Benchmarks & observations
Development measurements from the technical paper—Apple Silicon macOS, indicative not formal benchmark suite.
Unique frames vs total
0 / 129
Smoke PDF deduplication
Full paper at 1080p
0s
~6.4 min of narration
Automated tests
0
Timing, mapping, camera
Max camera zoom
0×
Word-mode close-up
Conversion times
| Document | Resolution | Audio | Frames | Wall time |
|---|---|---|---|---|
| 1-page smoke PDF | 640×360 | ~4 s | 129 | ~2 s |
| 1-page smoke PDF | 1920×1080 | ~4 s | 129 | ~2 s |
| Academic paper (CSN) | 1920×1080 | ~6.4 min | ~11,458 | ~36 s |
Unique-frame deduplication
For a 4-second smoke PDF, only ~14 unique composites are rendered versus ~129 timeline frames at 30 fps—roughly an order-of-magnitude reduction in static-camera mode.
Parallel workers
OCR and frame compositing scale with CPU core count. AAC re-encoding overlaps with video rendering.
Camera mode trade-off
Smooth pan-and-zoom requires per-frame compositing and scales with total frame count—ideal for short demos, heavier on long documents.
Correctness
Eleven automated tests cover text indexing, cue timing, highlight mapping, and camera planner behavior.