Every AI narration job has a limit on what it may spend. Who sets the limit depends on whose OpenAI and HeyGen accounts the job runs on.
On your own keys
On Free, and on Pro or Studio once you have saved your own keys, OpenAI and HeyGen bill you for what a job uses. Your "Most a job may cost (US dollars)" setting is the most Vistralo will let one job spend.
On the narration your plan includes
On Pro and Studio without your own keys, the job runs on Vistralo's accounts and you are not billed by OpenAI or HeyGen. Vistralo applies its own limit to each job. Your setting can lower that limit. It cannot raise it.
Setting it
Open Settings, go to "AI narration" and, under "Voice and spending", set "Most a job may cost (US dollars)". It starts at 1.00 and can be anything from 0.01 to 50. Outside that range the form says "Set a per-job cost cap between $0.01 and $50."
How it is applied
Before Vistralo sends anything to a provider, it estimates what the request will cost and compares that with the job's limit. The Settings page puts it this way: "A job whose estimate is higher is refused before anything is sent. The same request is never sent twice."
What it cannot promise
The cap is checked against Vistralo's estimate. On your own keys, the bill comes from the provider. The two are usually close and are not guaranteed to match. For a limit that cannot be exceeded, also set a budget inside your OpenAI and HeyGen accounts.
When a request's result is unknown
If a request is sent and no clear answer comes back, Vistralo does not send it again by itself, so that one job is not charged twice. The job stops and you decide whether to retry.
What costs nothing
Recording, uploading, trimming, sharing and a website analysis without a narration script do not call an AI provider.