where that standard is not enough, and where a single value should be left visible on purpose because covering it would make the message useless while protecting nothing.

A five-question audience sheet

Run this before you open the editor. It takes about a minute and it is the step that decides what the boxes are for.

  1. Who is the named reader, and what do they already hold? Write it out rather than gesturing at it: your name, your address, the account, the thread, the org chart, the other side of the exchange, the room. This list is what your redaction cannot take back from them.
  2. Who else can hold this without any further action from me? Forwarding, group membership, shared access to wherever it lands, and the plain fact that anyone who can see an image can screenshot it. If the honest answer is I do not know, that is the answer to plan against.
  3. What does my image, combined with what the reader already has, produce that neither had alone? This is the question that catches the real leaks. It is not what is sensitive in this picture; it is what this picture completes. An address plus a schedule. A name plus a diagnosis. A grievance plus a date plus a small group.
  4. Have I covered the thing that matters to this reader, or the thing that identifies me to strangers? Habit pushes hard toward the second. Name out loud the single item you would most mind this particular audience having, and check that it is actually under a box.
  5. Does the reader need any covered value in order to act? If a support agent cannot find your account, or a counterparty cannot tell which transaction you mean, the image has failed at its job and you will send a second, less careful one. Decide which single value stays visible, deliberately, instead of discovering it under pressure.

Question three is the one worth slowing down for. Most redaction checklists, this site included, ask what is sensitive in the frame, which is a property of the image alone. Conditional disclosure is a property of the image and the reader, and it is where images leak most quietly, because nothing in the frame looks dangerous on its own.

What the editor above can and cannot do for this

No tool knows your audience, so nothing here decides any of the above for you. A few things about how this editor actually behaves, read from its own code, do matter once you have decided.

Solid fill removes structure; blur and pixelate transform it. Black Box paints a flat fill over the selection with no dependence on what was underneath. Blur redraws the region from a copy scaled down by a factor of ten and back up, and Pixelate averages into blocks sized at a twelfth of the region's shorter side, with a floor of six pixels. Both therefore keep coarse layout: how long the value was, where the gaps fall, how many lines there are. Against a reader who is confirming rather than reading, that residue is often sufficient, which is why the safety notice on the toolbar is worth taking literally for anything you are hiding from someone with prior knowledge.

Two audiences, one session, in the right order. Coverage is painted onto the working canvas, Undo steps back through saved snapshots, Clear redraws from the original image the page still holds, and Download writes the current canvas as a freshly encoded PNG named with a timestamp — and can be pressed repeatedly. Producing a strict version and a looser version from one sitting is therefore easy, but do it strict-first: cover everything, export, then uncover only what the second audience is genuinely allowed, so that the loose file's covered regions are a subset of the strict file's. Built the other way round, the two exports overlay into the original for anyone who ends up with both.

The view is scaled, so your judgement of legibility is not evidence. The canvas is set to the image's true pixel dimensions while being displayed fitted to the panel. Text that looks like grey mush to you here is stored at full resolution and reads normally to anyone who zooms. Judge coverage at full size in a viewer, not from the preview.

Small selections are silently dropped. A rectangle under about four image pixels in either direction, or an oval under six, is discarded rather than applied, and a freehand lasso needs at least three points. If you tapped a small box over a short value and moved on, confirm it is really there before exporting.

All of this happens on the device the image is already on. The page reads the file in your browser, draws it into a canvas and hands the result back as a download; the picture does not reach this site, and there is no account or history for it to sit in. The working copy exists only while the tab is open.

Common mistakes and misconceptions

"They already know it, so I do not need to cover it." True about that reader, false about the file. The named recipient is the first holder, not the last, and everyone downstream starts from less than they do.

Treating a private channel as a bounded audience. A direct message bounds who you are speaking to. It does not bound who holds the image afterwards, because forwarding, screenshotting and shared access all happen without your involvement.

Covering the label and leaving the handle. Names are not the only way people are identified. Position in a list, message ordering, avatar colour, timing, role, phrasing and the shape of a signature all point at a person for anyone in the same room as them.

Assuming the stranger test is the strict one. It is a test against a reader with no prior knowledge. A reader with a prior needs far less residue, because confirming a guess is cheaper than reconstructing a value.

Copying the masking conventions you have seen on documents sent to you. Those conventions exist so that a document can be matched by the person who already owns the data. Reproducing them for a general audience reuses a pattern designed for the opposite situation.

Forwarding a file that was redacted for a different audience. A redaction is a decision about one reader set. Passing the same export on to a second one silently reuses a judgement that was never made about them, and this is how an image redacted for a colleague ends up attached to a public thread.

Assuming today's relationship is the permanent one. The coverage that feels right for a friendly recipient has to survive that recipient becoming a former colleague, a former partner or an opposing party, with the file still in their possession and nothing you can do about it.

Over-covering until the image cannot do its job. The failure mode is not harmless: an unusable image gets replaced by a hastier second one. Decide up front which single value stays visible, and leave it visible on purpose.

Redact for the Reader You Actually Have

A cover removes information from a reader only in proportion to what that reader did not already hold. Redaction advice almost always assumes a stranger — someone arriving with nothing — but most images go to a landlord, a support queue, a claims file, a group chat or an opposing party, all of whom arrive holding part of the answer already. Against them, covering the name can remove nothing while the amount, the date or the association sits in plain view.

A knowledgeable reader is also cheaper to satisfy. A stranger has to reconstruct a value from whatever survives; someone with a prior only has to confirm a guess, and coarse residue — the length of a word, the shape of a field, the position in a list — is often enough for that. At the limit, a reader holding the unredacted original learns nothing from the covered content and everything from the covers themselves, which map what you considered sensitive.

The working rule: your audience is not the recipient, it is the union of everyone who will hold the file over its life, so calibrate to the widest plausible reader and make one deliberate exception for any value the immediate reader needs in order to act. Ask what your image, combined with what the reader already has, produces that neither had alone — then cover for that, not for the habit. Everything in the editor above runs on the device the image is already on.

Frequently asked questions

The person I am sending this to already knows my name. Should I still cover it?

Usually yes, and for a reason that has nothing to do with them. Covering a value your reader already holds removes nothing from that reader, so if they were the only person who would ever open the file you could leave it. But the reader is rarely the audience. The file can be forwarded without asking you, stored in a ticket system or a shared inbox, attached to a case that other people handle, or re-sent months later to someone who was not in the original conversation. Every one of those readers starts from less than your named reader does, and the name is doing full work against them. The exception worth making is when the value is what makes the image useful to the person you are sending it to - an account number a support agent needs to look you up, a date that identifies which order you mean. Leave that one visible on purpose, cover everything else, and accept that you have chosen to trust where the file goes next.

Is the stranger test not the strictest one? How can someone who knows me get more out of an image?

Because they are doing a different task. A stranger has to read your image and reconstruct something from it, so degraded detail genuinely costs them. Someone who already holds part of the picture only has to confirm a guess they arrived with, and confirming needs far less than reconstructing. A blurred word of the right length in the right position, a message at a recognisable time, a covered region whose shape matches a form they have seen before - none of that is legible, and all of it is enough to move a hypothesis from probably to certainly. The stranger test is not the strict test. It is a test against a reader with no prior, and most of your readers have one.

Can the redaction itself tell someone something?

Yes, in two ways, and both matter most for a reader who already holds the original. First, a cover announces its own location and size: it says a sensitive value exists here, it is roughly this long, and it sits in this field. Against an empty-handed stranger that is minor. Against someone holding the same document, the boxes are the only part of your file that is new, and the set of them is a readable map of what you consider damaging. Second, in a dispute the choice of what to cover can be read as an admission about what the covered part says. If that is a risk in your situation, the fix is not a better box - it is covering a uniform region rather than surgically boxing the sensitive words, or not sending the image at all and saying what you need to say in text.

I have to send the same screenshot to two different audiences. Two versions or one?

Two versions is usually the right answer, with one rule attached: make the looser version a subset of the stricter one. If a reader ever holds both files, the safe parts of each fill in the covered parts of the other, and what they end up with is the union - effectively the original, with your own boxes marking where the interesting parts were. So build them in that order. Cover everything for the strictest audience first, export that, then remove coverage only where a wider allowance is genuinely justified for the second audience, so that every region covered in the loose version is also covered in the strict one. The editor above makes this cheap: coverage is painted onto the working canvas, Clear redraws the original the page is still holding, and Download can be pressed as many times as you like in one session.