Home / Benchmarks
Measured, not claimed.
These are measurements from development recordings and exported sample files. Competitor promotional footage is not a controlled test of their recording engines. Results depend on the source, export settings and hardware.
| Measurement | Result | How |
|---|---|---|
| Network connections during a full licensed export | 0 | Every TCP connection from LaunchLoop processes was logged during a complete export. |
| Time to render a 34-second launch trailer | ≈ 70-110 s | 1080p, 30 fps, offscreen frame capture plus encode, on an RTX laptop. |
| Shots planned on a 3-minute web tutorial with no cursor data | 10 | From picture activity alone, ignoring burned-in captions. |
| 3-minute tutorial cut to a trailer | 17 s | Product Trailer pass, quality score 100/100 after repair. |
| Formats from one recording | 4 | 16:9 film, native 9:16 film, a looping X clip per shot, GIF. |
| Camera jerk on auto-zoom (lower is smoother) | 9.4 vs 11.5 | LaunchLoop vs Screen Studio's published demo videos, same optical-flow metric (median zoom jerk). |
| Stuttered frames in our launch trailer | 0% | Frames that repeat while the picture moves. Measured the same way: Canvid's promo 5.8%, Arcade's 5.2%. |
| Launch film loudness | -13.3 LUFS | Social platforms normalise to about -14. Canvid's promo measured -20.6, Arcade's -22.2, Supademo's -17.0. |
| Launch film frame rate | 60 fps | Frame-exact page capture on virtual time, with 3-sample motion blur in Hype films. |
| Automated tests in the app | 4,400+ | Unit tests run on every change. |
How to interpret these results
Camera jerk and repeated-frame measurements describe the specific sample files, not every recording a product can make. Competitor source footage, codecs, frame rates and editing differ. Loudness is a file property, not a quality score. A zero-connection local export does not cover activation, model downloads or optional AI writing.
The sample trailer renders and tutorial tests used different inputs and settings. They should not be combined into a universal speed claim. Test a short project on your own hardware before estimating a deadline.
The privacy test
We ran a complete, licensed export and logged every TCP connection opened by any LaunchLoop process. The count was zero. This test covered local exporting; optional AI tools, model downloads and licence activation can make network requests.