Image Compressor

Real, and labeled per file. PNGs go through OxiPNG, a dedicated lossless optimizer compiled to WebAssembly, so the pixels stay identical while the encoding gets tighter. JPEG and WebP use the browser encoder at your chosen quality, labeled as such because bytes can vary slightly between browsers. Every result card names the codec that produced it, and if a re-encode does not beat your original, the original is kept byte-for-byte and the card says so.

Tick "Target file size" and enter the limit. The tool runs a bounded quality search (at most 7 encodes) on JPEG or WebP output and reports the result honestly: the quality it landed on and whether the target was met. When a target is impossible at the minimum quality, it says so and tells you the smallest achievable size, with the fix, reducing dimensions, named instead of pretending.

In PNG and WebP output, yes, untouched. JPEG has no alpha channel, so converting a transparent image to JPEG must flatten it: this tool detects transparency with a full-resolution scan, lets you pick the matte color, and states on the result card exactly what was flattened and to what. The output format is never changed behind your back: if this browser cannot encode WebP, the option is disabled up front rather than silently producing a JPEG.

For your tab’s survival, with the reason stated. Dimensions are read from the file header before any decoding, and a file whose decoded pixels would blow the memory budget (40 megapixels on desktop, 16 on constrained devices) is rejected with the estimated memory cost, because a 400-megapixel PNG can be 2 MB on disk and 4.8 GB decoded. Animated GIF and WebP are refused too: a still-image pipeline would silently keep only the first frame.

It is stripped: re-encoding goes through a canvas, which drops EXIF, GPS location, and other metadata, and orientation is normalized into the pixels first so photos do not rotate. That is usually what you want for web publishing (smaller files, no location leaks), and each result card states it. If you need metadata preserved, keep the original file for your archive.

Up to 20 per batch, processed through a bounded queue (one at a time on phones, two on desktops) so a big batch cannot decode everything simultaneously and crash the tab. Results download individually or as one ZIP with unique names and an optional manifest.csv recording sizes, formats, codecs and notes for every file. Everything runs locally; your images never leave the device.