Pixelated WhatsApp photo: what can really be recovered

Why WhatsApp photos look pixelated: what the app does when you hit send

When you send a picture as a normal photo, WhatsApp does not pass on the file from your camera. It makes a new, much smaller one. Two things happen in a fraction of a second, and both are destructive.

First, it shrinks the image. A modern phone shoots 12 to 50 megapixels, with a long side of 4,000 pixels or more. WhatsApp scales that down so the long side is somewhere around 1,600 pixels in standard quality. That alone throws away most of the pixels: a 4,000 × 3,000 photo becomes roughly 1,600 × 1,200, which is less than a sixth of the original pixel count. Fine detail such as eyelashes, text on a sign or the weave of a jumper simply has nowhere to live any more.

Second, it re-encodes the result as a JPEG at low quality. JPEG splits the image into blocks of 8 × 8 pixels and describes each block with a handful of numbers. At high quality the blocks are invisible. At the quality WhatsApp uses, the fine variations inside each block are rounded off, so you get the typical side effects:

• Blockiness: flat areas such as sky, walls or skin break into visible squares, especially in shadows.
• Ringing or halos: a faint ghost outline around high-contrast edges, like dark hair against a bright window or black text on white.
• Colour bleeding: JPEG stores colour at lower resolution than brightness, so red and blue edges smear into their neighbours.
• Smeared texture: grass, gravel and fabric turn into a soft, painterly mush.

That is why a photo that looked crisp on your friend's phone arrives as a file of 80 to 200 KB that falls apart the moment you zoom in. If the photo was a screenshot, or had already been forwarded a few times, each hop can repeat the process and add another layer of damage.

The key point for everything that follows: the missing detail is not hidden in the file you received. It was discarded on the sender's phone. No app, filter or AI can read back information that is no longer there. What software can do is tidy up the damage and make a convincing, consistent larger image from what is left. That is useful, but it is a different thing from recovering the original.

Send photos on WhatsApp without losing quality: ask for the original first

Before you try to improve a blurry WhatsApp photo, ask the person who took it to send it again as a file. It takes a minute, and nothing you do afterwards will beat having the real thing.

When a photo travels as a document rather than as a photo, WhatsApp treats it like any other attachment and does not resize or recompress it. You receive the exact file from the camera, with its full resolution and its metadata.

On iPhone:

1. Open the chat and tap the + button next to the message box.
2. Choose Document, then Choose Photo or Video.
3. Select the picture and send. It arrives as a file with its name and size shown.

On Android:

1. Open the chat and tap the paperclip icon.
2. Choose Document, then browse to the photo (often under Gallery, DCIM or Camera).
3. Select it and send.

Recent versions of WhatsApp also offer an HD option when you send a photo normally. It produces a larger image than standard quality, but it is still resized and recompressed, so for anything you want to print, archive or edit, the document route is the safe one.

A few practical notes:

• iPhones often save photos as HEIC. Sent as a document, the file may arrive as .heic. If your computer cannot open it, our guide to convert HEIC to JPG covers it, and Airpic's tools accept HEIC directly anyway.
• Originals are large, often several megabytes each. If you need to share them somewhere with a size limit after you have them, compress images with control over quality and target size, rather than letting a messaging app decide for you.
• For a whole batch, such as wedding or holiday photos, a shared cloud folder is usually easier than sending dozens of documents one by one.

If the original no longer exists, because the phone was lost, the sender deleted it, or the only copy is the one in the family group, then you are working with what you have. The rest of this article is about getting the most out of it.

Can you recover the quality of a WhatsApp photo with AI? What it can and cannot do

Modern image models are trained on millions of pairs of clean and degraded pictures. They have learned what JPEG blocks, halos and soft edges look like, and what the underlying image most probably looked like before the damage. That gives them three abilities that are genuinely useful on a pixelated WhatsApp photo.

• Removing compression artefacts. The 8 × 8 grid, the mosquito noise around edges and the colour smears are regular, recognisable patterns. A model can identify them and replace them with smooth, plausible surfaces.
• Rebuilding edges. Where a contour has turned into a staircase or a fuzzy band, the model draws a clean, continuous edge in the place where the evidence says it was.
• Multiplying pixels coherently. Instead of stretching each pixel into a bigger blurry square, an AI upscaler generates new pixels that fit the edges and surfaces around them. A 1,600-pixel photo can become 3,200 or 6,400 pixels wide and still look sharp at normal viewing distance.

What it cannot do is bring back texture that was erased. If the compression flattened the bark of a tree into a brown smudge, the model can add a texture that looks like bark, but not the bark that was actually there. If a number plate or a line of small print turned into a few grey blocks, no tool will make it legible again: anything it shows you would be a guess dressed up as a fact.

The same is true of faces, and it matters more there. A face that is 60 pixels wide in the received file holds very little information about the person's real features. A face model will happily draw sharp eyes, teeth and skin, but if you push it too hard it draws a face, not their face. Section four explains how to keep it on a short leash.

A sensible expectation, then: the result should look like a clean, larger photo of the same scene, good enough for a frame on the shelf, a screen, a slideshow or a modest print. It will not match a photo that came straight from the camera, and it should not be used to read details that are not visible in the received file.

How to improve a blurry WhatsApp photo: the order that works

The order of the steps matters more than most people expect. Upscaling enlarges everything in the image, including the JPEG blocks and the noise. If you enlarge first, you hand the next step bigger, more convincing artefacts to deal with. So the rule is: clean first, enlarge second.

1. Judge the damage at 100 % zoom. Look at a flat area (a wall, the sky, a cheek) and at a strong edge. Visible squares and grain in the flat areas mean the image needs cleaning. A uniform softness across the whole picture, with no squares, is more likely camera shake or missed focus.
2. Remove the noise and compression artefacts. For grain, low-light noise and blocky shadows, denoise a photo with moderate strength. Too much smooths skin and fabric into plastic; a middle setting usually removes the blocks and keeps some natural texture. If the main problem is softness from movement or focus rather than blocks, try unblur a photo instead.
3. Upscale. Once the image is clean, enlarge it. ×2 is enough for most screens and small prints; ×4 when you want to crop or print larger. Upscaling a WhatsApp photo to poster size will show its limits, so match the factor to how you will actually use it. Our guide to upscaling an image for print explains how many pixels each print size really needs.
4. Treat faces separately, and gently. If there are people in the picture, turn on face restoration. Then lower the strength (in Airpic this slider is the fidelity of the faces). At full strength the model rebuilds the face almost from scratch and can turn a heavily compressed face into a generic, too-perfect one that looks like somebody else. Somewhere around the middle of the slider usually keeps the person recognisable while removing the mush. Compare the result with the original side by side and trust your eyes, especially if you know the person.
5. Stop before it looks processed. Running the same tool twice on its own output rarely helps: it tends to add artificial crispness rather than real detail.

One extra trick for faces: if you have other, better photos of the same person, use them. Airpic's old photo restorer accepts up to three reference photos to guide the resemblance, which helps a lot when the WhatsApp version is tiny or badly compressed.

Step by step in Airpic: upscale the image and, if there are faces, restore them

Airpic runs in the browser with no account and no subscription. Processing and viewing the result is free; the full download of the first image each month is free too. Files are deleted one hour after download, one day at most, are never used for training, and the GPUs that process them are in the European Union.

Before you start: save the photo from WhatsApp to your phone or computer. On a desktop, WhatsApp Web lets you download it from the chat; on a phone, it is usually already in your gallery. JPG, PNG, WebP, TIFF, HEIC and AVIF are all accepted.

Photos without faces (landscapes, objects, screenshots of documents):

1. If the image shows clear blocks or grain, run it through the denoiser first and download the cleaned version.
2. Open the upscaler and drop the photo in, or paste it with Ctrl+V.
3. Choose ×2 or ×4. For large prints, the print mode works out the size you need.
4. Press Process and drag the divider on the comparison to check edges and flat areas. Use «View at 100 %» to see real pixels, not a scaled preview.
5. Download the result.

Photos with people:

1. For a normal recent photo where faces are reasonably large, use the upscaler with the «enhance faces» option switched on.
2. If the faces are very small, heavily blocked, or the photo is old or damaged, use the old photo restorer instead (images up to 20 megapixels). It rebuilds faces, enlarges the background and can recolour if needed.
3. Lower the strength slider before processing, then raise or lower it and compare until the person looks like themselves.
4. If you have other photos of the same person, add up to three as references to guide the likeness.
5. Check the eyes, mouth and teeth at full size, then download.

For everything else that tends to go with a forwarded photo, the same site can remove an object from a photo, such as a date stamp or a stray elbow, or turn a black-and-white family picture into colour. But for the pixelated photo in your chat, start here: upscale an image with AI, check it at 100 %, and keep the result only if it still looks like the moment you remember.

FAQ

Keep reading