One credit, one clip: why Rush pricing is flat
Per-minute and per-token meters cannot be budgeted before a shoot, so Rush charges one flat credit per clip and carries the variance.
You should know what an AI pass costs before you press go. Not an estimate, not a range, a number. That requirement sounds modest, and it eliminates nearly every pricing model the AI industry has settled on. It is why Rush pricing is flat: one credit analyzes one clip, up to fifteen minutes, and the app shows the total before anything is submitted.
You cannot budget a shoot in tokens
Saturday game, 412 clips on the cards, and the producer asks what the analysis pass will cost. On a per-token meter the honest answer is "it depends": on how many frames get sampled, on how verbose the model felt like being, on how many retries the difficult clips ate. "Somewhere between forty dollars and four times that" is not a line item a producer can carry, so in practice nobody prices it, the bill arrives as a surprise, and the surprise gets remembered longer than the results.
Per-minute meters are at least computable in advance, but now you are summing durations across 412 clips at 11pm, and the meter punishes exactly the clips that need analysis most: the long, unlogged ones you were never going to scrub by hand. Either way you are pricing the model's behavior or the camera operator's trigger discipline, not your job.
Meters change how you work
A variable price on every clip turns logging into rationing. You analyze the A camera and skip the B-roll, run half the day and promise yourself the rest later, and the archive ends up partially indexed, which is to say it is still the old archive: the one you scrub through by hand. The whole point of an AI pass is coverage. A meter that makes you stingy about coverage is working against the product it is attached to.
Flat pricing makes the count the price. The grid says 412 clips, so the pass costs 412 credits, and the app says "412 clips → 412 credits" with your balance next to it before you commit. Anyone on the team can price a job by counting, which is a thing people can do in advance, standing in a hallway, without a spreadsheet.
The honest margin story
Some clips cost us more to process than the credit brings in. A fourteen-minute 4K clip is far more work than a twenty-second insert, and both are one credit. We priced the average with headroom, so across any real batch it evens out: we lose a little on the monsters, keep a little on the inserts, and the fifteen-minute cap keeps "one clip" from meaning "one feature film". This is not generosity. It is the ordinary bet behind every flat rate you have ever liked, and we would rather carry the variance than make you model it per clip. The cap has a workflow answer too: for the three-hour event tape, promote the segments that matter into clips of their own and spend credits on those, not on the dead air. Failed or cancelled clips refund automatically, because a credit buys a result, not an attempt.
What flat does to trust
The numbers, on the record: a one-time 25-credit trial, Pro at $20 a month with 150 credits, Studio at $49 a seat with 500 pooled per seat, and packs at 100 for $12, 500 for $50, and 1000 for $90 that never expire. Two principles hold the scheme together. First, the cost shows before every run, every time. Second, you pay to go faster, never to get your own work back: browsing, logging, on-device AI, sidecars, and every export are free without an account, so credits buy speed and larger models, not access to your own metadata.
The trust effect is the point. A meter you cannot predict is a meter you learn to fear, and nobody builds a workflow on something they fear. Flat is not the cheapest possible price on every batch; it is the one you can defend in a budget meeting.
How to budget AI work on any tool
The test travels beyond us. Ask one thing of any metered AI product: whether you can compute the bill from something you can count before pressing go. Clips, you can count in the grid. Minutes, you can sum, tediously, if the tool shows durations. Tokens, you cannot know until the model has already run, which means the price is whatever it turns out to be, which means it is not a price, it is a receipt. If the vendor cannot quote the number up front, the meter is doing your budgeting for you, after the fact, at its own convenience. Run the same test on your own rate card while you are at it; clients budget shoots too.
A price you learn after the run is a receipt; budget your shoots on numbers you can count before you press go.