Guide
One sentence carries this page: a photo of a screen contains two information sets — the screen’s and the room’s — and the screen’s half arrives tilted, soft and glared, which means most of the fix belongs at capture time rather than in any editor.
Why you have a photo and not a screenshot
It is worth starting here, because the usual advice stops at “take a screenshot instead” and that advice assumes a device you control. A great many screen photos exist for reasons that no amount of good practice removes.
The screen may belong to someone else — a colleague holding up their phone, a clerk turning a monitor toward you, a customer showing you an order. Screen capture may be blocked, logged, or against policy on a managed machine, or the content may be protected so that a capture comes out as a black frame. The display may have no capture path at all: a kiosk, an ATM, a parking meter, a point-of-sale terminal, a car head unit, a lab or medical instrument, a departures board, a television. You may be looking at a machine you are not allowed to put a file onto or take a file off of, which makes a photo the only way the information crosses the gap. And very often it is simply speed: an error dialog that was going to vanish, a toast notification, a loading state you could not reproduce, and the camera was the thing already in your hand.
All of those are legitimate, and the page below assumes you are in one of them. The point is not that the photo was a mistake. It is that a photo of a screen needs a different first move than a screenshot does, and that first move is not drawing a box.
The frame is the bigger half of the problem
Before you redact anything inside the screen, look at everything outside it. A screenshot cannot contain a room; a photo of a screen almost always does, and the contents of that room identify the place and the person far more reliably than most of what is on the display.
Work outward from the screen edge:
The glass itself. A screen is a mirror with a picture on it. Dark regions of the display reflect whatever is in front of them — you, holding the camera; the window behind you and the street through it; a second monitor with another application open on it; the ceiling, the lights, the layout of the room. Reflected content is real content, and because it is low-contrast you are unlikely to notice it on the small preview. The same mechanics are covered in more detail on blur a mirror selfie reflection, and they apply to a monitor just as much as to a bathroom mirror.
The bezel and the hardware. Asset tags, inventory stickers, serial numbers, a service label, a logo that names the employer, a webcam cover with a company sticker on it, the model of phone someone is holding.
The desk. This is where the real leaks live. Sticky notes, an open notebook, a badge or lanyard, a letter or envelope with an address on it, a prescription bottle, a parcel label, a second phone with its lock screen showing notifications, keys, a bank card lying face up, a calendar, a name plate.
Hands and people. A hand holding the phone carries a ring, a tattoo, a watch, a visible wrist band. A reflection or a shoulder brings in a bystander who did not agree to be in a support ticket.
The room. Wall colour, art, a view, a floor plan, a doorway with a room number on it. Any one of them is nothing; together they put a named person in a named building.
The important asymmetry: the screen contents are the thing you are concentrating on, so you redact them carefully. The room is the thing you are not looking at, so it goes out unmodified. That is the wrong way round for how much each one gives away.
Soft focus, glare and moire are not redactions
This is the misconception that does the most damage, and it is specific to optical capture because a screenshot never produces it. A photographed screen is unevenly readable by nature: the near side is sharp and the far side is soft because the plane of the screen is at an angle to the plane of focus, a highlight sits across one corner, a regular interference pattern may appear where the camera’s sensor grid and the screen’s pixel grid disagree, and a line of small type is simply at the edge of what the lens resolved.
The temptation is to leave those regions alone, on the grounds that they are already illegible. Three reasons not to.
First, you are judging legibility at the wrong size. The editor on this page sets the canvas to the image’s true pixel dimensions but displays it scaled to fit the page, inside a scrolling area capped at about seventy per cent of the window height. A recent phone photo is many thousands of pixels across, so the view you are reading is a substantial reduction. An illegible smudge at that reduction is often clean type at one hundred per cent.
Second, softness and glare are contrast problems, not data problems. The characters are still present as a low-contrast signal, and contrast is adjustable by whoever receives the file. Sharpening, levels, auto-tone in an ordinary image viewer — none of those recover anything from a region of uniform fill, but they can absolutely pull readable text out of a region that merely looked finished to you. The test that matters is not “can I read it” but “is the information gone”, and for an optically soft region the answer is no.
Third, the audience may not be a pair of eyes. Automated readers do not need the character to look crisp to a person.
The rule is the same one that applies to a weak blur filter: if you would have covered the value when it was sharp, cover it when it is soft.
A photographed screen is not axis-aligned — and two of the three shapes are
Here is where the tool mechanics stop being abstract. The editor above offers three selection shapes, and they are not interchangeable on an angled capture.
The rectangle is drawn axis-aligned to the image: you drag two corners and you get a box whose sides are parallel to the edges of the file. The oval is the same — it is drawn from the bounding box of your drag, with its rotation fixed at zero. Neither can be tilted. On a screenshot that is never a limitation, because every row of text in a screenshot is already horizontal. On a photo of a screen taken from anywhere but dead square, text rows run at a few degrees of tilt, and the rows at the top and bottom of the screen converge toward a vanishing point, so they are not even parallel to each other.
Against a tilted row, an axis-aligned rectangle can only be wrong in one of two directions. Keep it tight enough to avoid swallowing the line above and the end of the value sticks out past the corner; make it tall enough to catch the whole value and it covers a wedge of the surrounding interface as well. The first failure is the dangerous one, and it is easy to miss because the leftover is a couple of characters at an edge rather than a conspicuous fragment in the middle.
The freehand lasso takes an arbitrary outline, which is the answer: trace along the row at whatever angle it sits at, following the tilt, and close the loop. It needs at least three points and a bounding box of more than a pixel or two in each direction, otherwise the drag is discarded without a message. It works by touch as well as by mouse, so this is practical on the phone that took the photo.
The fallback, when the tilt is severe, is to stop being precise and cover generously. An axis-aligned box that takes in the whole row, both neighbours and the margin either side is a perfectly good redaction; it just hides more context than you wanted. Overcovering is cheap. Clipping a value is not.
How the effect strength is computed, and why that matters at an angle
One more detail that is invisible until it bites. For Blur and Pixelate, the editor copies the bounding box of the region you drew, applies the effect to that copy, and then clips the result to the shape. So the strength of the effect is set by the bounding box, not by the shape inside it.
Blur downsamples the bounding box by a factor of ten and scales it back up. Pixelate uses square blocks sized at roughly a twelfth of the bounding box’s shorter side, with a floor of six pixels. Both of those are relative measures, which means the same gesture gives you different absolute strength depending on how big a box your selection happens to imply.
On an angled capture that interacts with the geometry in an unhelpful way. A long, thin selection traced along a tilted row has a bounding box inflated by the diagonal — much taller than the row itself — so the effect applied inside it is stronger than the same trace would produce on a horizontal row. Conversely, a short horizontal strip over a single field has a very shallow bounding box, so the pixelate blocks fall to the six-pixel floor and each block is small enough that the shape of a character can survive across several of them. None of this is a reason to learn the formula. It is a reason not to form an intuition about how strong Blur and Pixelate are on this kind of image, and to use Black Box for anything that genuinely has to go. A solid fill has no strength parameter to get wrong.
A running order when the capture is a photo of a screen
1. Ask whether a screenshot or a copy-paste is available after all. Thirty seconds of checking saves all of the rest. If the content exists as text somewhere, pasting the text and redacting a sentence beats redacting a photograph of the text.
2. Retake the photo before you edit it. This is the step people skip, and it is the only one that fixes the frame. The editor above has no crop, no resize and no rotate — every operation replaces pixels inside a region you draw, and nothing changes the boundaries of the image — so framing is a capture-time decision and cannot be undone later. Move closer, square up to the screen, and fill the frame with just the part of the display you actually need. A retake that excludes the desk removes an entire category of work.
3. Kill the reflection on the retake too. Change your angle, step out of the bright part of the room, turn off the lamp behind you, or tilt the screen. Coping with a reflection by redacting it is far more work than preventing it, and the reflection you failed to notice is the one that matters.
4. Now list what has to go, inside the screen and outside it. Two lists, counted. The second list — the room list — is the one that will surprise you.
5. Cover with Black Box, lasso on tilted rows, and overshoot. Extend into the surrounding chrome deliberately. On a photo there is no pixel-exact edge to aim at anyway, because the characters have soft edges to begin with.
6. Verify on the exported file at full size. Download the PNG, open the saved file rather than trusting the canvas you were drawing on, and zoom well past one hundred per cent — particularly on the soft and glared regions you decided to leave, and particularly along the far edge of the screen where the type is smallest. This is the check that catches the clipped corner of an axis-aligned box.
7. Send the export, not the original. Download PNG re-encodes the canvas from scratch, so the downloaded file carries none of the camera file’s tags. The original keeps its capture time, device details and any location data, so it should not be the file that goes into the thread.
What the editor above does and does not do
The tool on this page is region-based: three methods — Black Box, Blur, Pixelate — across three selection shapes, rectangle, oval and freehand lasso. Each operation replaces the pixels inside the region you drew. Black Box fills it with a fixed #111111; Blur downsamples the region heavily and scales it back up; Pixelate replaces each block with a single sampled colour. Undo steps back one cover, Clear restores the untouched image and discards all of them, and Download PNG writes the current canvas out as hideshot-<timestamp>.png, re-encoded from scratch rather than copied from the source file. Everything happens in your browser on your own machine; nothing is sent anywhere, which is also why there is no stored copy and no sharing link.
Four limits matter specifically for a photographed screen. There is no crop, resize or rotate of any kind — the canvas is set to the source image’s exact pixel dimensions and stays there — so you cannot trim the room out of the frame here and you cannot straighten a tilted capture. There is no perspective or deskew correction, which is why the lasso rather than the rectangle is the tool for an angled row. There is no sharpen, levels or contrast control, so the honest test of whether a soft region is really illegible has to be done in a viewer or editor after you export. And the export is a PNG at the source’s pixel dimensions, which for a photograph means a notably larger file than the JPEG you started with.
Common mistakes and misconceptions
“It is just a photo of a screen, so I treat it like a screenshot.” The content is screen content but the file is a camera file, and the two pipelines behave differently all the way down — metadata, compression, sharpness, framing. The broader comparison is on redacting a screenshot vs redacting a photo; this page is the case where both sets of problems arrive in one image.
“The parts that came out blurry are effectively redacted.” No. Optical softness is unverified, not removed, and you are judging it on a downscaled view of a much larger file. Cover it.
“Glare already wiped that corner out.” A highlight compresses contrast, and contrast is recoverable in a way that a uniform fill is not. Only a completely clipped region is genuinely empty, and you cannot tell which it is by looking at it on the page.
“I redacted everything on the screen.” Quite likely true, and quite likely not the problem. Go back to the frame: the glass, the bezel, the desk, the hands, the room.
“The rectangle is fine, the tilt is only a couple of degrees.” A couple of degrees across a long row is several characters of vertical drift, which is exactly enough to leave the end of an account number outside the corner of a box that looked right while you were drawing it.
“I will zoom the browser in so I can place the box accurately.” Worth doing, but it does not change what you are checking. The verification that counts happens on the downloaded file, because that is the thing the recipient opens.
“Pixelate looks heavy enough on a photo.” Block size is derived from the bounding box of your selection, so a small or shallow selection gets small blocks regardless of how strong the result feels. For anything sensitive, use the solid fill and stop estimating.
“Taking a second photo is more work than editing this one.” It is almost never more work, and it is the only step that can fix the frame, the angle and the reflection at the same time. If you can retake it, retake it.