JPEG quality controls an encoder's tradeoff between file size and changes to the image. It is not the percentage of original detail retained. Quality 80 can be a useful starting point for a photograph, but it is not a guarantee of invisible loss, and quality 100 is not a promise of an unchanged original.

Compare your actual exported files at their intended viewing size and at 1:1. Check bytes alongside texture, lettering and gradients. A single quality number cannot replace those checks.

What the slider changes

Common photographic JPEG encoding represents image detail in blocks and uses quantization to store some information less precisely. Coarser choices can use fewer bytes while changing more of the decoded result.

The quality control guides the encoder, but its mapping is not standardized as one universal 0-to-100 visual scale. It is inaccurate to say every application uses one exact quantization table at 50 or halves every divisor at 75.

Browser canvas export accepts a quality parameter for formats that support variable-quality encoding. The HTML specification defines that interface; it does not promise identical output bytes from every browser.

Why 100 is not an archival format

A high setting can preserve more detail than a low setting in the same workflow. It does not undo changes made by a crop, resize, previous JPEG export or source conversion.

If you need an editable master, keep the original photograph or editing project. If you need to preserve a rendered raster without another lossy JPEG step, use a suitable lossless format such as PNG and understand that it preserves the pixels you already have, not missing earlier detail.

“JPEG” also names a family of standards and workflows. The claim here concerns the ordinary browser photo export used by this tool, not every possible lossless operation or standard carrying the JPEG name. JPEG's official overview is the primary format reference.

Compare a few settings on the same source

Use one difficult source with fine texture, smooth gradients and high-contrast edges. Export fresh copies from that same source rather than re-saving each output into the next one.

  1. Set the intended dimensions first.
  2. Export JPEG copies at quality settings such as 60, 80 and 100.
  3. Record each actual file size and dimensions.
  4. Compare at the intended display size, then inspect difficult areas at 1:1.
  5. Choose the smallest result that remains suitable for your use.
  6. Repeat with another source type before applying the setting to a large collection.

This is a comparison method, not a claim that those three settings always produce a fixed saving or one quality threshold. A photo of foliage and a flat diagram can respond very differently.

A reproducible comparison

On 20 October 2026, we exported the site's synthetic 1200×1600 test image through the actual browser tool in Chromium 151.0.7922.34. Each fresh export used Custom mode, Keep original pixels, JPEG, sharpening off and no byte cap. The only changed setting was JPEG quality: 35, 65 or 90.

FileExact bytesDimensionsInspect the full file
Synthetic source PNG2,301,3621200×1600Download source PNG
JPEG quality 35107,1491200×1600Download quality 35
JPEG quality 65164,4641200×1600Download quality 65
JPEG quality 90373,9091200×1600Download quality 90
The same synthetic-image crop in the original PNG and JPEG exports at quality 35, 65 and 90; lettering, thin lines and texture show different artifacts.
Each panel uses the same 200×112-pixel crop without resizing. The 117,052-byte figure is assembled as lossless WebP so it adds no new lossy compression. Open the figure at full size for 1:1 pixels; a narrow page may reduce its display size. Full images above download only when chosen.

The source is a synthetic fixture with text, gradients, fine lines and texture, not a camera photograph. Compare the colored lines and the region around the white letters. The lower-quality export shows more visible changes in this crop, while the higher setting uses more bytes.

The script captures actual JPEG downloads, checks their dimensions and SHA-256 against the previewed outputs, and verifies that the figure retains the decoded crop pixels. The manifest records file hashes, browser version, crop coordinates and processing settings. The original PNG is linked rather than automatically loaded.

These measurements apply to this one synthetic source and browser. They do not predict average savings for photographs, establish a universal quality threshold, or measure Instagram or another platform. Different browser encoders or source images can produce different results.

What artifacts to inspect

Look at smooth skies or walls for blotchy transitions and banding. Look around lettering, branches and other sharp edges for ringing. Inspect noisy shadows and fine fabric for lost texture.

The point where these become unacceptable depends on the image and the display. A small avatar can tolerate a different compromise from a product detail view or a document with fine text.

Some JPEG encoders also reduce color detail through chroma subsampling. Small colored lettering can therefore need special care. This tool does not expose a universal subsampling control; do not infer it from the quality number alone.

Repeated exports and metadata

Decoding and lossy re-encoding can introduce additional changes, particularly after a resize, crop or filter. Some operations can instead copy or transform JPEG data without a new lossy encode. Avoid the blanket claim that every file save always loses a fixed amount.

The practical rule is to keep the original and create each delivery copy from it. Increasing the quality of a damaged copy does not restore detail discarded earlier.

Original EXIF can contain camera or location information. This tool creates a new raster export rather than copying the source EXIF. Do not describe all metadata as useless: rights information, color profiles and provenance can be valuable. Keep the original when you need them.

JPEG, WebP and PNG choices

FormatUseful starting useImportant limit
JPEGOpaque photographic delivery copyLossy export; no transparent pixels
WebPWeb delivery when the destination accepts itThis tool's export is opaque; size and appearance depend on the encoder
PNGTransparency, diagrams or lossless storage of rendered pixelsPhotos can be large; this tool has no PNG quality slider

Google's WebP documentation and the PNG specification describe their capabilities. A published format comparison does not guarantee the same percentage saving for your files.

JPEG and WebP output here place transparent areas on white. Only PNG export in this tool preserves source transparency. WebP as a format can support alpha, but that capability is different from this tool's current opaque export. Check the downloaded file in the receiving application.

Automatic cap or manual quality

Use the JPEG compressor when a suitable delivery copy is the goal. Use a KB mode when an accepted byte cap is the requirement. If dimensions can be reduced to meet that cap, inspect the resulting dimensions too.

For manual comparison, choose Custom, the desired dimensions and JPEG output, then adjust quality. To isolate quality alone, keep crop, dimensions and sharpening the same between exports.

Try it on your own photoSocial-ready photos, smaller files, nothing uploaded
Compress a photo

The tool works on your device. Its output size and appearance depend on the source and browser. No local quality slider guarantees how Instagram, WhatsApp or another service later processes the file.

Limits and checklist

If a smaller-looking export uses more bytes, the larger-file troubleshooting guide separates dimensions, format and quality. If a logo gains a white background, check transparency and output format before adjusting quality again.

  • Keep the original or editable master.
  • Set dimensions for the intended use.
  • Compare actual bytes and difficult image regions.
  • Judge small-size appearance as well as 1:1 detail.
  • Avoid fixed savings or universal quality thresholds.
  • Check the format and transparency behavior of the download.

Editorial review: 20 October 2026. This revision removes unsupported savings, universal quantization mappings and claims that high-quality differences are always invisible.