Measured examples · Updated October 7, 2026
Real files. Visible tradeoffs.
Six everyday decisions, backed by actual downloads. Real CC0 photographs and clearly labeled synthetic graphics show what changed, what stayed useful, and when a smaller result is the wrong choice.
A real photograph · a specific limit
A photo under 200 KB, with the tradeoff shown.
This detailed coffee-bean photo needed fewer pixels as well as compression. JPEG, Balanced, 50% dimensions, and a 200 KB limit produced this download.
Photo by Roger Burger, CC0, via Wikimedia Commons. Previews are reduced-size thumbnails; open the files to judge full-resolution detail. Measured 2026-10-07; environment and settings are in the example record. This is one result, not a promise for every photo.
See settings, original files & other outcomesWhy not just compress at the original dimensions?
Original-size Balanced JPEG produced 1124.4 KB. Original-size Balanced WebP produced 785.5 KB. Neither meets 200 KB. The 50% JPEG result does, but removes three quarters of the original pixel grid. Do not use that resize if your destination requires larger dimensions.
Two photos, two different outcomes.
With the same Balanced JPEG settings and original dimensions, the coffee photo changed from 4154.7 KB to 1124.4 KB, while the apple photo changed from 1328.6 KB to 475.8 KB. Detail and the starting encoding matter: one percentage cannot describe the whole batch.
Review each completed item with Compare. Download individually or use Download all (.zip) for successful results. A failed item stays out of the ZIP; the other items are not blocked.
The separately measured two-photo ZIP is 1600.5 KB including archive overhead. Open it to inspect both actual JPEG files.
A smaller logo still needs clean transparent edges.
The synthetic emblem has partially transparent edges. PNG keeps the prepared pixels; WebP changes them through lossy compression. Inspect both results against light and dark backgrounds, not only a checkerboard.
JPEG cannot retain alpha. Choose “Fill transparency with a background color” deliberately; a white rectangle will remain visible on a dark page once flattened. The app rejects transparent-to-JPEG output until you make that choice.
Sometimes the honest saving is zero.
Reprocessing our already compact synthetic PNG produced 96 → 96 bytes: the original was kept unchanged. That is a usable result, not a reason to display a made-up saving.
This retention is allowed only when format and dimensions match and the input metadata scan confirms no metadata to remove. An input with metadata can need re-encoding even if it becomes larger.
A failed limit needs a different decision, not a misleading download.
The coffee photo at original dimensions could not meet 20 KB within the tested JPEG settings. The result was “Needs attention”; no over-limit file was offered for download.
Keeping that same 20 KB limit but explicitly resizing to 10% produced 298 × 397 and 19.6 KB. It meets the limit, but is only suitable for a small placement. If your form requires a larger grid, raise the limit if permitted or choose a supported alternative—not this tiny result.
- Read the destination’s minimum dimensions and accepted formats.
- Change Resize only if those requirements permit it.
- Choose Apply settings to all, inspect Compare, then verify the actual download.
Photo files, attribution, and measurement method
Photographs were processed through the public workspace and browser worker, downloaded, then reopened for dimensions. Measured 2026-10-07 in Chromium 151.0.7922.34 on win32. Balanced is the starting preset; target mode may choose a lower feasible quality. These are measured examples, not predicted results for your files.
Small page previews are separate 640-pixel JPEG thumbnails. They are not the measured output and cannot prove full-resolution quality. Use the original/result links for that inspection. No third-party image host is contacted when this page loads.
- Green arabica coffee beans — Roger Burger, CC0 1.0. Source-page license checked 2026-10-07.
- Red apple on a plate 2017 A — Fructibus, CC0 1.0. Source-page license checked 2026-10-07.
coffee jpeg — 1,124,377 bytes
Starting quality: Balanced. Format: JPEG. Original dimensions. No limit.
COMPLETE coffee-beans.jpg 73% smaller ORIGINAL 4.2 MB RESULT 1.1 MB DIMENSIONS 2,976 × 3,968 → 2,976 × 3,968 FORMAT JPEG Metadata removed Compare Download
coffee target — 181,937 bytes
Starting quality: Balanced. Format: JPEG. Scale: 50%. Limit: 200 KB (1 KB = 1,000 bytes).
COMPLETE coffee-beans.jpg 96% smaller ORIGINAL 4.2 MB RESULT 182 KB DIMENSIONS 2,976 × 3,968 → 1,488 × 1,984 FORMAT JPEG Target met: 181,937 bytes ≤ 200,000 bytes · Tested quality 39 Metadata removed Compare Download
coffee webp — 785,454 bytes
Starting quality: Balanced. Format: WEBP. Original dimensions. No limit.
COMPLETE coffee-beans.jpg 81% smaller ORIGINAL 4.2 MB RESULT 785 KB DIMENSIONS 2,976 × 3,968 → 2,976 × 3,968 FORMAT WEBP Metadata removed Compare Download
apple jpeg — 475,820 bytes
Starting quality: Balanced. Format: JPEG. Original dimensions. No limit.
COMPLETE red-apple.jpg 64% smaller ORIGINAL 1.3 MB RESULT 476 KB DIMENSIONS 4,494 × 3,371 → 4,494 × 3,371 FORMAT JPEG Metadata removed Compare Download
coffee too small — target not reached
Starting quality: Balanced. Format: JPEG. Original dimensions. Limit: 20 KB (1 KB = 1,000 bytes).
No preview NEEDS ATTENTION coffee-beans.jpg Target not reached with the tested settings. Try smaller dimensions, another format, or a higher KB limit. No over-limit file was added to downloads. Adjust the output settings, then choose Apply settings to all. The rest of the batch is not blocked.
coffee recovered — 19,591 bytes
Starting quality: Balanced. Format: JPEG. Scale: 10%. Limit: 20 KB (1 KB = 1,000 bytes).
COMPLETE coffee-beans.jpg 100% smaller ORIGINAL 4.2 MB RESULT 20 KB DIMENSIONS 2,976 × 3,968 → 298 × 397 FORMAT JPEG Target met: 19,591 bytes ≤ 20,000 bytes · Tested quality 59 Metadata removed Compare Download
optimized repeat — 96 bytes
Starting quality: Balanced. Format: PNG. Original dimensions. No limit.
COMPLETE optimized-repeat.png Already optimized — original kept unchanged ORIGINAL 96 B RESULT 96 B DIMENSIONS 160 × 100 → 160 × 100 FORMAT PNG Compare Download
Earlier synthetic measurements: inspect individual files
Each source was selected in the same public image workspace, processed by its browser worker, downloaded, and reopened to check dimensions. Results below were measured in Chromium 151.0.7922.34 on Linux on 2026-09-19 using 0.1.0 + image-apex-2026-09-19. A different browser decoder or app version can produce different bytes; these figures describe these files, not a guaranteed saving for your images.
Sources and results are original MallowDesk test graphics released under CC0-1.0. The orientation tag is fabricated and contains no real location or camera data. Download the measurement record and SHA-256 hashes.
Synthetic fine texture; original dimensions; balanced JPEG. — 95,870 bytes
Synthetic fine texture; original dimensions; balanced JPEG.
1,974,750 → 95,870 bytes. Output: 960 × 640 JPEG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE texture.png 95% smaller ORIGINAL 2.0 MB RESULT 96 KB DIMENSIONS 960 × 640 → 960 × 640 FORMAT JPEG Compare Download
Same texture, 50 KB limit; actual feasible output. — 40,492 bytes
Same texture, 50 KB limit; actual feasible output.
1,974,750 → 40,492 bytes. Output: 960 × 640 JPEG. Starting quality: Balanced. Original dimensions. Limit: 50 KB (1 KB = 1,000 bytes).
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE texture.png 98% smaller ORIGINAL 2.0 MB RESULT 40 KB DIMENSIONS 960 × 640 → 960 × 640 FORMAT JPEG Target met: 40,492 bytes ≤ 50,000 bytes · Tested quality 39 Compare Download
Same texture at 50% width and height, balanced JPEG. — 27,123 bytes
Same texture at 50% width and height, balanced JPEG.
1,974,750 → 27,123 bytes. Output: 480 × 320 JPEG. Starting quality: Balanced. Scale: 50%. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE texture.png 99% smaller ORIGINAL 2.0 MB RESULT 27 KB DIMENSIONS 960 × 640 → 480 × 320 FORMAT JPEG Compare Download
Same texture, balanced WebP. Quality numbers are not equivalent across codecs. — 109,280 bytes
Same texture, balanced WebP. Quality numbers are not equivalent across codecs.
1,974,750 → 109,280 bytes. Output: 960 × 640 WEBP. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE texture.png 94% smaller ORIGINAL 2.0 MB RESULT 109 KB DIMENSIONS 960 × 640 → 960 × 640 FORMAT WEBP Compare Download
Smooth synthetic gradient: inspect banding and color transitions. — 5,991 bytes
Smooth synthetic gradient: inspect banding and color transitions.
272,353 → 5,991 bytes. Output: 720 × 480 JPEG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE gradient.png 98% smaller ORIGINAL 272 KB RESULT 6.0 KB DIMENSIONS 720 × 480 → 720 × 480 FORMAT JPEG Compare Download
A synthetic emblem with partially transparent edges, optimized losslessly. — 7,628 bytes
A synthetic emblem with partially transparent edges, optimized losslessly.
13,907 → 7,628 bytes. Output: 480 × 480 PNG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE transparent-emblem.png 45% smaller ORIGINAL 14 KB RESULT 7.6 KB DIMENSIONS 480 × 480 → 480 × 480 FORMAT PNG Compare Download
The same transparent emblem with lossy WebP encoding. — 5,522 bytes
The same transparent emblem with lossy WebP encoding.
13,907 → 5,522 bytes. Output: 480 × 480 WEBP. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE transparent-emblem.png 60% smaller ORIGINAL 14 KB RESULT 5.5 KB DIMENSIONS 480 × 480 → 480 × 480 FORMAT WEBP Compare Download
Synthetic text screenshot. Inspect small characters and lines. — 13,565 bytes
Synthetic text screenshot. Inspect small characters and lines.
48,331 → 13,565 bytes. Output: 960 × 540 PNG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE text-screenshot.png 72% smaller ORIGINAL 48 KB RESULT 14 KB DIMENSIONS 960 × 540 → 960 × 540 FORMAT PNG Compare Download
The same screenshot in JPEG. Inspect ringing around text. — 20,790 bytes
The same screenshot in JPEG. Inspect ringing around text.
48,331 → 20,790 bytes. Output: 960 × 540 JPEG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE text-screenshot.png 57% smaller ORIGINAL 48 KB RESULT 21 KB DIMENSIONS 960 × 540 → 960 × 540 FORMAT JPEG Compare Download
Fabricated EXIF orientation 6 is applied; visible dimensions become 200 × 320. — 163 bytes
Fabricated EXIF orientation 6 is applied; visible dimensions become 200 × 320.
1,357 → 163 bytes. Output: 200 × 320 PNG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE orientation.png 88% smaller ORIGINAL 1.4 KB RESULT 163 B DIMENSIONS 200 × 320 → 200 × 320 FORMAT PNG Orientation applied · metadata removed Compare Download
A compact flat-color source. The original is kept if re-encoding is no smaller. — 96 bytes
A compact flat-color source. The original is kept if re-encoding is no smaller.
264 → 96 bytes. Output: 160 × 100 PNG. Starting quality: Balanced. Original dimensions. No target limit.
Inspect the actual output
This is the downloaded file, scaled to fit this page. Open the output link to inspect it at its original size. A checkerboard background helps reveal transparent edges.
Recorded result and applied settings
COMPLETE optimized.png 64% smaller ORIGINAL 264 B RESULT 96 B DIMENSIONS 160 × 100 → 160 × 100 FORMAT PNG Compare Download
What these cases teach
The texture and resized texture use the same source, separating dimension reduction from quality changes. The text screenshot makes edge artifacts easy to inspect. The transparent emblem tests partially transparent edges. The orientation case checks whether removing metadata leaves the visible image correctly rotated. The compact flat-color case shows why starting size alone cannot predict further savings.
These synthetic patterns complement the photographs above. Noise, detail, color, and the destination’s format support change the tradeoff. Choose your own image and inspect Compare before using the output.