Video compression testing methodology
How our compression cases are measured: source files, settings, byte sizes, timing, matching frames, and limitations.
Last updated: 2026-10-23
Find the evidence for your question
We publish four sets of measurements. The September 16, September 23 and October 23 sets used the live website; the September 30 Discord comparison used local FFmpeg. The dates below are the test dates, not the date this page was updated.
How much smaller can a video get?
See the September 16: three live quality-mode cases: a screen recording, a downloaded phone clip, and a six-minute film excerpt. The 66.9% reduction in the film case includes a resolution decrease from 1920 × 818 to 1276 × 544. It does not establish a typical reduction or lossless quality.
Did the downloaded files stay under their targets?
See the September 23: two live target-size cases: the downloaded files were 2,012,378 bytes against a 2,097,152-byte target, and 19,803,861 bytes against a 20,971,520-byte target. Each input was tested once. These MiB targets do not establish acceptance by an email or chat service.
What happened with decimal 10MB, 19MB and 25MB targets?
See the October 23: three live decimal-MB cases. The downloaded results were 9,484,147 bytes for a 10MB target, 18,546,631 bytes for 19MB, and 24,303,584 bytes for 25MB. Each input was submitted once. The screen recording was looped to 240 seconds without re-encoding; the film input was a continuous prepared five-minute excerpt. The film output also used a lower resolution. These runs do not establish an 8MB result or acceptance by an email or chat service. Read the exact measurements and preparation and license notes. Upload, queue, encoding and total times were not independently captured; missing times remain null.
What changes between a 9 MB and 19 MB target?
See the September 30: local Discord size comparisons: three inputs at two targets, with four encoded outputs and two cases where the original already fit. The comparisons show size, resolution, and matching frames. Local encoding times exclude website upload, queueing, and delivery; no Discord upload acceptance was tested.
For a decimal upload limit, use the 8 MB target page, email compressor, or Discord compressor. A 20 MiB target equals 20,971,520 bytes, which exceeds a decimal 20 MB limit of 20,000,000 bytes. A target is a byte ceiling; it is not a promise about visual quality or platform acceptance.
What a measured case means
The homepage cases are individual runs through the live compressor. Each lists the source, settings, date, input and output size, and measured times. They are not averages, independent reviews, or guarantees for other videos.
A camera-original file, a downloaded copy, and a prepared excerpt are different inputs. Our cases identify which was used. The size before compression always means the file actually uploaded, including any preparation described in the source notes.
Measurements and quality comparison
- Identify a source that can legally be shared, record its origin and license, and inspect its duration, dimensions, codecs, audio, and dynamic range.
- Submit it to the live compressor and retain the task's actual settings and result metadata.
- Download the result, verify its bytes and media properties, and compare them with the task record.
- Compare source and result frames at the same media time. Publish the settings and limitations with the result.
We calculate size reduction as (1 − result bytes / source bytes) × 100%. MiB uses 1,048,576 bytes; decimal MB uses 1,000,000 bytes. A target is an upper bound, not a requirement to fill every available byte. For sharing, leave room below the receiving service's limit.
Matching still frames help inspect text and fine detail, but do not establish motion or audio quality. Reducing resolution can account for part of the saving. A quality-based run does not prove performance in target-size mode.
Timing and compression credits
Encoding time is the server's encoding measurement. Total task time also includes preparation, waiting, validation, and delivery; download time is excluded. Network conditions and server load affect these values. Where upload time is reported, it is measured separately.
Compression credits depend on the source and chosen settings. A case's recorded consumption applies to that run, not to every video of the same duration. The upload area shows the current estimate before you start.
Public records and source permissions
The September 16, 2026 cases include an original test-page recording, an iPhone 7 recording downloaded from Wikimedia Commons, and a six-minute excerpt of Sintel. Read the measurement data and preparation, timing, and license notes. The longer case uses a lower output resolution; the phone case is not a camera-original file.
We retain failed runs in our test records. A successful shorter excerpt does not establish that a full-length input succeeds. Results and source links on the homepage let you inspect the published successful runs directly.
The September 23, 2026 target-size tests reuse the screen and phone inputs with 2 MiB and 20 MiB targets. The results include verified download sizes, consumed credits, matching frames, and output hashes. Their total task time is measured from server task creation to its completed update; it excludes local checks before task creation and downloading.
The corresponding September 23 measurement data preserve the live-run records; the results page explains the inputs and checks. The separate September 30 measurement data and local method and source permissions document the Discord comparison. Reusing a source in another mode or at another target does not add an independent source to the sample. We do not combine these sets into a success rate or a speed average.
Using these results
Choose a case close to your source and settings, then compare your own downloaded result before sharing it. Use pricing for current plans and About for editorial responsibility and corrections.