Why three files give three different answers

A noisy image spends bytes on fine detail. A flat transparent graphic has repeated pixels that a lossless encoder can describe efficiently. A native-text PDF can already be compact. The same “make it smaller” request therefore needs different settings and different checks.

Our examples are synthetic files created for testing. They contain no customer data and are free to download and reuse. They are not a representative benchmark, a passport-photo sample, or a promise about the reduction you will get.

High-detail synthetic image

This deliberately noisy image is a stress case, not a typical photograph. Reaching the target required resizing from 1800 × 1200 to 854 × 569 and lossy JPEG encoding. The smaller file does not preserve the original pixels.

Original
2,596,260 bytes
Output
198,911 bytes
Saved
92.3%
Original synthetic noise pattern, 1800 by 1200 pixels
Original · fitted preview
Lossy compressed noise pattern, resized to 854 by 569 pixels
Output · open the files to inspect at full size

Settings: Lossy compression allowed; JPEG output; maximum 200,000 bytes; resizing allowed; metadata removal enabled; OCR off.

Checks passed: min bytes, max bytes, mime, width, height, metadata removed. A technical pass does not certify visual readability or acceptance by an upload portal.

Transparent flat graphic

A single-colour translucent rectangle compresses very well. Lossless WebP retains the decoded RGBA pixels and 600 × 400 dimensions. This large percentage saving should not be extrapolated to detailed photographs. The receiving portal must accept WebP.

Original
1,891 bytes
Output
318 bytes
Saved
83.2%
Original translucent green rectangle, 600 by 400 pixels
Original · fitted preview
Lossless WebP rectangle retaining the original pixels and transparency
Output · open the files to inspect at full size

Settings: Strict lossless; WEBP output; maximum 10,000 bytes; resizing disabled; metadata removal enabled; OCR off.

Checks passed: min bytes, max bytes, mime, width, height, metadata removed, lossless pixels. A technical pass does not certify visual readability or acceptance by an upload portal.

One-page native-text PDF

This tiny PDF contains native text, not a scanned page. Structural cleanup saves only 45 bytes. Page count and extracted text remain unchanged. A compact source has little redundant data to remove; dramatic savings are not always possible.

Original
719 bytes
Output
674 bytes
Saved
6.3%

Settings: Strict lossless; PDF output; maximum 50,000 bytes; resizing disabled; metadata removal enabled; OCR off.

Checks passed: min bytes, max bytes, mime, page count, text preserved, unencrypted. A technical pass does not certify visual readability or acceptance by an upload portal.

A failure is sometimes the correct result

We also asked the noisy image to fit into 512 bytes as a lossless PNG without resizing. The engine rejected the request. The required pixels cannot be preserved within that budget. In that situation, choose a larger maximum or explicitly allow lossy changes; the tool should not silently discard information.

How to reproduce and review these results

  1. Download an original above and open the free workspace. The homepage’s “Try a sample image” button loads the first example and its settings for you.
  2. Select On-device only, match the output format and exact byte ceiling, then match the compression mode and resize setting. For the image example, 195.3125 in the workspace’s binary KB control equals 200,000 bytes.
  3. Process the file. Inspect the reported bytes, dimensions or page count, and download the result. Compare the original and output in your own viewer. Use strict lossless for the graphic and PDF examples.

These measurements were generated by running the shipped WebAssembly module locally in Node.js. Browser speed and memory availability vary; the published numbers are file sizes, not timing claims. The machine-readable report includes all constraints, validation checks and SHA-256 hashes of the inputs, outputs and engine.

For the transparent graphic, we also decoded both files independently and compared all RGBA pixels. For the PDF, we compared extracted text and page rendering. Lossless refers to the checked content, not identical file bytes or retention of metadata.

Choose your next step

For a photograph, start with the 200 KB image tool and inspect fine details. For a signed or interactive PDF, keep the original and follow the destination’s instructions before modifying it. Read our guides to image dimensions, scanned PDFs and exact byte limits.