We Ran 10 Photos Through the Pipeline: The Real Numbers

What we tested

Ten photographs, each 3000 × 3000 px, downloaded from a public sample set. Each was resized to 2000 px on the longest side with Lanczos, then encoded three ways at quality 82: progressive JPEG with optimization, WEBP with method 6, and PNG with optimization.

The JPEG files then went through the site’s own metadata injector, which writes Title, Subject, Tags and Comments (the Windows XP tags). Each output was re-read with piexif to confirm the fields existed. Finally, each tagged JPEG was converted to WEBP the same way the converter does it, and re-read to see whether the metadata survived.

Everything ran on the same machine, with Pillow 13 and piexif, on October 22, 2026. The script and the sample set are described at the end so anyone can repeat it.

The raw measurements

All sizes are KB. “Original” is the 3000 px download; the three output columns are 2000 px. The last two columns count how many of the four metadata fields were present (out of 4).

What the medians say

The WEBP saving is real but not uniform: one sample dropped 77%, another only 18%. That spread is the point. Averages hide it, which is why a single “WEBP saves 30%” claim is only a planning number — measure your own heaviest image before converting a catalogue.

PNG is not a photo format. At 7.4× the weight of JPG with no visible gain, a PNG product photo is the most expensive mistake in this table. Keep PNG for transparency and flat graphics.

Metadata: 10 out of 10 written, 10 out of 10 preserved

The injector wrote all four fields on every file. That part was expected; what mattered was what happened next.

In the first run, converting a tagged JPEG to WEBP produced a file with zero metadata fields. The converter was opening the image and saving it without passing the EXIF block through, so the keywords vanished silently — exactly the failure mode the new “verify metadata after upload” guide warns about.

We fixed the converter to pass the EXIF payload into JPEG, WEBP, AVIF and TIFF outputs. Re-running the same ten files: 10/10 kept all four fields. The lesson is not “our converter is perfect”; it is that metadata loss is usually a default in a library, not a platform conspiracy, and it is testable in five minutes.

What this means for a real gallery

Take a ten-photo listing at the medians above. As JPG, the gallery weighs about 4.5 MB. As WEBP, about 2.8 MB — roughly 1.7 MB less on one page. On a phone connection that is the difference between a page that feels instant and one that makes the buyer wait.

The saving is not free: WEBP is not accepted everywhere. For marketplace uploads, export JPG; for your own store, serve WEBP. The export matrix in the marketplace-formats guide covers which channel takes what.

Limitations, stated plainly

• Ten photographs is a sample, not a benchmark of every image type. Textured and noisy images compress worse than smooth studio shots.
• The sample images are not product photos on white; real catalogue images often compress better because they have less detail.
• Quality 82 is one setting. Quality 70 would be smaller and visibly softer on gradients; quality 90 would be larger with little visible gain.
• We measured file size and metadata presence, not perceived quality. Always compare at 100% zoom before publishing.
• The conversion test used our converter. A marketplace’s upload pipeline may still strip metadata; test it with the round-trip method.

Reproduce it in ten minutes

• Pick ten images from your own catalogue, including your heaviest hero shot.
• Run them through Speed Boost at 2000 px and quality 82, once as JPG and once as WEBP.
• Compare the ZIP sizes and note the median, not the average.
• Run the JPG batch through the SEO optimizer with four fields enabled.
• Open two outputs in the metadata inspector and confirm all four fields.
• Convert one tagged JPG to WEBP, then inspect the converted file. If the fields are gone, your converter is stripping them.
• Write the numbers in a note next to the batch. Next quarter, repeat and compare.

Frequently asked questions

Is WEBP always smaller than JPG?

Almost always for photographs, but the saving ranges from about 18% to 77% depending on content. Measure your own heaviest image.

Why did my metadata disappear after conversion?

Most image libraries do not copy EXIF unless you explicitly pass it. Test the output file, and use a converter that preserves it.

Should I use PNG for product photos?

No. In this test PNG was 7.4× heavier than JPG with no visible benefit. Use PNG only for transparency or flat graphics.

Does this test prove Etsy will keep my metadata?

No. It proves what our pipeline does. Marketplace uploads run their own processing; use the round-trip test on your own account.

What quality setting should I use?

75–85 for photographs is the usual sweet spot. Below 70, gradients and skin tones start to band.