Measured, not estimated
How long does it take to make an AI video?
A topic becomes a finished vertical video in a median of 217 seconds, measured across 52 production runs. Ninety percent finish within 262 seconds. The fastest completed in 71 seconds; the slowest took 391.
These are our own production figures, read from the database on 3 September 2026. They describe Reelyr, not the category — we have not benchmarked anyone else, and we are not going to publish numbers for software we did not measure.
Which part of AI video generation is slowest?
Image generation is the slowest stage at a median of 90 seconds, followed by rendering at 61 seconds. Research and script writing together account for under 40 seconds. Generating pictures, not writing words, is what you wait for.
| Stage | Median | Mean | Runs | What happens |
|---|---|---|---|---|
| Images | 89.9s | 74.5s | 60 | Ten to thirteen 2K stills, generated in parallel |
| Render | 61.3s | 69.1s | 54 | Composited to 1080x1920 H.264 on AWS Lambda |
| Script | 26.7s | 25.6s | 113 | Narration, shot list and caption timing |
| Research | 12.8s | 13.7s | 114 | Live web search, facts gathered with sources |
| Direction | 9.3s | 9.6s | 114 | Format, voice and visual treatment chosen |
| Voice | 7.8s | 7.1s | 58 | Narration and per-word timings (concurrent with images) |
Why do some videos take four times longer than others?
Completed runs range from 71 to 391 seconds. Videos built from cards generate no images at all and finish fastest. Videos built from ten or more generated stills take longest. Shot count is the variable that moves the total.
How is this measured?
Every pipeline stage writes its own duration to a database row as it completes. These figures are read from that table directly, excluding dry runs and failed stages.
Voice is excluded from the end-to-end total because it runs concurrently with image generation — adding it would double-count time nobody waits through. Stage counts differ because not every format uses every stage: card-based videos generate no images, which is why that row shows 60 runs against 114 for research. A run counts as complete when at least five stages recorded a duration.
Frequently asked
- How long does it take to make an AI video?
- A topic becomes a finished vertical video in a median of 217 seconds, measured across 52 production runs. Ninety percent finish within 262 seconds. The fastest completed in 71 seconds and the slowest took 391.
- Which part of AI video generation is slowest?
- Image generation is the slowest stage at a median of 90 seconds, followed by rendering at 61 seconds. Research and script writing together account for under 40 seconds. Generating pictures, not writing words, is what you wait for.
- Why do some videos take much longer than others?
- The range is 71-391 seconds. Videos built from cards generate no images and finish fastest; videos built from ten or more generated stills take longest. Length and shot count drive the difference.
- How is this measured?
- Every pipeline stage writes its duration to a database row as it completes. These figures are read from that table, excluding dry runs and failed stages. Voice is excluded from the end-to-end total because it runs concurrently with image generation.