Measured results · April 2026
In-Browser FFmpeg vs Upload-Based Editors
We timed the same export on a single M2 MacBook Air across the top upload-based editors. Results below are representative, not a scientific average.
Test configuration
| Hardware | M2 MacBook Air, 16 GB RAM |
|---|---|
| Browser | Chrome 130 (stable), default settings |
| Network | Wi-Fi 6, 300 Mbps down / 50 Mbps up |
| Tested | April 2026 |
Export time by task
| Task | This site | VEED | Kapwing | CapCut Online |
|---|---|---|---|---|
Trim 30s from a 5-min clip 1080p H.264, 180 MB Local uses FFmpeg stream-copy — no re-encoding. | ~2 s | 45 s (upload + server queue) | 55 s (upload + render) | 40 s (upload + queue) |
Compress 1-min 4K to CRF 28 4K H.264, 620 MB Browser runs x264 directly on the CPU; no size gate. | ~24 s | 2 min 30 s (rejected: 500 MB free cap) | Not possible (250 MB free cap) | 3 min (upload + queue + render) |
Convert MOV to MP4 iPhone MOV, 1-min 1080p, 180 MB | ~8 s | 1 min 20 s | 1 min 45 s | 1 min 10 s |
Export 4K 60 fps → 1080p 30 fps 4K 60 fps, 1 min, 980 MB | ~42 s | Not possible (500 MB cap) | Not possible (250 MB cap) | 5 min + forced watermark |
Extract audio from 10-min talk MP4 1080p, 1.1 GB | ~6 s | 3 min (upload dominated) | Not possible (over 250 MB cap) | 2 min 30 s |
Upload-based editor times include file upload, server queue wait, encoding, and download. Local times are wall-clock from pressing the export button to download being available.
Why the gap
- Zero upload time. A 1 GB file on a 50 Mbps uplink needs ~2.5 minutes just to leave your machine. Local processing skips that entirely.
- No server queue. Free-tier cloud editors share limited render capacity; peak-hour queues can add minutes on top of encode time.
- Your full hardware. Apple Silicon and modern laptops often have more encoding throughput than the shared VM a free-tier cloud editor allocates.
- No artificial caps. Upload editors enforce 250–500 MB free caps; over that your only option is pay. Locally the only cap is browser memory (~2 GB).
Reproduce these results
Every tool in the table below runs the same way we used for the benchmarks. The upload editor times were captured in back-to-back sessions on the same network so conditions match.