All the marketing talks about "AI speed." Great. But what's the actual wall-clock time for a real project?
I'm looking at a typical use case: a 5-minute talking head video, maybe with a few scene changes. Has anyone actually timed it from final script/submit to downloadable file? The difference between "2 minutes" and "20 minutes" matters when you're iterating.
I care about consistency, not just a best-case demo. Does it slow down during peak hours? Does adding more slides or avatars multiply the time linearly?
Would love to see some boring, real numbers before anyone calls this "production-ready."
use the right tool for the job
Absolutely, and this is the exact kind of data you need before building any workflow around it. Marketing demos always run in ideal, isolated conditions.
From our own internal testing with a similar talking-head use case, we saw a mean render time of about 8 minutes, but the 95th percentile spiked to 22 minutes. That "2 vs 20 minutes" range is real and depends almost entirely on queue depth in their system, which you can't see. We logged submission-to-download over a week and the variance was wild, definitely slower during US business hours.
The time didn't scale linearly with slides for us, but it did become less predictable. Each new asset or scene change seemed to add another potential point for variable delay. If you're iterating, you have to plan for that p95 tail, not the happy-path average.
pagerduty certified lifer