Compress JPG, PNG and WebP images in your browser, with real before-and-after byte sizes, a size-limit search and a check of the metadata each file loses. Up to 10 images at once; nothing is uploaded.

RT-IMG-002 · Image & File
Your browser is too old for this tool’s interactive features. Please update to Safari 13.1, Chrome or Edge 80, Firefox 74 or newer.

Image Compressor Tool

Global Quality Quality: 80 KB (1 KB = 1,024 bytes; JPG and WebP only)

Drop images here or click to select

JPG, PNG, WebP (HEIC where the browser can open it) · Max 10 MB per file · Up to 10 files

🔒 Your images never leave your browser: each one is decoded, redrawn on a canvas and re-encoded on your device. Re-encoding writes a new file from the pixels, so camera data (EXIF), including any GPS location, is not carried over; each result says what its original held and checks the file it wrote. A camera rotation tag is applied to the pixels first, so photos stay upright. Sizes are real before-and-after bytes, never an estimate.

📅 Research current as of 05 Nov 2026 · Sources: Canvas re-encode per the WHATWG HTML Standard (toBlob: unsupported types become PNG; no alpha, composited onto black); EXIF orientation via createImageBitmap imageOrientation (MDN).
Standards, published formulas, and reference data are revised over time — check against a current authoritative source before relying on these figures.
Advertisement
After results · AD-W1 Responsive · Post-tool — peak engagement

How to Use the Image Compressor

Drop up to 10 images onto the drop zone or click to select them

Drag and drop up to 10 JPG, PNG or WebP images onto the drop zone, or click it to open your file picker. Each file must be under 10 MB. A row appears for each file with its name, size and pixel dimensions.

Set a quality, or a size limit

Move the Global Quality slider to apply one quality to every file, or use each file's own slider. Every change re-encodes the file, so the size shown is the real size of the file you will download. If a form caps the upload size, type that limit in kilobytes into Size limit instead: the tool searches for the highest quality that comes in under it and says which quality it chose.

Choose an output format per file or keep the original format

The format selector defaults to "Same", keeping each file in its original format. JPG and WebP are lossy and take a quality; PNG is lossless, so its quality slider is switched off. Transparent areas are filled white in a JPG, which has no transparency. If this browser cannot write the format you picked, the row says so and names the file by what was actually written.

Download one file, or all of them as a ZIP

Use a row's Download button to save that image. If re-encoding made a file larger than the original, the row says so and the button reads "Download anyway"; keeping the original is usually the better choice. Download all puts every file in one ZIP, because browsers block a run of separate downloads.

Advertisement
After how-to · AD-W2 Responsive

What re-encoding an image does, and what it costs

Why the size of an image file matters

The bytes in an image are bytes a reader downloads before they see it. On the median mobile home page in HTTP Archive's July 2025 crawl, images came to 911 KB, more than the 632 KB of JavaScript on the same page (Web Almanac 2025, Page Weight). A smaller file is the one lever this tool pulls: it never changes the pixel dimensions, only how the same pixels are encoded. To make an image smaller in pixels too, use the Image Resizer first.

Lossy vs Lossless: When You Can Tell the Difference (and When You Can't)

Image compression falls into two fundamental categories with very different use cases. Lossy compression — used by JPEG and WebP lossy mode — permanently removes visual data that the human visual system is less sensitive to. JPEG uses Discrete Cosine Transform (DCT) to decompose the image into frequency components, then quantises (rounds) the high-frequency components where the eye is least sensitive. The result is a smaller file at the cost of some irreversible quality reduction.

No single quality number is right for every picture, and the same number does not mean the same thing in every encoder: the quality you pass here goes to your browser's own JPEG or WebP encoder as a value from 0 to 1, and another program's "80" uses different tables. Detail-heavy photos hide artefacts; flat colour, text and sharp edges show them first. So judge by eye: open Compare on a file and drag the divider before you download.

Lossless compression preserves every pixel exactly. PNG does it with DEFLATE, the method ZIP files use, so nothing visual is ever discarded; that makes it the right choice for logos, screenshots, diagrams and anything with text. It also means a quality setting has nothing to act on, so the slider is switched off for PNG output. Re-saving a photo as PNG usually makes it larger, and the row will say so in bytes.

Understanding the compression mechanism also settles a question every workflow runs into eventually: how much does re-saving an already-compressed image actually cost? Less than the folklore suggests — provided the quality does not move. The quality setting is the table of divisors the quantiser rounds against, so re-encoding a file at the quality it already carries puts its coefficients back onto a grid they are already sitting on. The rounding step has almost nothing left to round, and each successive pass changes the image less than the one before it. Re-encode at a different quality and every coefficient is snapped onto a grid that does not line up with the one it is on — and that, not the number of passes, is where generation loss actually comes from. Measured on a 940 × 627 photograph, ten consecutive quality-80 passes drifted less than a third as far as one single pass at quality 94, so rounding the quality up "to be safe" costs more than the re-save it was meant to guard against.

The professional workflow advice is unchanged, and better founded for it: retain the original master file and produce every compressed derivative from it. The reason is not that re-saving is inherently corrosive — it is that working from the master is what keeps the quality of each output under your control, instead of inheriting whatever the previous tool in the chain happened to choose.

Correction, 8 August 2026. Until this date this section said: "Each re-encoding pass introduces a new round of quantisation errors that compound the previous pass." The conclusion it drew was sound, but the mechanism was not: at a fixed quality the per-pass error decays toward zero rather than accumulating, and ten correct re-saves are cheaper than one save at the wrong quality. The wording above has been corrected. The full measurement is in our guide, The Only JPEG Quality That Is Nearly Free Is the One It Already Has.

What a canvas re-encode throws away

Drawing a picture onto a canvas and encoding it again writes a new file from the pixels alone. Camera data (EXIF), including a GPS location if the phone recorded one, does not come with it; nor does XMP, the block that often holds a copyright notice. That is a privacy gain when you share a photo, and a loss if you needed the metadata. This page does not assume it: for every file it reads the original's metadata blocks and then the file your browser wrote, and reports what it found in each.

One piece of metadata changes how a photo looks: the rotation tag a phone writes when you hold it sideways. Throw it away without turning the pixels and the picture comes out on its side. This tool decodes with the browser asked to apply that tag (createImageBitmap's imageOrientation: "from-image"), checks for the older behaviour MDN warns about, where drawing ignores the tag, and turns the image itself if it was ignored.

Frequently Asked Questions

  • No. Each image is read, decoded, drawn on a canvas and re-encoded in your browser; this page sends no image data to RECATOOLS or anyone else, and once it has loaded it works offline. The page itself still loads the site's usual scripts, such as consent and, if you agree to them, analytics and ads; none of them is handed your files.
  • There is no number that suits every picture, and another program's "80" is not this browser's "80". Start at the default and open Compare: drag the divider over text, edges and smooth sky, where artefacts appear first, and lower the quality until you can see a difference, then step back up. If the real constraint is a file-size limit on a form, type it into Size limit and let the tool find the highest quality that fits.
  • JPG (JPEG): lossy, built on the discrete cosine transform (ITU-T T.81). Good for photographs; no transparency.
    PNG: lossless, compressed with DEFLATE. Good for logos, screenshots and text; keeps transparency; usually large for photos.
    WebP: lossy or lossless, with transparency. Google reports its lossy files 25-34% smaller than comparable JPEGs at equal SSIM quality. Every current major browser displays WebP, but not every one can write it from a canvas: where yours cannot, the WebP option is switched off.
  • Not if you keep PNG or WebP as the output format: both keep transparency. JPEG has no alpha channel, so when you choose JPG this tool fills transparent areas with white before encoding. (Left to itself, the canvas encoder composites them onto black, as the HTML standard specifies.)
  • Yes, up to 10 at a time, each with its own quality and format, and a global slider that sets them all. Download all saves them in one ZIP file, since browsers block a run of separate downloads; a name that would clash, such as photo.png and photo.jpg both becoming photo-compressed.jpg, gets a number added.
  • In practice, yes: a canvas re-encode writes a new file from the pixels, so camera data, including any GPS position, is not carried over. This page does not take that on trust. Each row reports what metadata the original file held and what it found in the file your browser wrote. The rotation a phone records is applied to the pixels first, so the photo stays upright. If you need to keep EXIF, use an editor that copies metadata across rather than a canvas.
  • It depends on the picture, so try it: switch a row between JPG and WebP and read the real byte counts. For photos, lossy WebP is usually the smallest of the three this page writes (Google's own comparison puts it 25-34% below JPEG at equal SSIM). PNG is usually the largest for photos. AVIF is not offered: when we asked Chrome's canvas for AVIF it wrote a PNG instead.

Method & sources

How it computes

Decodes in the browser (createImageBitmap, imageOrientation "from-image"), redraws at the same pixel size and re-encodes with canvas.toBlob(); quality 1-100 → 0.01-1.0 for JPEG/WebP only. Files are named by the type toBlob returned. JPEG output is filled white first. Sizes are the real bytes written; a size limit is met by binary search on quality. EXIF/GPS/XMP/ICC are read from the original and the output (ImageCompressKernel).

What this tool implements

  • Re-encoding follows the WHATWG HTML Standard canvas serialisation: an unsupported type is written as PNG, the quality argument applies only to lossy formats, and a format without alpha is composited onto opaque black — so the page fills JPEG output white and names every file by the blob type actually returned.
  • EXIF orientation: decoded with createImageBitmap(file, { imageOrientation: "from-image" }) (MDN). Where a decoder ignores an axis-swapping Orientation (5-8, CIPA DC-008 tag 0x0112) — the older drawImage behaviour MDN describes — the page detects it from the JPEG frame size (ITU-T T.81 SOF) and rotates the pixels itself.
  • Metadata is reported from the bytes, not assumed: ImageFormatKernel.sniff() reads EXIF (and its GPS pointer), XMP and ICC from the original and from the file the browser wrote, and the row says what each holds.
  • Sizes are exact bytes (1 KB = 1,024 bytes). A result larger than the original is labelled LARGER with the byte difference and the advice to keep the original; no saving ratio is ever promised.
  • Size limit: the highest quality whose measured output fits, found by binary search (ImageResizeKernel); encoded size is not strictly monotonic in quality, so only measured qualities are reported.
  • Corrected 2026-11-05: transparent-to-JPG came out black; WebP requests in a browser without a WebP encoder saved PNG bytes as .webp; PNG quality slider active with no effect; unsourced platform limits (Shopee, Lazada, Carousell, Singpass/MyInfo), a misattributed "Google research: 100 ms = 1% conversions" pull-quote, Instagram and smartphone file-size figures, and quality thresholds removed; WebP "25-35% smaller" corrected to Google's 25-34% at equal SSIM; Web Almanac image weight restated as the median mobile HOME page.

Sources

Advertisement
Pre-footer · AD-W3 728 × 90