People treat “blur,” “pixelate,” and “black box” as three skins on the same hide. They are not. A blur is a filter: each output pixel is a weighted mix of its neighbors. The original structure is attenuated, not deleted. Pixelation is a resolution attack: the region is reduced to a grid of coarse cells, then those cells are drawn large. A black box is replacement: every pixel in the region becomes one color, usually opaque black. Those three operations leave three different residues in the file you download.
This page is about reversibility of on-image content — the samples in the JPEG or PNG. It is not about EXIF, GPS, or file headers. A perfectly blacked-out Social Security number can still sit next to a geotag in the same file; that is a different layer, covered on the metadata page in this batch. HideShot runs all three modes in this tab so you can apply the cover that matches the risk, then download a PNG with the cover baked into the pixels.
Guide
Redaction of a photograph is a claim about the file you will share: a reader should not be able to recover the hidden region from those bytes. “Looks unreadable on my phone” is not that claim. Phone previews are small, compressed, and viewed at arm’s length. The file you post can be saved, zoomed, contrast-stretched, and run through sharpening or deconvolution. The method you pick decides how much of the original region still exists as a puzzle.
A walkthrough of applying pixelation as a tool lives on pixelate an image online. A walkthrough of opaque cover as a tool lives on blackout an image online. Stay here for the mechanism: why a light Gaussian smear is a different object from a mosaic, and why a solid fill is the conservative choice when the hidden string is an identifier rather than a bystander’s face in a landscape.
What a blur actually does to the samples
A Gaussian blur (and the cheaper box blur many editors use as a stand-in) is convolution. Each pixel in the marked region is replaced by a weighted average of nearby pixels. High-frequency information — sharp edges, the strokes of a digit, the serifs on a name — is attenuated. It is not removed from the neighborhood. The operation is linear and, when the kernel is small and approximately known, it is the kind of thing deconvolution algorithms are built for. Wiener filtering and Richardson–Lucy deconvolution are the textbook names. Consumer tools phrase the same idea as “sharpen,” “clarity,” or “dehaze.” None of those reconstruct a perfect original. All of them can pull a high-contrast string back toward readability if you only used a few pixels of radius on a large, well-lit plate or a screenshot of text.
Risk goes up when three things are true at once: the blur radius is small relative to the feature size, the content is binary-looking (black text on white, a high-contrast license plate), and the saved file is still high resolution. A 4K phone JPEG with a 8-pixel blur over a plate is a different object from a 200-pixel-radius smear over a distant face. Do not use “I blurred it” as a synonym for “the samples are gone.”
JPEG recompression after a blur adds another wrinkle. Block-based compression around a soft region can leave ringing and DCT artifacts that follow the original edges. That is not a full reverse, but it is a hint. If you must blur, use a radius that makes the feature unreadable at 200% zoom on the file you will actually send, then check a contrast-stretched crop of that same file — not only the editor preview.
What pixelation destroys that blur keeps
Pixelation (a mosaic) downsamples the selected region to a coarse grid and then upsamples with nearest-neighbor (or a similarly blocky filter). Each visible square is one value — typically a mean of the original pixels that landed in that cell. The variation inside the cell is discarded. That is a loss of spatial resolution, not a smear of the same resolution. A deconvolution kernel cannot put back samples that were averaged away and then thrown out.
Pixelation is still parameterized. If each block is 2×2 on a sharp screenshot of an account number, each cell still encodes a lot of the original stroke. Research on “mosaic attacks” and super-resolution of very large pixelated faces is about recovering identity from coarse face mosaics, usually when the blocks are large relative to a face but the face still occupies many blocks. That literature is not a reason to panic about a heavy mosaic over a 12-digit number. It is a reason not to use decorative 4-pixel mosaics on high-stakes text. Make the block larger than the feature you are hiding. If a digit still has a unique silhouette across two or three squares, enlarge the cell size or switch to a black box.
Pixelation also looks like redaction. Viewers understand the grid as “this was covered on purpose.” A soft blur can look like atmosphere or a cheap filter, which invites someone to try to “fix” it. That social difference is not a cryptographic one, but it changes whether a stranger bothers to try.
Why a solid black (or opaque) fill is the conservative pixel cover
An opaque fill writes a constant color into every pixel of the region. After that write, those samples contain no function of the original intensities. There is nothing to deconvolve. There is nothing to super-resolve from inside the rectangle. That is why document redaction culture standardized on black bars rather than Gaussian privacy filters: the claim is destruction of the marked area, not attenuation.
Two caveats, both about the boundary, not the interior. If the box is a pixel short, a serif or a plate frame still sits in the file. Anti-aliased edges and JPEG ringing around a hard rectangle can carry a faint ghost of a high-contrast edge. Draw a margin. After export, zoom past 100% and look at the perimeter. If you can still name a character from the halo, enlarge the box and export again.
A black box is ugly. That is the point for sensitive content. For a bystander in a vacation skyline, a heavy pixelate or a strong blur can be the right trade: the scene still reads as a place, and the identity is no longer a portrait. For an SSN, a medical record number, or a password, ugliness is cheaper than a clever reverse. Do not pick blur because it “looks nicer” on a tax form.
Match the method to the risk, then bake it into the download
Casual privacy — a stranger in the background of a landscape, a coworker at the edge of an office snapshot you will send to a friend — can use a strong blur or a coarse pixelate if you verify at 100% zoom that the person is not recognizable. Genuinely sensitive on-image content — government ID numbers, full financial account strings, credentials, anything you would not read aloud in a cafe — should be an opaque fill, oversized, checked on the saved file. Mixed frames (a selfie with a bill on the table) can use both: pixelate the face if you want the photo to remain a photo; black-box the digits.
HideShot’s three modes are those three operations on a canvas in this tab. Choose the image, pick Blur, Pixelate, or Black Box, mark the region, and download a PNG. The PNG is a new raster. The cover is in the pixels of that download, not in a layer you can toggle off. Re-open the download, not the in-tab preview, before you treat the job as done. For a dedicated pixelate session, use pixelate an image online. For a dedicated opaque-cover session, use blackout an image online.
- Decide the risk class before you pick a mode. Identifiers get a box. Atmosphere can get a mosaic or a heavy blur.
- Mark a margin around the feature. Check the export at more than 100% zoom, including the edge of the region.
- If you used blur, try a contrast or sharpen pass on a copy of the export as an adversary would. If the string comes back, the radius was too small — switch mode or enlarge.
- Remember that this page does not strip GPS or camera tags. Pixel cover and metadata are separate files of work.
Mistakes that confuse a filter with a wipe
Using a 4-pixel Instagram-style blur on a sharp screenshot of a card number. That is concealment for a glance, not destruction. Black-box the digits.
Pixelating at a cell size smaller than a character. Increase the block size until the character is not a puzzle of two squares.
Judging the cover on a tiny preview. Open the downloaded PNG at 200%. Previews lie.
Treating a black box as a metadata wipe. The box replaces pixels. Headers are untouched. Strip tags in a metadata tool if the file will travel.
Leaving the unmarked original next to the marked PNG in the same album you will share from. People attach the wrong file. Keep the original out of the share folder.