Catch the small edits.
Compare revisions of a banner, screenshot or graphic. Use the reveal and overlay views, then inspect the highlighted pixels.
Compare an original with a revised image. Reveal the edits, measure the pixel differences and take a clear report into your next review.
No image selected
No image selected
Drag the reveal handle, use its arrow keys, or adjust the reveal and overlay controls below View. Click a preview to inspect a coordinate, or enter exact coordinates below. Views and reports do not measure image quality or authenticity.
Compare revisions of a banner, screenshot or graphic. Use the reveal and overlay views, then inspect the highlighted pixels.
Check a compressed image against its original. MAE and RMSE describe numeric channel error; judge visual quality separately.
Analyze a rectangle, adjust tolerance and export a mask or report. Every comparison uses original-resolution decoded pixels.
No. Both images are decoded and compared locally in your browser. No image or comparison history is automatically saved. Normal site analytics and feedback services may still load.
A pixel counts as changed when its largest absolute compared-channel difference is greater than the selected tolerance, from 0 to 255. Tolerance 0 detects every numeric difference; tolerance 255 counts none. MAE and RMSE always use all pixels in the selected area before tolerance filtering.
Visible colours mode composites each decoded RGBA pixel onto the chosen solid background in 8-bit sRGB and rounds each RGB channel before comparison. RGBA mode compares four decoded channels, including alpha. Fully transparent RGB may already have been discarded by browser decoding. Colour profiles and high-bit-depth files may be converted by the browser.
Equal dimensions are required by default. Top-left padding compares the union canvas with missing pixels treated as transparent. No image is stretched, centred or automatically aligned. A differently sized fully transparent border may have zero visual difference; check dimensions as well as metrics.
No. This tool compares decoded pixels in selected coordinates, not original file bytes, metadata, meaning or authenticity. Pixel differences are not a perceptual similarity score and do not say which image looks better. Compression, colour conversion and alignment can affect the results.
The image previews are reduced to fit. The reduced heatmap uses the maximum difference in each preview cell so isolated changes are not dropped. Heatmap gain changes display intensity only. Full-resolution heatmap and mask PNGs use one output pixel per comparison pixel; the numeric inspector reads original-resolution values.
Use two still PNG, JPEG or WebP images, each up to 20 MiB, 12 million pixels and 8192 pixels per side. The union comparison canvas must also fit 12 million pixels. Animated images, SVG, HEIC and RAW are unsupported. Large pairs may exceed some phones’ memory.
Download a full-resolution difference heatmap or binary change mask as PNG. The mask uses white for pixels above tolerance and black elsewhere. JSON and CSV reports contain filenames, dimensions, settings and numeric results. CSV prefixes formula-like filenames with an apostrophe, including leading control characters; spreadsheet applications may still reinterpret text. Use JSON to preserve filename text exactly. Coordinates start at zero. Regions outside the selected analysis area are excluded from metrics.