Chapter 50. Screens, Documents, Chats, and Graphic Packaging#

50.1 A graphic is a fact first#

Names, amounts, dates, results and identities must come from the project's fact ledger, not be typed in by a designer. Graphic assets reference fields, so one change propagates to the episode, the ads and every language version.

50.2 Information hierarchy#

The audience does not need to read a whole contract. Every graphic defines a first tier that must be read, a second tier of credible detail, and a third tier of decoration. On the authorization, the first tier is the fund name, the title and the signature; the other clauses supply realism without competing for the read.

50.3 Phone UI#

Build a fictional, consistent system: status bar, typeface, notifications, contacts, calls, chat and clock. Battery, time and unread counts continue across shots. Avoid imitating real application trademarks.

50.4 Documents and reports#

Design paper content as print layout, accounting for hand scale, folds, seals and signatures. Hero inserts use the high-resolution version; shots of a character holding it use a lower-detail version of the same layout.

Important text must never flow with generated video; use planar tracking or a separate static insert.

50.5 Chats and news#

Chat needs a sender, timestamps, read status and context — not a single plot line hanging on a screen. News headlines, outlets and dates fit the world's setting, with the source recorded as fictional.

50.6 Compositing realism#

Screen graphics match perspective, brightness, reflection, depth of field, refresh, moiré, motion blur and occlusion. A black screen should not be absolute black, and its glow should affect hands and faces.

Paper graphics match texture, shadow, curvature and finger occlusion.

50.7 Reading time#

Ask the audience to read only the minimum fields that advance the story. Test with a cold viewer at a single playback. If they must pause, add a close-up, repeat it in dialogue, or reduce the fields — rather than only raising the resolution.

50.8 Graphic style#

Unify typeface, numerals, dates, colors and emphasis across the series. Corporate systems, medical records and private chats can differ, and they belong to one world's product language.

50.9 Localization and editability#

Keep graphic text on layers separate from the base image, with layouts that allow text expansion. After translation, re-check wrapping, field widths and cultural formats. Never burn a single language permanently into an uneditable video.

50.10 Generate graphics from data bindings, not repeated typing#

Key graphics should render from structured fields. The authorization, bank statements, chat and news may appear across several shots, ads and language versions; typing each one by hand causes divergence at the first change. A template reads character ID, display name, amount, date, institution and status, and outputs a versioned graphic asset.

graphic_instance:
  id: GFX_AUTHORIZATION_E001_V04
  template: TPL_LEGAL_AUTHORIZATION_V03
  bindings:
    authorized_person: CH_LINYUN.display_name
    acquiring_entity: ORG_NORTHLIGHT_FUND.legal_name
    effective_date: TIMELINE_E001.acquisition_effective_date
  locale: zh-CN
  required_read_fields: [authorized_person, acquiring_entity]
  approved_by: story_editor

When a fact field changes, the system lists every referencing instance and re-renders candidates — without automatically overwriting an approved version already in a composite. New graphics still pass layout, reading and continuity checks.

50.11 A chat interface is a state timeline#

Chat is not a set of bubbles. It contains sender, device time, message order, read status, typing indicators, network state, deletions and recalls. After a cut, time and unread counts must advance; a message the character has not opened cannot show as read in the next shot.

When designing a chat scene, write the message ledger first and then decide which screen to show. You may compress unimportant history so the audience can read clearly, and you may not change the causal order. When a character screenshots, forwards or deletes, retain the original message ID so evidence can be traced later.

50.12 Proof graphics take three reading tests#

The first looks at the graphic file alone, checking hierarchy and facts. The second places it in the real shot with perspective, reflection, depth of field, occlusion and subtitles, checking actual visibility. The third plays once at normal speed on the target phone and asks someone who does not know the story to say what they saw.

If a cold viewer can only say it looked like a contract, you cannot claim the acquisition authority has been proven. Add an insert, reduce the fields, change the highlight, support it in dialogue, or hold longer. Raising the resolution rarely solves insufficient screen area and attention.

50.13 Continuity after graphics are composited#

An approved graphic does not mean every shot may scale it freely. Phone screen brightness, chat scroll position, document folds, seal orientation and finger occlusion are all shot state. Adjacent shots reference the same asset version while recording viewport position and visible region.

Screen light should change with content without exaggeration. The color temperature and brightness a white document interface casts on a face should continue into the reverse. In a blackout where the phone becomes the key light, graphics, character and environment update together.

50.14 Three versions of the authorization#

The same authorization needs three derived assets. The handheld wide version keeps only the true masses. The hero insert guarantees the title, the fund name, Lin Yun's name and the signature are clear. The advertising version re-lays the must-read fields for a shorter hold and the CTA safe area. All three reference the same fact fields and document version — they are not three independent designs.

The original approach used the high-resolution hero version in every shot, and small type flickered and moiréd during motion. After tiering, the wide is more stable, the insert is more legible, and the ad no longer has to be hard-cropped from the episode's picture.

50.15 SOP for graphics#

First, read facts from the evidence and timeline ledgers. Second, define the must-read hierarchy. Third, use the project's UI and document components. Fourth, produce editable layers for other languages. Fifth, track and composite reflections and occlusion. Sixth, run the single-pass phone reading test. Seventh, check every reference is consistent.

50.16 Fault tree#

Symptom: the same amount differs between shots. Graphics are not referencing the source of fact. Generate from centralized fields.

Symptom: the screen looks like a sticker. Perspective, reflection, interactive light and occlusion are missing. Redo the composite.

Symptom: the UI overpowers the characters. Hierarchy and color are too strong. Emphasize only the story fields.

Symptom: layouts break after translation. There are no expandable components. Re-lay the design rather than shrinking below legibility.

50.17 Checklist, exercises and deliverables#

Check that fields come from the fact ledger; that must-read information is limited; that phone state is continuous; that trademarks are safe; that compositing is convincing; that reading completes in one pass; that other languages are editable; and that every reference agrees.

Exercise one: build a three-tier information hierarchy for the authorization. Exercise two: design a continuous three-shot phone chat. Exercise three: produce a screen planar-tracking checklist.

Deliverables for this chapter: the graphic registry, the UI kit, document templates, fact bindings, tracked composites, the readability test, and the localization package.

50.18 The graphic renderer generates from the fact ledger, never from screenshots#

Every template in the graphic registry defines its fields, types, formats, required items and data sources. The authorization reads name, scope of authority, date and signature version. Phone chat reads sender, message ID, send time, read status and attachments. The renderer generates the image from the project's fact ledger so designers do not retype anything.

Field validation runs before rendering: amounts must be numeric with a currency, dates must fall inside story time, character names must come from the entity registry, and signatures may reference only the currently valid version. Missing fields raise an error and block; placeholders never reach the final picture.

Graphic output produces both a visual file and a data snapshot, so a reviewer can trace which fact version produced the amount in a given pixel. When a fact changes, the system lists every affected graphic and re-renders — nobody has to hunt for screenshots.

50.19 Screen compositing must obey the physical frame#

Screen content matches perspective, edge occlusion, brightness, reflection, refresh, depth of field and motion blur, and physical realism must not cost legibility. Produce a clean master first, then generate the composite for the camera position. Do not typeset directly inside a tilted, blurred frame.

Touches, scrolling and page changes become UI state events. The touch point, scroll distance and content position must agree; a page cannot change before the character taps. Split complex interaction into a hand shot, a screen insert and a reaction, rather than demanding perfect hands, lip sync and UI in one shot.

Finally, check whether screen black level matches the environment, whether text is reversed, whether reflections cover key fields, and whether thin lines survive compression. Verify at the actual target bitrate and phone size.

50.20 Chats, news and documents all carry time state#

A chat interface is not one static tall image. Messages appearing, delivered, read, recalled, edited, attachments opened and screenshots spreading are all events. Each shot reads the message list visible at that moment; future messages cannot appear early in the scroll history.

News pages likewise store publication time, update status, comment counts and screenshot versions; documents store draft, signed, tampered and public versions. When the story turns on one document having been altered, the versions must be visually distinguishable and the metadata must support the reasoning.

Before reusing a phone or document shot across episodes, check its state time. Repeating an old frame to save cost while bringing back a recalled message or the wrong battery and clock breaks the evidence chain outright. Continuity of graphic assets matters as much as continuity of characters.

A note on sources#

Text generated by image models is unreliable. Making graphics independent assets solves legibility while protecting story fact and enabling multilingual expansion.