Production Economics
Cost per Accepted Film Is the AI Video Metric That Matters
Film evidence
Price the accepted export, not an isolated generation
A three-minute historical film consumes more than raw model seconds. Research, narration, still candidates, motion attempts, score, assembly, review, failed operations, and human correction all contribute to the real production cost.
Use this export as the unit in an economics worksheet. Record settled provider spend, retries, review time, rejected output, and whether the final cut is actually publishable. A cheap first generation is irrelevant if the accepted film requires extensive replacement work.
Omaha Beach: The Reality of D-Day · 03:01 · Full generated cut. This is a finished first-party Onira production, not customer proof or archive footage.
View the full film and production notesWhat to watch for
- Number of distinct production stages represented
- Likely correction points visible at sequence level
- Whether the finished runtime delivers enough value to publish
AI video products are often compared by subscription price, credits, generations, or an estimated cost per minute.
Those numbers are useful inputs. They are not the final production metric.
A documentary creator needs to know the cost per accepted film.
Generated output and usable output are different
A provider can return a technically successful asset that the film cannot use.
The scene may contain:
- identity drift;
- historical error;
- malformed objects;
- misleading text;
- unusable motion;
- the wrong emotion or framing;
- a transition that breaks continuity;
- a beautiful image with no relationship to the narration.
The generation consumed money and time, but it did not advance the accepted production until the shot was repaired or replaced.
Human correction is a variable cost
The creator’s time is part of the production economics.
Record:
- brief preparation;
- source review;
- story correction;
- pronunciation and narration work;
- prompt adjustment;
- take selection;
- regeneration;
- continuity repair;
- caption correction;
- final review;
- external editing;
- support time.
A workflow with a lower provider bill can be more expensive when it requires substantially more active correction.
Reservation and settlement should be separate
A production system can show an expected-to-maximum preflight estimate, reserve the maximum allowed amount, and settle against actual operations after the run.
That is different from promising a deterministic quote in a probabilistic system.
The creator should understand:
- what assumptions produced the estimate;
- the maximum amount that can be charged for the run;
- what provider operations actually occurred;
- what was retried;
- what was restored after a failure;
- the final settled usage.
Clear accounting is part of trust.
Compare workflows with one brief
To compare Onira with another product or a manual stack, use the same:
- brief;
- source boundary;
- audience;
- target duration;
- quality standard;
- disclosure requirement;
- acceptance rubric.
Then record total provider or subscription cost, active human time, elapsed time, failed assets, regenerations, external tools, and the complete final output.
The most important denominator is not rendered minutes. It is accepted films.
Publication and repetition complete the picture
An accepted file that is never published may still have been a useful proof, but it is not the recurring customer outcome.
Track:
- completed production;
- usable first cut;
- accepted or downloaded film;
- published film;
- second paid production.
The strongest unit economics come from a segment that repeatedly turns production spend into work the creator publishes.
This is also why a cheap volume tool and an opinionated documentary studio should not be compared only by file count. They optimize different outcomes.
The honest answer early is a range
Before a meaningful customer cohort exists, Onira should publish assumptions and ranges, not universal savings claims.
Realized cost varies with duration, quality profile, provider behavior, retries, correction, and the acceptance standard. A transparent distribution from actual cohort projects will be more useful than a headline percentage borrowed from unrelated traditional productions.
The job now is to measure the operation carefully enough that pricing, packaging, and marketing can follow reality.
See the AI documentary quality-control checklist, the complete production workflow, and the comparison center.