Blur Image, Face or Background — 100% Private & Free

Image blur is the controlled distortion of pixels in part or all of a photo, used to obscure faces, license plates, sensitive text, or to soften a background and lift the subject forward. The technique dates back to 19th-century darkroom defocus tricks; today it's the single most common privacy edit on the internet, applied to roughly 1 in 7 photos shared on news sites and social platforms.

There are three families of blur used by this tool. Gaussian spreads each pixel into its neighbors with a bell-curve weight — the most natural-looking and most reversible. Motion blur smears pixels along a direction to fake speed. Pixelate replaces patches with solid color blocks — the most aggressive and most irreversible.

This page covers all three, plus four ways to apply them: to the whole image, to a rectangular region you drag, to every face automatically detected, or to the background while keeping the subject sharp using on-device AI. Nothing leaves your browser.

The two reasons people blur are completely different. Privacy is one: hiding a face, a license plate, an address bar in a screenshot, a chat name. The other is composition: blurring a noisy background so a portrait subject reads instantly.

Both reasons have grown sharply in importance. A 2024 Pew study found that roughly 56% of US adults have shared a photo online they later regretted, and the most common regret was that the photo "revealed something I didn't intend." Under GDPR and CCPA, photos containing identifiable people processed by businesses count as personal data, and incidental bystanders in business-shared photos have been the basis for several five-figure fines in the EU since 2022.

The aesthetic reason is just as concrete. Phone camera sensors can't physically replicate the shallow depth-of-field of a DSLR. Software background blur fills that gap. iPhone Portrait Mode does it with a depth sensor; this tool does it with on-device machine learning.

Face blur is the most common privacy edit on the internet, and the easiest to get wrong. The model used in this tool runs entirely in your browser and finds faces in roughly 200–800 ms depending on photo size. Each detected face appears with a small numbered handle on the canvas; click to keep, drop, or resize.

If a face is missed (an unusual angle, very small, partially hidden), switch to Region mode and drag a rectangle around it manually. Region and Auto Face additions are processed in the same pass.

For privacy, always use pixelate, not Gaussian blur, when hiding identity. Pixelate at 8–16 px block size is functionally irreversible. Gaussian blur up to about 10 px radius can sometimes be partially reversed by AI deconvolution, which is fine for casual aesthetic softening but not for hiding someone's identity. The default block size for Auto Face + Pixelate is 12 px, which works well on most phone-resolution faces.

Background blur in this tool uses MediaPipe Selfie Segmentation, Google's open-source on-device model for separating people from backgrounds. The first time you switch to AI Background mode, your browser downloads the model — about 6 MB. After that, the model is cached in IndexedDB and loads instantly.

Why the model loads at all: every other "AI background blur" tool uploads your photo to a server. That's 1–5 MB per photo, every time, plus your image sits on someone else's hard drive. The 6 MB one-time download is a one-and-done trade for never uploading another photo to anyone's server, ever. A 6 MB download takes about 3 seconds on a typical home broadband connection — roughly the same time as uploading one 4 MB photo to a cloud editor.

The model identifies subject and background in roughly 300–600 ms on a modern laptop. The subject stays at full sharpness; the background gets whatever Gaussian or Pixelate intensity you set. Edge accuracy on fine details like flyaway hair is the main limitation — true optical bokeh from a DSLR or iPhone Portrait Mode (which has a hardware depth sensor) is still better. For social media and casual sharing, the difference is invisible.

Three under-served use cases the tool handles directly:

License plates. Switch to Region mode, drag a rectangle around the plate, choose Pixelate at 8–16 px block size. EXIF metadata is automatically removed from the output, so blurring a license plate also strips the GPS coordinates that were embedded in the original photo. (This last part matters: many privacy-blur tools leave the GPS data intact, defeating the point.)

Signatures and addresses in screenshots. Region + Pixelate works the same way. For contract proof-of-receipt screenshots, signed documents, or address-of-record forms shared for verification, pixelate the signature line and the address field before sharing.

Chat screenshots and DMs. Region mode supports multiple rectangles in one pass. Box every visible username, profile picture, and timestamp, then bulk-pixelate. Pair with JPG to PDF to bundle blurred screenshots into a single shareable file.

Whole Image, Auto Face, and AI Background modes all support unlimited batch processing. Drop a folder, choose your blur type and intensity on the first photo, and the tool applies the same settings to every image. Mixed orientations are handled automatically; the AI background model runs per-image but the model itself only loads once per session.

Region mode is the exception — because every photo has a different composition, the rectangles you draw on one image can't be meaningfully applied to another. The tool shows a banner suggesting you switch to Auto Face for batch privacy work.

Competitor batch limits for reference: Lunapic processes one image per session (no batch); Pinetools has a small batch page with no advertised cap; BeFunky hides batch behind a Plus paid plan. This tool has no batch cap and no daily quota.

Sometimes yes, mostly no. The honest breakdown:

Light Gaussian blur (≤10 px radius) can sometimes be partially reversed by AI deconvolution tools. The reconstruction isn't perfect — typically enough to identify a face but not enough to read fine text. For aesthetic softening, this doesn't matter. For privacy, it does.

Heavy Gaussian blur (≥30 px radius) is functionally one-way; the deconvolution gets a vague blob with no usable detail.

Pixelate at 8–16 px block size is mathematically irreversible. The original pixel values are averaged into a single block — there's no information left to reconstruct from.

Pixelate at very small block sizes (2–4 px) is actually weaker than Gaussian for privacy. The blocks are small enough that the visual signal of a face is preserved. Always use ≥8 px blocks for genuine censoring.

The most famous reversibility incident is the 2018 demonstration that pixelated PDF redactions could be partially reversed when the pixelation was applied at low resolution to text. The lesson: pixelate harder than you think you need to.

A blurred image is rarely the final step. The most common follow-ups:

Add a watermark. Blur hides identity; a watermark claims ownership. Add a watermark after blurring for a complete privacy + branding pass.

Strip EXIF metadata. Blurring a face doesn't remove the GPS coordinates of where the photo was taken (EXIF is on by default in this tool, but if you turned it off, do it now). Remove EXIF as a final privacy pass.

Compress for the web. Use JPEG compression or target a size with compress to 500 KB.

Resize for social media. Resize with one-click platform presets.