AI video model index
AI video model index: render time and success rate by model (October 2026)
Last updated
The Clip Studio AI video model index is a monthly table of how AI video models behave in production: how long a clip takes from submit to finished, how often a generation finishes, and how long the clips people make are. It is built only from real generations in Video Lab, not from vendor benchmarks or our own test prompts.
First edition landing in a follow-up. This page publishes the method first. The October 2026 table goes here once the data pull below has run against production. Until then there are no numbers on this page, by design: we do not estimate, round in from vendor claims, or fill rows by hand.
Key takeaways
- Measured, not claimed. Every figure comes from completed and failed Video Lab generations in a 30-day window.
- Render time is wall-clock. It runs from the moment the job is submitted to the provider until the finished clip is ready, so it includes queue time.
- Success rate is completed ÷ finished jobs. Jobs still rendering when the data is pulled are left out.
- Small samples are not published. A model needs at least 20 completed generations in the window to get a row.
- Volumes are bucketed. We publish ranges (20–49, 50–99, 100–249 and up), not exact counts.
- Internal traffic is excluded. The operator account, staff mailboxes and throwaway mailboxes do not count.
What does the AI video model index measure?
For each Video Lab model it reports four things over the last 30 full days: how many generations completed (as a range), the success rate (completed divided by completed plus failed), the median and 75th-percentile time from submit to completed, and the median length of the clips that completed. It also notes the resolution people chose most often for that model, because resolution changes render time.
How is render time calculated?
Render time is the wall-clock gap between the job's submit time and its completed time, in seconds. When a job never recorded a separate submit time, its creation time is used. The median is the middle job; the 75th percentile is the time three in four jobs finished within. Queue time at the provider is included, because that is the wait a creator actually sees.
Which generations are included?
Included: completed and failed Video Lab generations created in the window, for models in the live AI video models catalog, across text-to-video, image-to-video, first and last frame, and reference modes. Excluded: Edit clip jobs (they restyle an existing video, so their timing is not comparable), jobs still rendering at pull time, the unlimited operator account, any clipstudio.ai mailbox, and disposable or watchlisted mailboxes we already block from free generations.
Why are some models missing from the table?
A model gets a row only with at least 20 completed generations in the window. Below that, one slow or failed job swings the median and the success rate too much to be useful. Newly added models usually appear in the edition after launch, once they pass the threshold.
How often is the index updated?
Monthly. Each edition covers the 30 full UTC days before the pull date, and the pull date is printed at the top of the page as the last-updated stamp. Earlier editions are replaced, not edited, so every row on the page comes from the same window.
Does the index include Topic Shorts?
Yes, as one extra line. Topic Shorts is a full pipeline (script, voiceover, stock footage, captions and music) rather than a single model call, so it is reported separately as the median and 75th-percentile time from created to completed, with the same window, threshold and exclusions.
Methodology
The aggregation is read-only SQL kept in the Clip Studio repository and run once per edition. It reads generation status and timestamps only. It never reads prompts, images, or who made a clip.
October 2026 table
First edition landing in a follow-up. The table will list each model with its completed range, success rate, median and 75th-percentile render time, median clip length and most used resolution, followed by short observations drawn only from those numbers.
Sources
- Clip Studio Video Lab and Topic Shorts generation records, aggregated as described in Methodology. No third-party figures are used on this page.
