Every finished render gets a score from 1 to 5. Four and five mean usable. Three means it exists. The keeper rate is the share scoring four or better.
Free · no account · recomputed from live data
What a usable clip actually costs
Every studio publishes a price per render. Nobody publishes the price per render you would keep — and that is the number that decides your bill. This page publishes Saga's.
Built from Saga's own finished renders, each scored 1–5 by hand, priced live at the settings you choose below. Every figure carries the number of renders behind it.
Headline finding
Reading the rated renders…
This page recomputes on every load. Nothing here is typed in by hand.
The board
Loading live prices…Video
| Model and frame | Rated | Keeper rate | Per render | Per keeper |
|---|---|---|---|---|
The highlighted row is the lowest cost per keeper on this board — which is not always the lowest cost per render.
Image
| Model | Rated | Keeper rate | Per render | Per keeper |
|---|---|---|---|---|
Image renders bill on the exact output size, so the per-render figure is the current quote for the size selected above.
Live at /api/render-odds — the same numbers this page reads, with no key required.
Project estimate
Budget the clips you need, not the renders you buy.
Nobody sets out to buy renders. They set out to finish a piece with a certain number of shots in it. Say how many you need and the keeper rate does the rest.
Render at exactly these prices.
Create a free account
A new account starts with a free balance.
How this is measured
Where every number on this page comes from.
A model or frame shape with fewer than ten scored renders is left off entirely rather than shown with a wide error bar. Two earlier readings of this data were withdrawn for exactly that reason.
The per-render column is the current customer price for the settings above, from the same source that quotes a real render. Nothing on this page is a stored figure that can quietly go stale.
What this is not. These are one operator's ratings of Saga's own renders, not an independent benchmark and not a claim about anyone else's platform. Taste is part of it: a render Saga scored 2 might be exactly what you wanted. Cost per keeper is an expectation drawn from past outcomes, not a promise about your next render. The honest way to read the table is as a starting point that is measured rather than asserted — and the sample sizes are printed so you can discount it yourself.
Why the board splits by frame shape and not by resolution. There are enough ratings to break these renders down by resolution too, and doing so would produce a tidy-looking table showing 480p at a fraction of the cost per keeper. It is deliberately not published, because the rating answers “was this a good take?” and resolution mostly answers “can I deliver it?” — two different questions. A 480p clip can be a perfect take and still be unusable for the job it was made for, so a per-keeper figure that rewards dropping resolution would be measuring the wrong thing and quietly recommending the wrong choice. Frame shape is different: it changes what the model composes, which is exactly what the rating judges. The same restraint applies to clip length.
Questions
Asked plainly.
Why does vertical cost more per usable clip — is the model worse at 9:16?
No. Saga tested it rather than assuming: three vertical renders across three models, on a prompt written for a tall frame, all executed correctly — including the model the ratings appear to blame. The measured gap reflects what was shot vertically — wide, multi-subject, lateral choreography that needs width — not a limitation of the model. Compose for the tall frame and the penalty is not one you inherit.
Why publish a number that makes some of your own renders look bad?
Because the alternative is a price list that quietly understates what a project costs. Nineteen percent of Saga's rated renders scored 1 or 2. Any studio's would show something similar; most simply do not count. A price you can only verify after you have spent it is not a price you can compare.
Does a low keeper rate mean the model is bad?
Not on its own. It means that at these settings, on this kind of work, more attempts were needed. A cheap model with a low keeper rate can still be the cheapest route to a usable frame — the per-keeper column is there precisely so you can check rather than assume, and on the current data it sometimes comes out that way.
Can I see the underlying renders?
No. Renders belong to the accounts that made them and are never published. What leaves the database is counts and averages: model, frame shape, how many were rated, and how many were kept.
How often does this change?
The odds change as renders are rated; the prices change whenever the underlying model rate does. The page recomputes on load and caches for ten minutes.