TinyOwl

BENCHMARK

Compression results from a mixed benchmark corpus.

This page records the current TinyOwl compression benchmark so the homepage can stay simple while the test setup, tradeoffs, and current numbers remain easy to update.

CURRENT RESULT

The same local batch can stay familiar, shrink fast, or push for the smallest export.

These numbers are meant to compare export paths, not chase one abstract percentage. Preserve-original keeps the workflow predictable, WebP q50 is the strongest high-compression option in this corpus, and AVIF q50 is the smallest-file path when longer runtime is acceptable.

106.56 MB
0 %

106.56 MB -> 75.24 MB

Preserve original q75

106.56 MB
0 %

106.56 MB -> 6.91 MB

Export as WebP q50

106.56 MB
0 %

106.56 MB -> 4.90 MB

Export as AVIF q50

Results

Mode Input Output Savings Time
Preserve original formats q75 106.56 MB 75.24 MB 29.39% 36.24s
Preserve original formats q50 106.56 MB 25.75 MB 75.83% 12.48s
Export as WebP q75 106.56 MB 9.93 MB 90.68% 21.79s
Export as WebP q50 106.56 MB 6.91 MB 93.52% 22.47s
Export as AVIF q75 106.56 MB 8.89 MB 91.66% 73.79s
Export as AVIF q50 106.56 MB 4.90 MB 95.40% 53.74s
PNG source only q75 (already-compressed PNG set) 78.25 MB 69.53 MB 11.14% 26.90s
PNG source only q50 (stronger PNG reduction) 78.25 MB 22.33 MB 71.46% 8.56s

Test Setup

  • Mixed local benchmark corpus combining real desktop screenshots and public format sample files.
  • Release backend profile, 8 threads.
  • Preserve-original runs used q75 and q50 while keeping the current output format for each source image.
  • WebP runs used q75 and q50 across the full mixed corpus.
  • AVIF runs used q75 and q50 across the full mixed corpus.
  • PNG rows are limited to source PNG files only, so unrealistic JPG/WebP to PNG expansion does not distort the main benchmark summary.
  • Results vary by image content, format, dimensions, and selected preset.

How To Read This

The main table excludes JPG or WebP to PNG because that is usually not a sensible export path and can make files larger. In the combined corpus, the PNG-only q75 result improves to 11.14%, which better reflects how real screenshots and public sample PNGs behave together. The better product story here is still the path comparison: preserve original for safety, WebP q50 for practical high compression, AVIF q50 for smallest output.