Controlled file-size tests, with the numbers shown.

October 10 recheck: a byte target can conflict with fixed dimensions

We ran the actual image compressor on the existing synthetic 800 × 600 WebP (35,018 bytes) with a deliberately tight 1,024-byte target. The reference run used Node 24.19.0 and @napi-rs/canvas 1.0.7, not a browser encoder.

The smaller result is not interchangeable with an 800 × 600 deliverable. Lock dimensions when the destination requires them, and accept a larger byte budget if necessary. Try the dimension-locked browser example to measure your own browser.

In the same reference recheck, the PDF function produced 27,153 bytes below a 30,720-byte target and preserved one 595 × 842-point page. Native FFmpeg 7.1.5 produced a 183,976-byte, three-second 1280 × 720 H.264 video and a 23,920-byte, three-second AAC audio file. Their targets were 204,800 and 24,576 bytes. Small differences from the September records are expected across encoder versions; the earlier records remain labeled by date.

Actual Tesseract.js 7 recognition in Node again matched the ten English lines in the three existing synthetic OCR images. That does not establish translation accuracy or difficult-photo performance. Read the recheck settings and measurements. Live-browser interaction and physical-device performance are not established by these adapter tests.

What this test is designed to show

Two images can have identical pixel dimensions and radically different storage requirements. Flat graphics repeat the same colors and edges; screenshot-style interfaces repeat structured shapes; photo-like scenes contain gradients, noise, and irregular detail. The encoder has a much harder job with the last category.

For this first Fitabyte Lab reference set, we generated three synthetic 1600 × 1200 RGB images with no personal data using Python 3.13.5 and Pillow 12.3.0. We encoded each source as optimized PNG and at JPEG/WebP quality settings 85, 70, and 50, then recorded the resulting byte counts. The generation is deterministic so the same source can be recreated for future tests. These measurements describe that reference encoder—not Fitabyte’s browser encoder byte-for-byte.

Reference scenes

Flat graphic

Large flat colors, simple cards, thin lines, and sparse text-like shapes.

1600 × 1200 source

Screenshot-style UI

White panels, interface rows, icons, lines, and repeated high-contrast edges.

1600 × 1200 source

Photo-like scene

Gradients, deterministic noise, foliage-like texture, and irregular natural-style detail.

1600 × 1200 source

Measured file sizes

The table reports the actual encoded file size for this controlled test set. Quality numbers are encoder settings, not percentages of the original image quality.

What the numbers mean

Limitations and reproducibility

These are synthetic reference images, not a representative sample of every photograph or screenshot. The reference data was produced with Pillow rather than the browser encoder used by Fitabyte Compress. Different encoders can map quality values differently. Browser engines, operating systems, color profiles, metadata, transparency, and source content can all change the result. That is why Fitabyte treats the measurements as explanatory evidence rather than a universal benchmark.

Read How Fitabyte Tests Its Tools for our testing principles, or compare the formats in WebP vs JPG vs PNG. If you have a real size limit, use Fitabyte Compress and inspect the resulting file rather than assuming a preset will fit.