Discord 9MB / 19MB comparison — measured 30 September 2026

Scope
Local FFmpeg encoding on macOS arm64, FFmpeg 8.1.2. These are not live
website runs, production latency measurements or Discord acceptance tests.
No user videos, paid processing or uploads were used.

Method
- Same complete source for both targets; no bitrate padding or duration changes.
- Targets are 9,000,000 and 19,000,000 bytes (decimal MB).
- Shared product functions: planAdaptiveResolution (adaptive-v1),
  planTargetSizeBudget and targetSizeCorrectionBitrate.
- MP4 / H.264 / yuv420p, x264 veryfast, two-pass, two encoder threads.
- Preserve source frame rate; choose resolution with adaptive-v1. Lanczos
  scaling and square pixels. AAC mono 96 kbps / stereo 128 kbps for these inputs.
- 96% media budget reserves room for container overhead. If the first encode
  exceeds the cap, one corrective two-pass encode is attempted. Both game
  outputs needed correction; neither phone output did.
- Every encoded result is measured in exact bytes, checked against the cap,
  and fully decoded with FFmpeg -xerror. H.264 and AAC tracks are checked.
- Source and output SHA-256 hashes are in measurements.json.
- One run per target. Encoding seconds include both passes and size correction,
  exclude probe, validation, SSIM calculation, screenshots, upload and download.
  They are local measurements affected by concurrent work, not a speed claim.
- SSIM is computed from decoded videos, with the result upscaled to the source
  dimensions, square pixels and timestamps reset before comparison. It is a
  structural metric, not a perceptual quality score or quality guarantee.
- Snapshots: game 30s, phone 10s, screen 24s. JPEG q:v 2 at decoded dimensions;
  no sharpening, color edits or other enhancement. The page resizes them for
  display. Click to inspect full dimensions. JPEG compression can itself lose
  detail. Stills cannot demonstrate motion, sound or audio/video synchronization.
- The H.264 MP4 screen source already fits both limits; no re-encoding is done.
  Its original bytes and frame are repeated for comparison, not claimed as
  newly compressed results. A fitting file can still have playback constraints.

Sources and licenses

Game: Benchmark SuperTux Kart Base Profile 2025.07.04 - 10.19.10.02.webm
SuperTuxKart authors. Recording uploaded by PantheraLeo1359531.
https://commons.wikimedia.org/wiki/File:Benchmark_SuperTux_Kart_Base_Profile_2025.07.04_-_10.19.10.02.webm
Uses Wikimedia's 1920x1080 VP9/Opus derivative, 70.834s, 60fps, not the
8K original. The source page supplies game credits and licensing information.
Selected license: Creative Commons Attribution-ShareAlike 4.0 International.
https://creativecommons.org/licenses/by-sa/4.0/
Changes: recompression to H.264/AAC MP4, downscaling to 852x480, and JPEG
frame extraction. These game derivatives are distributed under CC BY-SA 4.0.
Attribution does not imply endorsement by the game authors or recorder.

Phone: La rivière Druyes et la rue du capitaine Coignet à Druyes-les-Belles-Fontaines
Benoît Prieur; the Commons description identifies iPhone 7 as capture device.
https://commons.wikimedia.org/wiki/File:La_rivi%C3%A8re_Druyes_et_la_rue_du_capitaine_Coignet_%C3%A0_Druyes-les-Belles-Fontaines.webm
CC0 1.0: https://creativecommons.org/publicdomain/zero/1.0/
Uses the existing public case-study WebM copy, not a camera-original file.
Changes: H.264/AAC recompression, adaptive scaling (720p at 9MB; original
1080p at 19MB), JPEG frame extraction.

Screen: original test webpage recording with fictional data and scrolling.
Source prepared for the 16 September 2026 website case study, no customer data.
https://assets.video-size-reducer.com/case-studies/2026-09-16/screen-source.mp4
No re-encoding. JPEG frame extraction at 24s.
Original recording method:
/case-studies/2026-09-16/methodology.txt

Reproduction
Repository script: scripts/benchmark-discord-targets.ts
Input paths and download commands: docs/benchmarks/discord-targets-2026-09-30.md
Exact measurements: measurements.json (same directory as this file).
Full command log and original output videos remain locally under
 data/discord-comparison/; videos are not bundled into the website.