Editorial Method
Why AI Filmmaking Needs Editorial Selection
Film evidence
Selection turns generated material into a character film
Forgotten Memories depends on a restrained sequence of accepted shots: routine cleaning, an abandoned station, a discovered drawing, and a shared emotional response. Any one of those images could exist as an attractive generation, but the film only works because the edit selects a causal progression.
Notice what is absent as well as what remains. The cut avoids extra locations, competing robot actions, and showy camera moves that would dilute the discovery. Editorial selection protects the emotional idea instead of displaying every usable candidate.
Forgotten Memories · 01:00 · Character short. This is a finished first-party Onira production, not customer proof or archive footage.
View the full film and production notesWhat to watch for
- Repeated robot identities remain legible across changing angles
- Every selected shot advances routine, discovery, or remembrance
- The quiet ending is stronger than an additional spectacle shot
Generation is productive because it creates options. It is dangerous when the production treats every option as output.
An AI film needs an editorial authority that decides what becomes true.
Candidates are not the film
An image request may return a composition with the right subject but the wrong object state. A motion attempt may preserve appearance but fail physics. A voice take may pronounce every word correctly while missing the intended performance. A music cue may be beautiful but compete with narration.
None of these candidates should become canonical merely because a provider returned success.
Technical success means the operation completed. Editorial acceptance means the result belongs in this film.
Selection creates production truth
When an editor accepts a character reference, still, motion take, voice performance, or cue, that decision affects downstream context.
The accepted character appearance informs later scenes. The accepted final frame defines a transition boundary. The selected voice track owns timing. The selected take determines where sound effects belong. The final timeline determines what the audience actually experiences.
This is why stable identifiers matter. A production must know exactly which candidate became canonical, especially after retries and repairs.
Dailies should answer explicit questions
Reviewing every generation without criteria creates fatigue. Each production phase should ask bounded questions.
For a still:
- Is the required subject present and recognizable?
- Does the setting match accepted state?
- Are composition and light appropriate for the beat?
- Are there critical anatomy, text, or object defects?
For motion:
- Does the intended action occur?
- Does the start match the accepted still or prior boundary?
- Does the end establish usable continuity?
- Are camera behavior, physics, and performance acceptable?
For the rough cut, the questions move to story, timing, repetition, geography, sound, and resolution.
Rejection needs a next action
A useful review does not only say no. It identifies why the candidate failed and which repair is permitted.
Some problems justify changing the exact request or prompt. Some require a different reference. Some indicate that the authored fallback should be used. Others reveal a story or rights issue that generation cannot solve.
The repair scope should remain bounded. A motion failure should not silently rewrite the story. A continuity problem should not authorize arbitrary changes to every scene.
Selection makes economics measurable
Raw generation cost is not enough. A cheap candidate rejected by the editor did not create a publishable second of film.
Track accepted still rate, accepted motion rate, alternate takes, correction time, settled cost, and first-cut acceptance. Over time, those measures reveal which briefs, routes, and review policies lead to complete films efficiently.
The AI film production workflow places selection between generation and final sound. The complete-film benchmark explains why accepted output should be the denominator for both quality and cost.
AI can generate a large volume of plausible material. Filmmaking is the discipline of deciding which material belongs, what it means, and how it changes the sequence.