A shot came back wrong. Because a lot is a list, fixing it costs one shot, and the take you are replacing stays exactly where it is, because you paid for it.
But before you spend anything, two things are free and one of them is the most common avoidable charge in this product.
First: did it actually fail?
Ask the provider again before paying to do it twice.
A take that timed out on our side may have been delivered on theirs. Re-checking costs nothing and asks the provider once more whether the clip exists. Re-rendering pays for the same clip twice.
Second: is this actually a crop?
“I need this in 9:16” is usually not a re-render at all. Delivering a lot in another aspect is ffmpeg, free, instant, and stored, so a second delivery of the same aspect costs nothing either.
It is honest about what it is: a centre crop with a chosen focus, not subject tracking. That is good enough surprisingly often, and when it is not, it is not. A 16:9 frame cropped to 9:16 keeps under a third of the picture, which turns a two-person shot into a one-person shot. The paid version paints new edges with a model instead. Reach for the free one first and look at it.
Spoken shots have no picture, so they are skipped, and named rather than silently dropped.
Rolling again, versus fixing what you have
These are genuinely different decisions and the right one depends on what is wrong.
| What is wrong | Reach for |
|---|---|
| The performance, the framing, the whole idea of the shot | Render it again. There is nothing here worth keeping |
| The grade, but the performance is right | Restyle it, motion, pacing and performance survive exactly |
| One object is wrong | Edit it by sentence. The people, their clothes, the setting and the timing are kept |
| One thing needs to be gone | Erase it by name, and the background is rebuilt behind it |
| It is too small | Upscale it |
Re-rendering throws away the take you have and rolls the dice again. A fix keeps it and changes one thing. Neither is always right, sometimes a fresh render genuinely is cheaper, and the quote for one is the number a fix is competing against.
Every fix lands as another take
This is the part that makes the whole thing safe. A restyle, an edit, an erase, an upscale: each one lands as another take of the same shot, beside the original. The two can be compared. Mount whichever wins. Nothing you paid for is replaced.
The generative reframe is the one exception, and it is worth knowing: its result lands as a delivery format, where the free crops land, rather than as a take. So an export in that aspect picks it up and the shot’s own takes are untouched.
Read the tier before you pick the pass
Every paid fix runs on a tier, and the row you land on decides three separate things, none of which is obvious from the name of the pass.
| The row decides | Why it bites |
|---|---|
| Whose key pays | A workspace can hold a key for one tier and not another. Picking one you have no key for is refused by name before anything is queued |
| How much of the clip survives | The default window returns about five seconds whatever you give it. A twelve-second take comes back cut |
| What the price is per | Source second, output megapixel, or started second. They do not scale the same way |
Read the windows and prices off the rows rather than from any sentence, including this one. A row is current; a sentence goes stale.
Three things about price that change what you choose
- Upscaling bills on the output. For stills that is per output megapixel, so doubling the factor quadruples the price. The tiers are also not the same product: one is faithful, one invents detail, one reinterprets.
- Erasing has a flat price, the same for a two-second clip and a twenty-second one, which nothing else paid here does. On a long take that makes it the cheapest pass by a wide margin.
- Anything never measured cannot be quoted. A take whose length or dimensions were never recorded comes back unknown rather than priced, and one tier refuses such a clip outright rather than submitting it and letting the provider decide.
Every one of these can be priced without spending. Do that, compare it to a fresh render, and then decide.
Say what you mean, once
Erasing is the only pass in the product driven by free text about the footage itself, so the sentence is the whole instruction. Name one thing, the way you would point at it: “the man in the red coat”, “the logo on the wall”.
And describe what should be there once it is gone. That part is optional and leaving it out is worse than it looks: the fallback is the shot’s own prompt, and that prompt describes the shot with the thing still in it. The fallback is actively the wrong description.
Editing an object holds best between objects of similar size and place. Where shapes differ a lot, the new object can grow out of the old one, which is a re-roll, not a loss, because the original take is still there.