Most tools in this category bill in credits, which is a way of not telling you what a cut cost. This one bills in dollars and seconds, because you are paying your own providers directly and there is no internal currency to convert.
Two things spend money.
Rendering a shot, and rendering a lot. Plus the paid fixes on a take you already have. Everything else is free, including every way of finding out what something will cost. Planning, creating, attaching references, choosing a model, reading a brief and pricing the whole campaign are all free, and you can do all of them and walk away having spent nothing.
Where the number comes from
Not from a table of averages. A quote is produced by the adapter that would actually run the job, the same code path that will submit it, priced for the thing being bought.
Which is not one shape. Video is usually per second, but not always the second you think: some rows bill per started second, so a 6.2-second clip is billed as seven. Stills are sized by token budget rather than by pixel height. Speech is per character. Some video edits bill per megapixel of generated video, width times height times frames, so length and resolution both move the number. And three rows bill in frames, which means the same clip costs twice as much at 60fps as at 30.
You do not have to know which of those applies. You do have to read the quote rather than extrapolating from a different shot.
The price you are shown is the price of what will be made
This is the part that makes the quote worth trusting, and it is easy to skim past.
A shot written for seven seconds, on a model that only makes four, six or eight, is quoted for eight, because eight is what you will get and eight is what you will be billed. A resolution tier the model does not offer is downgraded, and quoted at the tier that will actually run. The estimate follows the render rather than the request.
The brief flags every one of these on a line of its own, so a shot whose provider will hand back a different length, shape or resolution than it asked for says so before you commit.
Quote the real shot, not a sketch of it
A quote without your references is a clean quote for the wrong render.
Pricing describes a hypothetical shot, so it can only report problems with what you describe. Leave out the references and the one incompatibility that costs a real take, a face silently dropped by a model that takes no images, cannot be reported at all. The render succeeds, the price is right, and your creator is not in it.
Same for continuity: a shot chained to the one before it cannot run on a model that will not start from another shot’s last frame, and only a quote that knows the shot is chained can tell you.
Three tools people merge into one
| Question | Where the answer is |
|---|---|
| Can this provider do it, and is there a working key? | The provider list |
| What does it charge for this shot? | A quote. Nothing else prices anything |
| How has it behaved for us? | Provider health, from this workspace’s own jobs |
The confusion worth avoiding is the first two: capability is not price. A provider being able to make a ten-second 1080p clip with audio says nothing about whether it is the cheapest, the dearest, or the only one that can.
Health says nothing at all about providers you have never used. An absent row there is an unknown provider, not a bad one.
When a price cannot be known
Point a gateway at a model slug we have no catalogue row for and it will still submit, but the estimate comes back unknown, not invented.
That is deliberate rather than a gap. A wrong price corrupts three things at once: the dry run you decided on, the kept-versus-discarded meter on Usage, and your spend cap. A number that might be wrong is worse than no number, because you would act on it.
What happens next is your workspace’s choice, in Settings · Spend caps: refuse unpriced models, or let them run knowing their cost cannot be counted in advance.
The order, before any money moves
- Can this provider do it, and is there a key that has not just failed?
- What does it cost, and what will it change? Quote at least two candidates on the same shot rather than assuming direct is cheaper than a gateway, or the reverse.
- Read the incompatibilities. Say them out loud to whoever is paying.
- Is there room under the cap?
- Render one shot, look at it, and only then queue the rest.
Everything above step five is free, repeatable, and costs nothing to get wrong.
The list prices themselves
This lesson is about how a quote is built. What each lab actually charges, read off its own pricing page and dated, is on how much AI video generation costs, with the AI video cost calculator to run your own spec across every key, and Kling vs Veo cost for the two most-asked-about labs side by side.