# Media editing examples: method and limits

Measured October 22, 2026. These examples demonstrate a specific edit and export. They are not general quality scores, product rankings, or promises about other files.

## Video compression

Source: QuestStudio's existing eight-second cinematic portrait example, `/static/blog/uploaded-videos/cinematic-push-in.mp4`.

- Source SHA-256: `8824f58197a6882ec9f7ec18a186e7616dddce25ac073cdec1a07d58d01d0f62`
- Source: H.264, 720 × 1280, 24 fps, with stereo AAC audio.
- Encoder: FFmpeg 7.1, libx264, medium preset. Frame size and frame rate unchanged.
- Audio: stream-copied in both exports. Encoded audio payload hashes matched the source.
- No file-size target was set. These are constant-quality experiments, separate from the guide's hypothetical 20 MB calculation.

| Copy | Bytes | Decimal MB | Reduction | Full-clip SSIM against source |
|---|---:|---:|---:|---:|
| Source | 2,281,398 | 2.281398 | Reference | Reference |
| CRF 26 | 1,074,667 | 1.074667 | 52.9% | 0.985213 |
| CRF 40 | 338,981 | 0.338981 | 85.1% | 0.946827 |

SSIM compares aligned decoded frames. It is not a percentage of perceived quality and does not certify an acceptable delivery. All three files decoded without errors. The article's comparison uses the same 300 × 300 crop at 4.000 seconds, starting at pixel x=170, y=420. Crops were exported as lossless WebP without resizing, sharpening, or selective enhancement. The smaller CRF 40 copy shows softer texture and coarser shading in this crop. Inspect motion and your own intended viewing size separately.

Reproduction commands, using the source as `source.mp4`:

```text
ffmpeg -i source.mp4 -map 0:v:0 -map 0:a:0 -c:v libx264 -preset medium -crf 26 -c:a copy -movflags +faststart crf26.mp4
ffmpeg -i source.mp4 -map 0:v:0 -map 0:a:0 -c:v libx264 -preset medium -crf 40 -c:a copy -movflags +faststart crf40.mp4
ffmpeg -i crf26.mp4 -i source.mp4 -lavfi "[0:v]setpts=PTS-STARTPTS[a];[1:v]setpts=PTS-STARTPTS[b];[a][b]ssim" -an -f null -
```

Repeat the last command for CRF 40. Output bytes can differ with another encoder build. See the [FFmpeg encoding documentation](https://ffmpeg.org/ffmpeg.html) and [SSIM filter reference](https://ffmpeg.org/ffmpeg-filters.html#ssim).

## MP3 edit

The speech was synthesized locally with the installed Microsoft David Desktop voice through Windows SAPI. It is not a recording of a person, a cloned voice, an Audacity test, or a QuestStudio generation. The script contains a false start, two copies of the first intended sentence, then the second intended sentence. Deliberate pauses were inserted between segments and at the end.

- Format of both MP3s: 128 kbps, 22,050 Hz, mono.
- Source: 320,199 bytes; decoded duration 19.920136 seconds.
- Edited result: 141,313 bytes; decoded duration 8.737551 seconds.
- Source SHA-256: `e844451f410a52fb3354b338978f997de5caa9fd97d7b871d0b09d65c6c772a6`
- Result SHA-256: `7a21b1cebd37b5fa108097fcefea4f134a244308e6e7b1c2b5c988b1f802d431`

The source MP3 was decoded once to 16-bit PCM. The following sample ranges, with exclusive end points, were retained in order:

| Start sample | End sample | Meaning |
|---:|---:|---|
| 108571 | 191451 | First intended sentence |
| 191451 | 202476 | Half-second gap |
| 296381 | 395139 | Second intended sentence |

Those ranges were joined in PCM and encoded once to the final MP3. The pre-export PCM matched the selected samples byte for byte. Decoding the final MP3 produced the expected 192,663 samples. This checks selection and duration; it does not establish unchanged sound after lossy encoding or an inaudible transition. Some players display rounded or differently compensated MP3 durations. Use the supplied transcript and listen across the join before judging the result.

Transcript of the edited file: “Keep the original recording before you make any changes. Export the final file once, then open it in another player.”

There has been no subjective listening-panel assessment. The example intentionally keeps its source sample rate and one pause; it is not a recommendation to downsample another recording or remove every breath.