WebP vs JPEG: Measured at Equal SSIM
On this NASA photograph and these FFmpeg encoders, real files selected by six-step bisection near SSIM 0.960 were 2,174,520 bytes for JPEG (SSIM 0.961625) and 2,224,544 bytes for lossy WebP (SSIM 0.960073). JPEG was 2.3% smaller on this photograph only. This is a measurement of one image and two encoder settings, not a general format verdict.
Short answer
On our NASA photograph at SSIM near 0.960, JPEG came out 2.3% smaller than lossy WebP, 2,174,520 bytes against 2,224,544. That is one image and two encoder settings, not a general verdict, and only WebP supports transparency.
Measured data
The image run used a NASA public-domain photograph, 3.1 MB, decoded once to a fixed RGB reference. Results describe this photograph and these exact encoder settings.
| Format | Exact encoder setting | Real file bytes | Measured SSIM |
|---|---|---|---|
| JPEG | q:v 4, huffman optimal | 2,174,520 | 0.961625 |
| WebP lossy | quality 85, compression level 6 | 2,224,544 | 0.960073 |
Each format was tested in six bisection steps and the chosen row is a real encoded file within SSIM 0.960 ± 0.002, not an interpolation. JPEG arguments: `-c:v mjpeg -q:v 4 -huffman optimal`. WebP arguments: `-c:v libwebp -q:v 85 -compression_level 6`.
The fixed reference is a NASA public-domain photograph, 3.1 MB, decoded once to RGB. Measure a representative image before treating either result as a format rule.
When to use each option
Use the result only for this photograph
JPEG was 2.3% smaller at this matched score on this photograph and these encoder settings.
Use WebP after measuring your own image
WebP can win on other images and settings. Test the actual asset, decoder path and quality target instead of relying on a generic percentage.
Use neither result for graphics
A photograph does not represent text, line art or transparency. Measure those assets separately.
Questions people ask
Is WebP always smaller than JPEG?
No. JPEG was 2.3% smaller in this one matched test; other images and encoder settings can differ.
What does equal quality mean here?
Both real files fall within SSIM 0.960 ± 0.002 against one decoded RGB reference.
Why should I test my own image?
Content, encoder and target quality all affect byte size. This page measures one NASA photograph only.
Can JPEG store transparency?
No. Use WebP or PNG when an alpha channel is required, then test the target that consumes it.
Measured vs sourced
Cells marked measured are Koldflux Media Lab measurements. Cells marked sourced derive from each platform / codec standards body. Cells marked guidance are expert guidance, not measured.
Sources
- Google WebP compression documentation (read 2026-10-08)
- NASA image record: Buzz Aldrin salutes the U.S. flag (read 2026-10-08)
- FFmpeg codec documentation (read 2026-10-08)
Verified at 2026-10-08.
Next steps
Inspect my file: /tools/validate
Convert / make correct version: /tools/convert
Create all target versions: /tools/pack