Controlled Image Compression Benchmark
A reproducible synthetic-image benchmark comparing optimized PNG, JPEG, WebP, and AVIF outputs across six visual patterns.
What was measured
Six 1200×800 synthetic images were generated to represent a gradient/photo-like scene, flat illustration, text-heavy screenshot, line art, noisy texture, and transparency. Each source was encoded with the same documented settings.
How to interpret the table
The smallest file is not automatically the best deliverable. Review visible quality, alpha support, decoding, browser support, and whether a source pattern resembles the real image.
Reproduce rather than generalize
The downloadable table records this controlled run. Publishers should repeat the test with their own assets and encoder versions before setting a site-wide rule.
How to read the downloadable table
The table is organized around “Synthetic sample” and “PNG KB.” Begin with the row labeled “gradient-photo,” then read across the full row before comparing it with another case. The cells are designed to preserve context; copying a single number or phrase without its row label can change the meaning.
Use the last row, “transparent-shapes,” as a completeness check rather than as an automatic conclusion. Where the table contains scores, thresholds, examples, or suggested actions, they apply only under the method and limitations stated on this page.
A repeatable application workflow
Measure a real page under a documented device, network, and cache condition. Change one variable at a time, repeat the run, and use field data when available instead of assuming a single laboratory score represents every visitor.
For controlled image compression benchmark, keep a short review log containing the date, page or file tested, input values, result, reviewer notes, and any source that changed the interpretation. This turns a one-time check into an auditable workflow.
What evidence to preserve
Save the exact version of the input, a screenshot or exported result where appropriate, the source URL, and the date accessed. If the result depends on software, include the browser, library, encoder, calculator version, or device conditions that could affect reproduction.
When publishing a conclusion, distinguish an observation from an inference. An observation reports what the documented test produced; an inference explains what that result may mean. Readers should be able to see which is which.
Maintenance and citation practice
Cite this resource using the full title, ToolnixHub, the reviewed date, the canonical report URL, and the relevant table or section. Link to the report page rather than an isolated download so readers can see the methodology and limitations.
Recheck the underlying primary references before using the resource for a time-sensitive decision. Standards, rates, browser support, product behavior, and official guidance can change after the reviewed date, even when the general workflow remains useful.
| Synthetic sample | PNG KB | JPEG KB | WebP KB | AVIF KB | Smallest output |
|---|---|---|---|---|---|
| gradient-photo | 15.6 | 40.4 | 22.4 | 9.9 | AVIF |
| flat-illustration | 6.2 | 18 | 4.5 | 2.8 | AVIF |
| text-screenshot | 5.7 | 31 | 4.5 | 1.8 | AVIF |
| line-art | 40.9 | 257 | 167.3 | 24.7 | AVIF |
| noise-texture | 3116.8 | 637.3 | 657.7 | 407.3 | AVIF |
| transparent-shapes | 13.7 | 19.4 | 9.5 | 8.9 | AVIF |
Methodology
Generated 1200×800 images were encoded with Pillow 12.2.0 using PNG optimize, JPEG quality 82 optimize, WebP quality 80 method 6, and AVIF quality 55. Byte sizes were measured in memory and converted to KiB.
Limitations
Synthetic images do not represent every camera photo, illustration, encoder, browser, or quality threshold. File size alone does not measure perceptual quality.
Questions about this report
Are these industry averages?
No. They are results from a documented synthetic test run.
Why include a noise texture?
High-entropy content helps show that compression behavior changes with visual complexity.
How often should this resource be reviewed?
Review it whenever a primary reference, rate, standard, browser behavior, tool implementation, or decision context changes. For time-sensitive use, verify the sources on the day of the decision.
May a publisher reuse the dataset?
Publishers may quote or summarize a reasonable portion with clear attribution and a link to the canonical report. Republishing the complete dataset as a substitute for this resource requires permission.