Methods and image sources

Check where a recommendation comes from, what was measured and how to reproduce the published examples.

Four layers of information

Official source. Platform references link to the current marketplace documentation and state when it was checked.
Reference measurement. The downloadable encodings record the actual file dimensions, byte counts and hashes from one specified implementation.
Worked example. Crop geometry and pixel-buffer estimates are identified as calculations or controlled examples. They are not device-memory or marketplace benchmarks.
Workflow advice. The guides explain how to make a file decision and verify the destination preview. Their recommendations remain separate from platform requirements.

Download and reproduce

The six-input format comparison publishes its normalized inputs and actual output files. It uses two CC0 photographs and four clearly identified generated fixtures. The photographs were already JPEG-compressed before normalization. Every format starts from the same opaque 1200 × 900 pixels, with white compositing; this comparison does not test alpha preservation.

Measured 2026-10-20 using Node 24.13.1, sharp 0.35.4 and libvips 8.18.6. JPG/WebP use quality 85 and PNG uses default lossless encoding. Equal numeric quality parameters do not imply equal perceived quality.

Download the manifest of dimensions, bytes and SHA-256 hashes. The generator and verification script are in the repository. Browser output is a separate test because its decoder, scaling and encoder can differ.

Photograph credits

Coffee Cup, accession 39.18.3, The Metropolitan Museum of Art, CC0 1.0. Used as third-party test material; fitted, resized and re-encoded for the examples.
Coffee cup and coffee bean, mcfoodie, CC0 1.0. Used as third-party test material; fitted, resized and re-encoded for the examples.

Text, pattern, transparency and seeded-noise fixtures are generated test images. They are not photographs of goods sold by this project. Synthetic stress tests illustrate specific properties and do not establish a universal format ranking.

Revision rules

Every marketplace reference includes a last-checked date and direct source links. We update that date only after reviewing the linked source or changing the tested method. A date is not a promise that the platform will remain unchanged. When guidance conflicts across official help pages, we show the difference rather than inventing one universal threshold.

Editorial integrity

We do not invent seller histories, testimonials, user counts, credentials or approval results. AI may assist with drafting code or prose, but the explanation must be checked against the implemented workflow and cited source.

Technical scope

Image Prep, Image QA and the format-comparison tool run in the browser. Selected files are not sent to a Seller Operations Lab server. Local processing still depends on device memory, browser behavior and the user's download settings.

Corrections

Report a reproducible technical issue or source correction through the contact page. Do not attach private customer files to a public issue.