Each row shows the original size, the new size and the saving.
Shrink one photo before sending it, or drop a folder of product shots and watch each file get lighter. Both jobs live on the same page: single mode gives you a compare slider, batch mode builds a table with the old size, the new size and the saving for every image.
Quality, maximum width and output format can be set per file, and anything that has no headroom left is flagged instead of being quietly inflated. Originals on your disk are never touched. All processing runs locally in your browser. Your data is not uploaded to any server. 100% Local
JPG and WebP divide the image into blocks, transform each block into frequency coefficients and then round those coefficients more coarsely as quality falls. At 75 to 85 most photographs look identical at normal size while losing a large share of their bytes. Below 60 blocking and colour bleeding become visible around edges, and repeated saves stack the damage because each one re-quantizes an already degraded image. The slider never changes the pixel grid, only how precisely those pixels are stored.
PNG uses lossless DEFLATE, so the encoder cannot discard detail the way JPG can. A screenshot exported by a design tool is usually optimised already, and re-encoding may add a few bytes for a fresh header. Real reductions come from changing the pipeline instead: convert to WebP, flatten a mostly photographic PNG into JPG, or reduce the dimensions. Keep PNG for logos, interface captures and anything that needs an alpha channel.
A catalogue of two hundred shots at four megabytes each will slow every collection page. Batch mode with a 1600 pixel width cap and quality 80 typically brings each photo under three hundred kilobytes while keeping zoomed detail acceptable. Rename nothing before exporting; the rows keep the original file name and append a suffix, so the mapping back to your spreadsheet stays clear.
Mail servers reject or delay messages with oversized attachments, and readers on mobile data abandon heavy newsletters. Compressing hero images to JPG at quality 75 and capping width at 1200 pixels keeps a full issue in a few hundred kilobytes. Keep the logo as PNG if it needs transparency, and remember that some clients still ignore WebP.
Large images are usually the heaviest part of a page and the main cause of a poor Largest Contentful Paint score. Export at the widest size the layout actually uses, compress to WebP where supported, and keep a JPG fallback. A batch run over the whole media library makes the improvement measurable instead of fixing one hero image at a time.
Only when you choose a custom maximum width. With the default keep-original setting the canvas keeps every pixel and the saving comes from the encoder alone. Setting a width of 1600 pixels does both jobs at once, which is usually the cheapest way to prepare photos for a website.
PNG is lossless, so the encoder cannot throw away detail the way JPG can. Screenshots that were already optimised often gain a few bytes because the browser writes a fresh header and colour profile. Convert those files to WebP or JPG, or resize them, instead of compressing the PNG again.
Yes. Every image is redrawn on a canvas and re-encoded, so EXIF fields such as camera model, GPS position and capture time do not survive. Orientation is applied to the pixels before that step, so rotated phone photos still appear upright. Keep a copy of the original if the metadata matters.
No. Encoding runs inside a Web Worker so the interface keeps scrolling while images are processed. One or two files are handled at a time to keep memory in check, and the queue reports progress. Very large panoramas can still take a few seconds each on a laptop.
For photographs, 75 to 85 removes most invisible weight and is hard to distinguish from the original at normal viewing size. Use 90 or higher for artwork with fine gradients, and lower values only when a thumbnail is the final destination. Check each important result on screen before publishing.