Analyze does not guess at highlights. It cuts the source into segments, transcribes them with our own Whisper service, captions each one with a vision model, and leaves you with a timestamped index you can search in plain language. Picking moments out of that index and exporting them as a short is the last and cheapest step. The expensive step is the first one, and this page starts with what it costs, because the number surprises people.
Analyze is the most expensive thing on this platform per minute of input, and it is the one lane where the free green wallet is no help at all. The frame pass runs on a vision model and is billed straight out of the gold wallet. There is no green rung, no partial payment, and no fallback.
Which plan: Starter and up. Below Starter the Analyze section shows a plan pitch instead of the workspace, and the billing settles it anyway: Analyze is charged 100% in ✦ gold, and signing up grants zero gold.
The frame pass costs about ✦0.07 per analysed frame at default sampling, and a moment search over the finished index is a flat ✦1. That works out to roughly ✦4 per minute of source. Set against the monthly gold grants, it looks like this:
| Plan | Gold a month | Analyze minutes a day | Video the grant covers |
|---|---|---|---|
| Free | none | 10 | none — no gold to spend |
| Starter | ✦50 | 60 | about 12 minutes a month |
| Creator | ✦380 | 120 | about 90 minutes a month |
| Pro | ✦900 | 300 | about 3.5 hours a month |
| Ultra | ✦10,000 | 900 | about 39 hours a month |
Read that table twice, because it says something the daily minute meter hides: on every plan below Ultra the gold wallet runs out long before the daily minutes do. Starter's 60 minutes a day is not a budget you will reach — ✦50 is. Buying gold credits is what actually extends Analyze, and buying credits does not change your tier.
Watch out: Minutes are charged up front when a run starts and are not refunded if it fails — a failed run still spent the vision calls. Start with a short source, or trim before you upload, rather than testing settings on a feature-length file.
Analyze is metered twice, in two different currencies, and both have to clear. The credit charge covers what the vision pass costs to run. The minutes-per-day allowance exists because the captioning rides a shared, non-elastic subscription rather than an elastic API — uncapped use would starve every other user of it, including chat. A moment search bills a flat two-minute equivalent against that allowance whatever the length of the video, because a search spends a handful of calls rather than scaling with the source.
Open Analyze — Opens the Analyze workspace. Nothing is analysed and nothing is charged until you press the run button.
The finished record is not a summary. It is a set of addressable objects, and they are what make the export step cheap:
That last point is the practical one. The expensive pass happens once per video. Everything afterwards — more searches, more exports, a different short from the same source next week — reads the cached index. Repeat queries on an already-analysed video skip the analysis entirely.
One setting decides both how good the index is and how much it costs, and it is the segment length. Segments can run anywhere from 0.2 to 600 seconds — it is a free-form number, not a ladder of presets — and the cost of the pass tracks the number of frames analysed, which tracks how finely you have cut the source.
| Segment length | Index granularity | Cost | Good for |
|---|---|---|---|
| Short (a few seconds) | Fine — a caption every few seconds | Highest | Fast-cut footage, finding brief moments |
| Auto | Chosen from the source | Balanced | The default, and the right first try |
| Long (a minute or more) | Coarse — one caption per span | Lowest | Talks and interviews, where the transcript carries the meaning |
The rule of thumb: if what you are looking for is spoken, go long and let the transcript do the work. If what you are looking for is visual and brief, go short and pay for it. Getting this wrong in the cheap direction is recoverable — you can re-run the pass — but re-running means paying for the pass twice, so it is worth thinking about for the thirty seconds it deserves.
The export cuts the segments you ticked out of the source and joins them in timeline order. What comes out is your source's aspect ratio, at your source's resolution — the export is a straight cut, not a reframe, which is precisely why it costs nothing.
That means a 16:9 source produces a 16:9 short. Reframing it to 9:16 is a separate job: crop it in whatever editor you use, or on Ultra, in the in-app Timeline editor, which does crop and reframe among much else. Do not plan on the Timeline from a lower plan — it is an Ultra-only section and is not shown below that tier at all.
A useful sequence for anyone doing this repeatedly: analyse once, ask three or four different questions of the same index at ✦1 apiece, export several candidate shorts from those answers for free, and only then decide which one is worth reframing and captioning. The economics of this feature reward asking a lot of questions of one expensive index rather than analysing a lot of videos.
No. Analyze shows a plan pitch below Starter, and it is billed entirely in gold credits, which a free account has none of. The green wallet cannot pay for any part of it.
At default sampling, roughly ✦4 a minute, so about ✦250 for an hour — two thirds of a Creator month's ✦380 grant spent on a single video. Longer sources are the case for buying credits, or for trimming before you analyse.
No. The export is a straight ffmpeg cut and join from the source video — no model call and no charge. The cap is 40 segments per export request.
The minutes are still spent, because the vision calls were made. Credits are deducted for the pass that ran. This is why the workspace quotes the cost before starting and why a short test source is worth the two minutes.
Yes. Picked segments can go out as one short, be staged as LoRA training clips, or seed a storyboard. The Timeline editor also accepts them, but Timeline is an Ultra-only section.