JPEG artifacts: what they are, why they happen and how to fix them

JPEG makes images small by throwing away information people are unlikely to notice. It does this in three steps.

First, it usually stores colour at a lower resolution than brightness, because our eyes are less sensitive to fine colour detail. This is called chroma subsampling. Second, it splits the image into blocks of 8 × 8 pixels and describes each block as a mix of simple wave patterns, from smooth to fine. Third, it rounds those descriptions off. Smooth patterns are kept fairly accurately; fine ones are mostly discarded.

The quality setting controls that rounding. At high quality, little is lost and the result is hard to tell from the original. At low quality, each 8 × 8 block is reduced to a few broad patterns. Every JPEG artifact comes from this process, and the discarded information is not stored anywhere in the file.

Blocking is the easiest to recognise. Each 8 × 8 block is compressed on its own, so neighbouring blocks stop matching at their borders and a faint grid of squares appears. You see it most in smooth areas such as sky, walls and skin, and in dark shadows.

Ringing, also called mosquito noise, appears around sharp edges. A hard edge needs fine wave patterns, and when those are discarded the edge is drawn with ripples that spill into the flat area beside it. The result is a halo of specks around text, logos, hair against sky and building outlines. Screenshots and graphics suffer most, because they are full of hard edges.

Banding turns smooth gradients into visible steps. A sky that fades from light to dark becomes a series of flat stripes. Heavy editing of an 8-bit image can cause banding too, but strong JPEG compression makes it much more likely.

Colour bleeding comes from chroma subsampling and from compressing colour harder than brightness. Strong colours spill past the edges of objects, red and blue details look blurry while their outlines stay sharp, and small coloured text gets hard to read.

Platforms that handle huge numbers of photos need them small. Small files cost less to store, load faster and use less mobile data. So most platforms resize uploads to the largest size they display and save them again as JPEG, at a quality they choose. Instagram's help centre, for example, says photos are kept at a width of up to 1080 pixels. WhatsApp's help centre says media is usually compressed when shared, so quality can be lower than the original.

One round of compression at a reasonable quality is usually fine. The problem is generation loss. Every time a JPEG is opened and saved as JPEG again, the rounding happens again on top of the old damage. A photo sent on WhatsApp, saved, posted to Instagram, screenshotted and shared again may have been compressed four or five times.

Editing adds rounds too. Opening a JPEG, adjusting brightness and saving it as JPEG is one more. Doing that repeatedly in different apps is one of the most common ways a good photo turns blotchy.

A common tip is to convert a JPEG to PNG to improve quality. PNG is lossless: it stores every pixel exactly. That prevents new damage, but it also stores the existing blocks, halos and smears exactly. You get the same damaged image in a much larger file.

PNG is the right format for saving an image after you repair it, and for screenshots and graphics from the start. It cannot undo damage that is already in the pixels.

Traditional deblocking filters smooth the borders between blocks and blur the ringing away. They help with mild damage but soften the whole image, because a simple filter cannot tell a block edge from a real edge.

AI artifact removal is trained on many pairs of compressed and clean images. It learns what the damage looks like (the 8 × 8 grid, the ripples next to edges, the stepped gradients, the shifted colour) and removes those patterns while keeping real edges.

Which tool to use depends on what the image is:

• Screenshots, graphics, logos and anything with text: Enhance removes the compression at the same size, and text it detects is protected rather than redrawn. It was built for photographs, so check flat areas and every important word against the original, and use a fresh PNG copy when you can get one.
• Phone photos that lost quality in a chat app, often with night grain on top: use Enhance, which removes compression and noise together. Our page on fixing WhatsApp photo quality walks through it.
• An image that is also too small: remove the artifacts first, then enlarge. An upscaler enlarges blocks and halos along with everything else, so an 8-pixel block becomes a 32-pixel block after a 4× upscale.

Find the least compressed copy you have: the original upload, the first copy in the chat, not a forward or a screenshot. Run the right tool from the list above and compare the result with the original at 100 % zoom, not fitted to the screen.

Save the repaired version as PNG, or as a JPEG at a high quality setting. Otherwise the next save adds a fresh round of artifacts on top of your clean result.

Artifact removal takes away the damage. It cannot always bring back what compression discarded. In a heavily compressed image, fine texture such as hair, grass or fabric weave may be reduced to flat patches. A tool that adds no texture will clean those patches and keep the remaining edges, but they stay smooth.

Some tools, including our Enhance model, add natural-looking fine texture where a photo lost it. That makes a photo look less flat, but the texture is plausible, not the original. It is fine for a holiday picture, not for evidence.

Small text is the clearest limit. If compression turned letters into smudges, removing the artifacts makes the smudges cleaner, not readable. None of our tools guess letters.

Most JPEG damage is avoidable with a few habits:

• Keep the original file from your camera or phone, and share from it rather than from saved copies.
• In WhatsApp, choose HD quality when you send, or attach the photo as a document to send the original file.
• Edit in an app that keeps your edits separate from the original, and export to JPEG once, at the end.
• Export JPEGs at a high quality setting. The file is larger, but the difference in artifacts is large too.
• Save screenshots, graphics and anything with text as PNG.
• Upload to social platforms at the size they display, so they have less resizing to do.

You get 3 free credits a day, with no sign-up and no card. One photo with one tool uses 1 credit; Upscale 8× uses 2. Free uploads can be JPG, PNG or WebP up to 3 MB and 8 MP, and free results carry a watermark. Pro ($9.99 a month, 200 credits) and Pro+ ($19.99 a month, 500 credits) remove the watermark and accept uploads up to 20 MB and 30 MP. Uploads and results are deleted within 24 hours and are never used to train AI.