Reproducible browser conversion tests

These results come from the same conversion code used by the public tools, including the current high-quality MozJPEG and libwebp encoders. They are dated measurements, not promised compression rates or universal codec claims.

Measured 3 October 2026 · Chromium 151.0.7922.34

Image test card

One source, four exports, measured in the browser

The source is a generated 960 × 640 PNG containing small text, a colour gradient, flat colour patches, narrow lines and a transparent outer border. Every export starts from that same source.

Export	Actual format	Dimensions	Bytes	Top-left pixel
Original PNGReference fixture generated in the browser.	image/png	960 × 640	172,423	Transparent (alpha 0)
PNG, unchanged sizeLossless re-export matched the reference byte size in this run.	image/png	960 × 640	172,423	Transparent (alpha 0)
JPG quality 75MozJPEG HQ with 4:4:4 colour detail; transparency was flattened to white.	image/jpeg	960 × 640	31,146	Opaque (alpha 255)
JPG quality 92MozJPEG HQ with 4:4:4 colour detail. Higher quality increases size and preserves more encoded detail.	image/jpeg	960 × 640	54,085	Opaque (alpha 255)
WebP quality 85libwebp HQ with sharp-YUV conversion; transparency remained available.	image/webp	960 × 640	18,340	Transparent (alpha 0)

Encoder quality check

MozJPEG HQ versus the browser JPEG encoder on the same fixture

At quality 95, with the 960 × 640 test card flattened onto white and no resize or enhancement, the high-quality JPEG path produced a closer pixel match to the reference fixture in this Chromium run.

Encoder	Quality setting	File size	PSNR vs white-flattened reference
MozJPEG HQ · 4:4:4	95	64,233 bytes	45.57 dB
Chromium canvas JPEG	0.95	58,065 bytes	42.79 dB

PSNR is a pixel-error measurement for this synthetic fixture only. A higher value here means the decoded JPEG was numerically closer to this reference image; it does not prove that one encoder will look better for every photograph or every file-size target.

What the run shows

Useful observations, with limits stated

PNG round-trip: The unchanged PNG export matched the reference byte size and retained the transparent test pixel in this Chromium run.
JPEG transparency: Both JPEG outputs were opaque because JPEG has no alpha channel. The converter intentionally flattened the source onto white for this test.
Quality and size: JPEG quality 92 produced a larger file than quality 75. Quality controls are encoder settings, not a percentage of original detail retained.
WebP result: WebP quality 85 produced the smallest file in this particular synthetic fixture while keeping the transparent test pixel. That does not make WebP 85 universally better than JPEG 75 or 92.

Important limitation

This is not a universal benchmark.

Different photographs, graphics, colour profiles, browser versions and encoders can produce different visual trade-offs and file sizes. The table is evidence for this fixture and environment only.

For an important asset, compare the real output at its intended display size and keep the best source file separately.

Method

How this dataset was produced

1
Generate the source. The site creates its known test card locally rather than downloading an unknown sample.
2
Convert from the same original. PNG, JPG 75, JPG 92 and WebP 85 each start from the reference PNG, with dimensions held at 960 × 640.
3
Measure the actual result. The browser reports MIME type, dimensions, byte length and the alpha value of the intentionally transparent top-left pixel.
4
Publish the environment. The browser version and date are recorded so a future run can be compared instead of silently replacing the result.

Next test

Use your own destination requirements.

A converter can only prepare a candidate file. The receiving website, editor, document system or print workflow decides which formats, dimensions and file sizes are acceptable.