Target-size compression: measured service results

Live target-size tests with exact downloaded bytes, settings, matching frames and source attribution, plus earlier MiB tests.

Last updated: 2026-10-23

Live tests: 10MB, 19MB and 25MB targets

Measured on 2026-10-23 using the public website. Each source was submitted once in target-size mode. The byte counts below come from the downloaded MP4 files.

Screen recording · text and scrolling · 10MB target

This source went from 13.27 MB to 9.48 MB with a 10 MB target: 28.5% smaller.

Source
13,265,235 bytes (13.27 MB) · 240s · 1920 × 1080
Compressed
9,484,147 bytes (9.48 MB) · 240s · 1920 × 1080

Settings: target-size mode · H.264 MP4 · Fast · Keep original

Credits used: 8

The screen fixture repeats our original 60-second fictional webpage recording four times using stream copy, making 240 seconds without padding or increasing bitrate. It is a prepared test fixture, not a newly recorded four-minute session.

Screen recording · text and scrolling: Source, frame at 24 secondsScreen recording · text and scrolling: Compressed, frame at 24 seconds
Source → downloaded result · matching frames at 24s Frames at 24s. Open an image to inspect its encoded dimensions. JPEG snapshots are illustrative; they do not establish motion or audio quality.

Original recording of a fictional webpage, prepared by repeating the original four times with stream copy. · Source and attribution · Test method and licenses

Gameplay · SuperTuxKart · 19MB target

This source went from 52.05 MB to 18.55 MB with a 19 MB target: 64.4% smaller.

Source
52,046,748 bytes (52.05 MB) · 70.834s · 1920 × 1080
Compressed
18,546,631 bytes (18.55 MB) · 70.833333s · 1920 × 1080

Settings: target-size mode · H.264 MP4 · Fast · Keep original

Credits used: 3

The game result keeps the source dimensions, but the frame at 30 seconds shows visible blocks in the sky, grass and road. Keeping 1080p does not mean keeping the same detail.

Gameplay · SuperTuxKart: Source, frame at 30 secondsGameplay · SuperTuxKart: Compressed, frame at 30 seconds
Source → downloaded result · matching frames at 30s Frames at 30s. Open an image to inspect its encoded dimensions. JPEG snapshots are illustrative; they do not establish motion or audio quality.

SuperTuxKart authors; recording by PantheraLeo1359531. Wikimedia 1080p VP9 derivative. Frames and recompressed derivatives: CC BY-SA 4.0. · Source and attribution · Test method and licenses

Sintel · prepared five-minute excerpt · 25MB target

This source went from 203.76 MB to 24.30 MB with a 25 MB target: 88.1% smaller.

Source
203,756,551 bytes (203.76 MB) · 299.419352s · 1920 × 818
Compressed
24,303,584 bytes (24.30 MB) · 299.416341s · 1276 × 544

Settings: target-size mode · H.264 MP4 · Fast · 720p

Credits used: 10

Sintel · prepared five-minute excerpt: Source, frame at 240 secondsSintel · prepared five-minute excerpt: Compressed, frame at 240 seconds
Source → downloaded result · matching frames at 240s Frames at 240s. Open an image to inspect its encoded dimensions. JPEG snapshots are illustrative; they do not establish motion or audio quality.

© Blender Foundation | www.sintel.org. Continuous prepared five-minute excerpt (stream copy from the September 16 H.264 MP4). CC BY 3.0. · Source and attribution · Test method and licenses

These individual tests do not predict other videos, prove acceptance by Discord or an email provider, or establish an 8MB result. MB means 1,000,000 bytes. Timings and server telemetry were not independently captured for these runs.

What happened in these two tests

On September 23, 2026, we submitted a 60-second screen recording with a 2 MiB target and a 30.769-second phone clip with a 20 MiB target to the live compressor. The downloaded results were 1.92 MiB and 18.89 MiB, respectively. Both stayed below their selected targets and each used 2 compression credits.

Each input was tested once. These results are not an average, a guarantee for other videos, or an independent review. Both used target-size mode, H.264, Fast speed, and the original 1920 × 1080 resolution. The screen recording has no audio; the phone result contains AAC audio. These are SDR inputs, not HDR or Dolby Vision tests.

File sizes, times, and credits

Measured results from September 23, 2026

MeasurementScreen recordingPhone clip
Source3,316,953 bytes (3.16 MiB)50,944,770 bytes (48.58 MiB)
Target upper bound2,097,152 bytes (2 MiB)20,971,520 bytes (20 MiB)
Downloaded result2,012,378 bytes (1.92 MiB)19,803,861 bytes (18.89 MiB)
Source → result duration60 → 60 seconds30.769 → 30.700 seconds
Encoding time66.5 seconds79.2 seconds
Total task time103.1 seconds143.9 seconds
Credits consumed22

Task time runs from server task creation to its completed update and excludes downloading. Encoding time is the server's encoding measurement. The result byte counts were checked against both the downloaded files and the task records. The machine-readable measurements include SHA-256 hashes for the outputs.

MiB and MB are different. A 2 MiB target is about 2.097 MB, so it does not guarantee a file below a 2 MB upload limit. A 20 MiB target is about 20.972 MB. For sharing, choose a target below the receiving service's limit; the compressor displays the conversion.

The two runs consumed 4 credits in total. Eligible new accounts receive 30 compression credits once at signup, valid for 30 days from receipt with visitor usage deducted, but this does not mean every account can compress 15 videos: the estimate depends on the source and settings. Check the estimate before starting and see current plans.

Screen recording: inspect the text at 24 seconds

This is an original recording of a fictional test webpage. It is the same input used in the September 16 quality-mode case, reused here with a target-size setting.

Source screen recording at 24 seconds
Source · 24 seconds
Target-size screen recording result at 24 seconds
2 MiB target result · 24 seconds

Open the source video · Open the target-size result

Phone clip: inspect the scene at 10 seconds

The source is Benoît Prieur's iPhone 7 recording from Druyes-les-Belles-Fontaines, provided under CC0. We used the downloaded WebM copy, not the camera-original file. The result's container duration is 30.700 seconds, compared with 30.769 seconds for the source.

Source phone clip at 10 seconds
Source · 10 seconds
Target-size phone clip result at 10 seconds
20 MiB target result · 10 seconds

Open the source video · Open the target-size result

Click a still for the full-resolution PNG. Matching frames help compare detail but do not demonstrate motion or audio quality; inspect the videos for those. Neither result establishes performance for longer videos, other devices, other targets, or HDR sources. Read our testing methodology, compare the quality-mode cases, or try your own source.

Apply the measurements to your sharing task

Use the Discord compressor, email attachment compressor, or 8 MB target page for a decimal MB target. These September 23 tests used MiB targets; they do not establish acceptance by a sharing platform.

For preparation and size checks, read the email guide or iPhone and HDR preparation guide. The separate 9 MB and 19 MB comparisons were encoded locally on September 30 and exclude website upload and queue time.