Chapter 14. Narration, Interior Voice, and the Narrated Comic#

14.1 Narration is not a caption for the picture#

A line telling us that Lin Yun walks into the boardroom and produces a document is reading out what the audience can already see. Narration should do the jobs the picture cannot do efficiently: compress time, supply a position, create information asymmetry, connect discontinuous events, reveal knowledge that cannot be spoken aloud, or change the meaning of an image through irony.

The test is to mute the narration. If the information is still complete and the meaning unchanged, the narration may be redundant. Only if muting it removes necessary context or a distinctive attitude is it doing work.

14.2 Distinguish four voices#

Objective narration describes the rules of the world and jumps in time. Interior voice supplies only what that character can know at that moment. Retrospective first person speaks from a later point and may evaluate the past in advance. The commentator has a pronounced attitude and may mock, take sides, or deliberately mislead.

Their knowledge permissions differ. Lin Yun's live interior voice cannot know that Zhou Lan has secretly arranged her mother's situation. A retrospective voice can say she did not know it then. An omniscient narrator can foreshadow, but doing so changes the structure of the mystery.

The project must establish a narration POV rule, forbidding convenient switching between omniscience and limited knowledge.

narration_rules:
  mode: first_person_retrospective
  narrator: CH_LINYUN_future
  knows: the facts as of the end of the season
  reveals: only the evaluation this episode needs; never the core answer early
  tone: restrained, precise, occasionally dry
  forbidden:
    - describing another character's undisclosed interiority
    - repeating an action the picture already shows
    - explaining what evidence means before the reversal

14.3 The six legitimate jobs of narration#

One, compress undramatic time — three years spent doing only one thing. Two, establish a subjective frame: Zhou Lan never lies, she simply decides for other people what counts as true. Three, bridge location or time. Four, supply rules the picture cannot present alone. Five, create irony by setting language against image. Six, in a narrated comic, control pace and attitude.

If the narration is completing the conflict for a character, announcing an emotion, or explaining a reversal that just happened, first try converting it into action, dialogue and reaction.

14.4 Four relationships between narration and picture#

Duplication: the voice says she closes the door while the picture shows her closing it — the least valuable. Complement: the picture shows the old pen while the voice explains it is the only thing her father left. Contrast: the narration says she always protects her family while Zhou Lan has the badge cut. Delayed explanation: an abnormal action appears first and the narration changes its meaning later.

Do not run one relationship throughout. If every sentence in a narrated comic has a literal illustration, it acquires the feel of automatic image matching. Alternate action, evidence, reaction, environment and contrast.

14.5 Organize long narration by thought, not by sentence#

The smallest writing unit in long narration is the thought segment: it proposes an idea, develops or reverses it, and lands on the next visual question. Twenty to forty-five seconds usually forms one production segment, though the real length follows meaning and emotion.

A worked example. Segment A explains how Zhou Lan controls the company's narrative. It establishes that she rarely orders anyone to lie outright. It develops by showing she explains everyone's motives for them first. It reverses on the observation that by the time facts arrive, people have already decided who is not credible. It lands in the boardroom, where Lin Yun's resignation statement has already been printed.

Generating sentence by sentence resets pitch, pace and emotion repeatedly. Preserve paragraph breath while writing, generate by thought segment, and cut at natural pauses.

14.6 Conversational is not the same as adding filler words#

A written sentence about failing to recognize someone's true intent given a complex set of circumstances is not speech. The spoken version says that at the time she only saw the false accounts and did not see why Zhou Lan needed her to see them. It has clear stress, a contrast, and breath.

Read-aloud testing checks whether it can be said in one breath; whether the stress falls naturally; whether three consecutive sentences are the same length; whether the pronouns are clear; and whether numbers, names and technical terms are awkward in the mouth. Delete written connectives — in fact, evidently, it is worth noting — and connect through the relationship between thoughts and through pauses.

14.7 Narration density and picture refresh#

Audio information density and visual change cannot both stay high for long. Use clear, stable images under complex narration; reduce explanation during fast action and let the voice supply rhythm or a single judgment.

Build a visual beat map per thought segment rather than one image per sentence:

narration_segment: NAR_E001_01
text: Zhou Lan never lies. She simply decides, in advance and for everyone, whose word is not worth trusting.
duration_estimate: 7.8s
visual_beats:
  - 0.0-2.0s: an old photograph of Zhou Lan straightening Lin Yun's collar
  - 2.0-4.8s: the date on the resignation statement, earlier than the meeting
  - 4.8-7.8s: directors avoiding Lin Yun's eyes
relation: contrast_then_evidence

The picture is not a word-for-word translation. It gives an abstract judgment its evidence.

14.8 Interior voice must change a choice#

Interior voice is often used to explain — she was afraid but could not back down. A better use is showing the character deciding between two strategies, or letting the audience know what she is concealing.

An interior line noting that Lin Wei is waiting for her to mention her father, and that the moment she does the meeting stops being about the accounts, explains the mechanism of a choice. Lin Yun then refuses the provocation and asks for the ledger instead. The line is not an emotional label.

Too much interior voice weakens the performance. Use it once at a key choice and let eyes, hands and pauses work everywhere else.

14.9 The dual-track structure of a narrated comic#

A narrated comic usually runs a narration track and a character dialogue track together. Narration skips low-value process, supplies attitude and links images. Dialogue keeps only relational collision, reversal and the memorable moments.

dual_track:
  narration: On her first day back, the badge went first.
  visual: badge fragments landing on the table
  dialogue:
    CH_LINWEI: Former employees have no right to attend.
  narration: The one thing Lin Wei miscalculated was whose agreement was already on that table.
  visual: the acquisition agreement and the authorization
  dialogue:
    CH_LINYUN: Then the acquirer's representative will attend.

Do not let narration steal the reversal from the dialogue. Narration should carry the audience to the edge of the conflict and let the character land the blow.

14.10 Commentary needs a target and a position#

Remarking that a woman is truly wicked, or that he never saw it coming, is noise. Effective commentary exposes the logic of a behavior, a social rule, or a character's self-deception — and holds a consistent position.

The commentator may take the protagonist's side, but must not turn every opponent into an idiot. The sharper the commentary, the more the picture has to supply evidence, or it reads as the author forcing an opinion on the audience.

Establish the commentator's vocabulary boundaries, kind of humor, and what is never mocked. That prevents personality drift across episodes and prevents a sudden joke over a serious injury.

14.11 Splitting long narration for production#

The writing draft, the generation draft and the edit draft are three versions. The writing draft keeps paragraphs. The generation draft adds pronunciation, pauses, stress and emotion. The edit draft references final audio timecode.

voice_segment:
  id: VO_NAR_E001_01
  text: Zhou Lan never lies. She simply decides, in advance and for everyone, whose word is not worth trusting.
  pronunciation: {}
  pace: controlled_medium
  emphasis: [never lies, whose, not worth trusting]
  pause_ms: [320, 180]
  emotion_curve: neutral_to_cold
  overlap_text_next: whose word is not worth trusting

Adjacent long segments can keep one overlapping sentence, generated separately so you can choose the more natural join. Derive final timing by forced alignment from the audio; never build final subtitles and shot timings from estimates.

14.12 Pronunciation dictionary and number rules#

Personal names, company names, abbreviations, amounts, dates and technical terms all enter the project pronunciation dictionary. Specify how a figure is spoken, how a name with an ambiguous reading is pronounced, and whether an abbreviation is spelled out or said as a word.

The pronunciation dictionary is an audio asset and carries a version. When a character's name changes, the system should find every narration, dialogue and subtitle reference.

14.13 Multiple languages and localization#

Do not lock long narration to a Chinese duration and then require every language to fit inside it. Information density differs by language. An international version maintains semantic segments and visual elasticity: which shots can extend, which pauses can compress, which cultural information needs rewriting.

Localization is not word-for-word translation. Forms of address, amounts, corporate structures and the phrasing of payoff moments have to land immediately for the target audience while the story facts stay identical.

14.14 Four playbacks for narration review#

The first reads the text only, checking viewpoint and knowledge permissions. The second listens to scratch audio with eyes closed, checking rhythm, voice and comprehension. The third watches with picture, checking duplication and whether the voice is stealing scenes. The fourth watches muted, checking whether the key actions are still comprehensible.

If turning the narration off makes the whole show incomprehensible, you may not have a narrated comic — you may have an audio story whose visual storytelling is unfinished.

14.15 SOP for writing narration#

First, fix the narrator's identity, temporal position and knowledge permissions. Second, keep only the legitimate jobs. Third, organize long text by thought segment. Fourth, define each segment's relationship to the picture. Fifth, leave the key reversals to character action or dialogue. Sixth, read aloud and mark stress and breath. Seventh, build the visual beat map. Eighth, generate audio by thought segment with overlaps retained. Ninth, force-align to final audio. Tenth, complete the four review playbacks.

14.16 Fault tree#

Symptom: it plays like a narrated slide deck. Narration maps sentence by sentence to static illustration, and the picture has no action, evidence or relational change. Switch to thought segments and vary the picture relationship.

Symptom: the voice speeds up and slows down across sentences. Generation was per sentence, or parameters drifted. Generate by thought segment and lock the voice bible and pronunciation dictionary.

Symptom: narration steals every reversal. It explains the result in advance. Let it establish peripheral information only and let action or dialogue land the blow.

Symptom: lots of interior voice and the character still feels empty. The interior voice announces emotion instead of explaining choice. Reduce the quantity and connect it to strategy and cost.

Symptom: subtitle timing drifts further and further. Timing came from text estimates rather than final audio. Re-align and make audio the master clock.

Symptom: the commentary is off-putting. The narrator's position and boundaries are unstable, or it is judging on the audience's behalf with no evidence. Rebuild the narrator voice rules.

14.17 Checklist, exercises and deliverables#

Check what the narrator knows; whether narration does work the picture cannot do efficiently; whether it is organized by thought rather than sentence; whether the picture relationships vary; whether reversals still land through characters; whether pronunciation, stress and pauses are recorded; whether final timecode comes from audio; and whether key action holds when muted.

Exercise one: delete every sentence in a narration passage that duplicates the picture. Exercise two: write the same scene three ways — objective, interior, and commentary. Exercise three: build a thirty-second thought segment with a visual beat map. Exercise four: build a pronunciation dictionary for ten names, numbers and technical terms.

Deliverables for this chapter: the narration POV rule, narrator voice rules, the thought-segment script, the dual-track script, the visual beat map, voice segments, the pronunciation dictionary, and the force-aligned draft.

14.18 The narrator's permission ledger#

The narrator ledger records, at each point in time, what the narrator can know, evaluate, withhold, and may not enter. A first-person live viewpoint cannot describe the antagonist's actions alone in a room. A retrospective viewpoint knows the outcome but must decide why it is still withholding a secret at this moment. Every change of permission has to be explained by the narrative frame.

Run narration lines through the permission check before rhetorical rewriting. A brilliantly written line that exceeds its permissions destroys both suspense and character credibility. When a narrated comic switches viewpoint, mark it with clear audio and visual signals — do not let omniscience appear occasionally just to patch a hole.

14.19 Auditing redundancy between picture and voice#

Segment by segment, tag the facts, attitude and time compression supplied by audio, and the evidence, emotion and irony supplied by picture. Where the two tracks fully duplicate, delete one side or make it complementary. Where both carry different critical facts, check whether the audience is overloaded.

Good relationships include: narration proposes a judgment while the picture supplies a counterexample; narration compresses three years while the picture holds on one decisive object; narration stays restrained while music and performance leak the emotion. Do not pair every noun with a matching picture and produce a slideshow of search results.

14.20 Building narration versions that can be replaced#

Split long narration by thought segment rather than arbitrary word count, each with an entry state, a core judgment and an exit state. When rewriting one segment, keep the overlapping sentences on either side and the target duration, which shrinks the rework radius across voice, picture and music.

Maintain full, short and locale-ready structures at the script stage rather than machine-deleting sentences later. The short version drops secondary proof while keeping the logical bridge; the multilingual version may reorder syntax and edit handles. All versions reference one shared source of fact.

A note on sources#

Experience producing narrated comics shows that narration can compress story substantially and reduce the burden of performance shots — while also making homogeneity and a slideshow feel easy to fall into. This chapter treats narration as an independent narrative track and an audio asset; Part VIII continues with voice generation, mixing and the relationship to music.