Client-Side File Compressor & Visual Difference Engine

File Options

Generate Visual Fidelity & SSIM Comparison

File Format Guide

Everything on this page runs in your browser. Your files are never uploaded.

WebP

A modern format that typically produces smaller files than JPEG at similar visual quality, and also supports transparency like PNG. Best choice for web use when the destination supports it.

JPEG

The universal photo format. Lossy compression works well for photos but can leave blocky artifacts on sharp text and edges, and has no transparency. Best when maximum compatibility matters.

PNG

Lossless, with transparency. Ideal for logos, screenshots and sharp graphics, but photos make very large PNGs. Here, PNG compression reduces the number of colors (like TinyPNG) to shrink files while keeping transparency.

Camera RAW

NEF, CR2/CR3, ARW, DNG, RAF, RW2, ORF, PEF and similar files hold sensor data plus a camera-rendered JPEG preview. This tool converts using that embedded preview (no sensor demosaicing), then compresses it to JPEG, WebP or PNG. The preview resolution depends on the camera and may be lower than the sensor's full size.

PDF

A container for text, fonts, vector graphics and embedded images. Scans and photo-heavy PDFs shrink a lot because the images are optimized; text-only PDFs (like a resume) are already compact and shrink very little.

Why does 100% quality still shrink my JPEG?

The quality number is an encoder setting, not "percent of the original data kept". Different programs map it differently, so "100%" in Photoshop and "100%" in a browser are not the same.
Chroma subsampling: 4:4:4 stores color at full resolution; 4:2:0 stores color at a quarter of the resolution while keeping brightness detail (which drives perceived sharpness) intact. High-quality exports from editors such as Photoshop typically use 4:4:4, so re-encoding can save a lot with little visible change.
Fine sensor grain in photos developed from RAW is costly to store and barely visible; re-encoding often discards some of it.
Metadata such as edit history, embedded previews and color profiles can add noticeable size.
Re-compressing an already-lossy JPEG never recovers lost detail, so use the SSIM score and heatmap to judge the result.

How to read SSIM and the heatmap

SSIM measures structural similarity from 0 to 1, where 1.0 means identical. Scores around 0.98 or higher usually look the same to most people on photos.
The heatmap is black & white: white areas barely changed and darker areas changed the most.