What to decide before episode one
Shape, cast, locations, language and clip ceiling all become expensive to change once episodes exist. Deciding them deliberately at the start costs an afternoon; discovering them in episode four costs re-generating everything before it. As of 2026-09-12.
| Decision | Cost of deciding late |
|---|---|
| Aspect ratio | Everything generated so far is reframed or reshot |
| Cast reference images | Episodes before the decision do not match |
| Clip ceiling | The shot list has joins nobody planned |
| Episode length | The format drifts visibly across the run |
| Language | Dubbing or regeneration, both after the fact |
Inclusion rule. Decisions that bind the whole season. Choices revisable per episode are not listed. Order. Alphabetical by decision.
1Aspect ratio, first and irreversibly
Vertical or widescreen is not a setting you revisit. Generated footage composed for one shape cannot be cropped into the other without moving faces out of frame, because nothing was framed with headroom for a crop that was not planned.
Decide it against the distribution path, write it down, and do not let an individual shot drift because it looked better wide.
2The cast, as images rather than descriptions
Every recurring character needs reference images before the first episode: front, three-quarter, full body, and the wardrobe they are in. Text descriptions are a weak lever against a picture, and a model given only words will invent a different person each time.
This is the single highest-value hour in pre-production for a generated series, and it is routinely skipped because episode one looks fine without it.
3Locations, the same way
Recurring rooms and streets need establishing images for the same reason faces do. A room re-described rather than reused comes back with different furniture and a different window, and across an episode that reads as the story having no place.
Decide time of day per location too. The same room at a different hour is a different picture, and a model has no reason to guess the hour you meant.
4Language, before any dialogue is written
If the series will ship in more than one market, that decision belongs at the start rather than after a cut exists. Dubbing a finished episode and generating it in the target language are different pipelines with different vendors and different schedules.
It also affects casting: a character needs a voice per language, and matching voices across languages is harder than matching within one.
5The clip ceiling shapes the shot list
Knowing the longest single generation before writing the shot list lets the shot list respect it. Shots written without that knowledge produce joins nobody planned, and joins are where continuity breaks and regeneration costs appear.
This is a cheap constraint to design around and an expensive one to discover.
6Episode length and how strictly it binds
Short-drama episodes run tight, and a format that drifts from ninety seconds to three minutes across a season reads as loss of control. Fix the target early and treat overruns as a writing problem rather than a delivery one.
It also fixes the budget arithmetic, since everything scales off delivered duration.
7Who owns the record
One person should own the continuity log, the reference images and the delivery record. Distributed ownership produces several partial files and no complete one, which feels like diligence and fails at exactly the moment somebody needs an answer.
Name them before episode one rather than after the first problem.
8What can safely be decided later
Grade, music, titles, and most of the sound design. Those attach to finished picture and can be revised without regenerating anything, which makes them the right place to leave creative room.
Knowing which decisions are cheap to defer is as useful as knowing which are not, and it stops a team trying to lock everything at once.
Craft notes, not a register entry. No line here is attributed to a tool, and none should be read as a claim about one. The sourced material is on the register. Related: Regenerating safely, Keeping a face.
9Continuity in practice
Serial work like “Planning a season” depends on continuity; SceneMixer's guides describe how it keeps it:
Every character gets an AI-generated full-body reference sheet at analysis time. Face and body crops from that sheet are injected into every video segment that features the character, so the same face, hair, and signature look persist across all episodes.
SceneMixer, Turn a Novel into a Video Series with AI — “How does SceneMixer keep characters consistent across episodes?” (read 2026-10-13)
SceneMixer generates each character once as a full-body reference sheet — 9:16, plain background, directional key light, identity features written before any technical instruction — and then attaches crops of that sheet to every shot the character appears in.
SceneMixer, AI Character Sheet Generator (read 2026-10-13)