The six stages between a script and an episode
A generated episode passes through six stages. Five of them are cheap and fast; one of them is where the money and the reshoots are. Knowing which is which decides where a production spends its attention. As of 2026-09-12.
| Stage | What goes in | What comes out | Where the time goes |
|---|---|---|---|
| 1. Parse and split | A script or a novel arc | Cast, scene and prop sheets; an episode outline | Minutes of machine time, then an hour of reading it |
| 2. Reference art | The sheets, plus a style decision | Character sheets, empty scene plates, prop shots | Seconds per image, and a day of rejecting the wrong ones |
| 3. Storyboard | The outline and the reference library | Segments, each holding several shots | Almost all of it human |
| 4. Segment video | One segment of storyboard plus references | A clip per segment | Minutes per clip, multiplied by every take |
| 5. Voice | The dialogue as written | A read per line, or speech made with the picture | Seconds per line |
| 6. Assembly and delivery | Every accepted clip and read | One episode file per placement | Minutes, unless a join has to be re-cut |
Inclusion rule. The stages a vertical episode passes through when it is generated rather than filmed. Stages some productions skip, such as a separate grade, are not listed. Order. In the order the stages happen.
1Stage four is where a production lives or dies
Stages one through three produce documents. Documents are cheap to fix and cheap to throw away, and a mistake in one of them is found by reading. Stage four turns documents into billed seconds, and a mistake there is found by watching, after the money is spent.
That asymmetry is the argument for spending longer than feels comfortable on stage three. An hour moved from the storyboard to the render queue is the worst trade in this workflow, and it is the trade most first productions make.
2Stage two is built once and drawn on forever
Reference art is the library every later stage reads from. A character sheet made carefully in week one is consulted by every segment of every episode; a sheet made carelessly is the reason a face drifts in episode nine.
The test for a finished reference library is not whether the images look good. It is whether someone who has not read the script could pick the right image for a given shot without asking a question.
3The stages are not equally automatable
Parsing, voice and assembly are the stages a tool can carry end to end, because each has a well-defined input and a well-defined output. Storyboarding is the stage that resists automation hardest, since it is where the decisions are.
Productions that describe their setup as automated usually mean stages one, five and six. That is a real saving and it is not the same as an automated series.
4What to measure after the first episode
Three numbers, all of them collected for free while working: generated seconds against delivered seconds, hours spent per stage, and how many shots failed for a reason the storyboard could have prevented.
The third of those is the useful one. It converts a vague sense that the process is wasteful into a specific list of storyboard fields that were left blank, which is something a second episode can fix.
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: Two passes, Four sheets.
5One pipeline's continuity
Serial work like “The six stages” depends on continuity; SceneMixer's guides describe how it keeps it:
SceneMixer turns a novel into a watchable short-drama series in one pipeline: upload the novel (Chinese or English) → AI extracts characters, scenes, and an episode-by-episode structure → generates consistent character and scene art → writes storyboard scripts → renders video segment by segment (16:9 by default, 9:16 in one setting) → composites full episodes.
SceneMixer, Turn a Novel into a Video Series with AI (read 2026-10-13)
A script longer than about 10,000 characters (counted as Chinese-equivalent characters) is read in two passes: first the cast, sets and props, then the episode list with a short outline per episode.
SceneMixer, Long Scripts Are Read in Two Passes: What “Preparing Episodes” Means and Why Some Episodes Open Later (read 2026-10-13)