Separate Spoken Sentences, Subtitle Cues, and Line Breaks
Handle these as three separate decisions:
- A spoken sentence determines the complete thought and breathing that the audience hears.
- A subtitle cue determines when a piece of text appears and disappears.
- A line break determines how one cue is divided across one or more lines.
One complete line of narration can appear on two subtitle lines without becoming two recordings. For example, the spoken line “After the tide went out, the boat repairer finally saw where the crack began” can appear as:
After the tide went out,
the boat repairer finally saw where the crack began.
“After the tide went out” and the rest still form one continuous thought. Breaking the narration into “The tide went out. Afterward. The boat repairer finally saw...” merely to shorten each line would fragment both the sound and the causal link. Line breaks serve reading; spoken phrasing serves the voice.
Lock Picture and Sound Before Detailed Layout
While cuts, narration placement, and speaking duration are still changing, subtitle in and out points will change with them. If you fine-tune every cue too early, moving one line of narration later can leave mistimed, obstructed, or outdated text behind.
Before detailed layout, confirm that:
- the shot order and main cuts are set;
- the narration text matches the actual recording in use;
- you know which original dialogue remains and which hard subtitles are already embedded;
- the positions of signs, documents, and interfaces the audience must read are marked;
- the delivery canvas and any overlapping titles or corner labels have an actual plan.
When the spoken content changes, update the narration and recording before changing the corresponding subtitle. Editing only the on-screen text does not change what the audience actually hears.
Preserve Complete Meaning When Breaking Lines
Find a meaning boundary first, then consider visual length.
| Prefer | Avoid |
|---|---|
| After a complete time or condition phrase | Leaving “not” on one line and the negated action on the next |
| Between complete phrases or clauses | Separating a name from its title, or a number from its unit |
| Before the person being addressed changes | Leaving a pronoun alone on one line |
| Where the causal point is already complete | Breaking abruptly before the key verb |
Each line should let the audience keep reading without stumbling. For a narrow frame and a genuinely long sentence, first try meaningful wrapping, cue division, placement, and readable type. If locked narration needs shortening, treat that as a copy revision and update audio and subtitles after the author accepts it. Do not keep shrinking the type, and do not turn one natural sentence into word-by-word flashes by adding more cues. How much fits on one line depends on the canvas, typeface, reading time, and occupied parts of the frame; there is no universal fixed character count.
Check the Layout with Five Types of Real Frames
Pull sample frames from this actual final cut. The worksheet below is a design plan; no images were generated, and it does not report a completed visual test.
| Frame | Record |
|---|---|
| Opening | Why inspect it: Often shares the frame with a title or hook; Main check: Whether subtitles compete with the title for space |
| Bright background | Why inspect it: Light areas can swallow light text; Main check: Whether the outline, shadow, or backing is sufficient |
| Dark background | Why inspect it: Dark areas can make a shadow ineffective; Main check: Letterform edges and color hierarchy |
| Character close-up | Why inspect it: Expression and lip movement carry information; Main check: Whether subtitles cover the face, gestures, or gaze |
| Longest subtitle | Why inspect it: Most likely to trigger tiny type or bad line breaks; Main check: Line count, margins, duration, and completeness of meaning |
A position that looks clean over an empty shot may cover important information once a character sits down or lifts a document. Sample frames help expose high-risk shots early. After the full export, play it normally as well, checking that moving characters, cuts, and subtitle entrances and exits do not create new obstructions.
Original Design Example: Three Subtitles During a Night Boat Repair
The following narration excerpt and layout are fictional examples created for this page. No real sample frames or finished video were generated. The shots show repairer Pei Zhao and an apprentice waiting for the tide, a lamp revealing the hull crack, and Pei drawing its endpoint with chalk before setting the chalk down. A repair folder is already beside the toolbox; next they retrieve the old order before comparing numbers. The bottom of the frame already contains hard subtitles from the source, while the toolbox number in the lower right is a story clue.
Narration one: “After the tide went out, Pei Zhao finally saw that the crack ran all the way under the old patch.”
Subtitle design:
After the tide went out,
Pei Zhao finally saw that the crack ran all the way under the old patch.
This keeps the time condition with the discovery and does not split “the crack ran all the way” in the middle. The character’s face is in the upper left and the crack is in the center, so the initial plan places this subtitle above the source’s hard-subtitle region. Its actual position must be checked against the real frame from that shot.
Source dialogue happens first: “Don’t patch it yet. Find the work order.” Pei says this to the apprentice beside him. The apprentice retrieves the old order from the existing folder by the toolbox and lays it on the work surface. Pei compares its seventeen with the toolbox number before this version’s second narration line.
Narration two: “The number seventeen on the toolbox matches the number on the old repair order.”
Subtitle design:
The number seventeen on the toolbox
matches the number on the old repair order.
The number in the lower right must remain readable, so the subtitle cannot cover the toolbox while this line appears. If the remaining space is insufficient, first move the subtitle to a checked region or use a shot that shows both the number and the character’s reaction. This version chooses disclosure after comparison, so subtitles follow that actual sound placement without revealing it earlier. Another work may deliberately lead with sound under its chosen information plan; subtitles follow that sound rather than redesigning the reveal.
Dialogue subtitles: Show the earlier “Don’t patch it yet. Find the work order” faithfully where it occurs rather than adding another decorative caption such as “He finds a problem in the repair records.” The type size, position, and duration of all three lines still need to be set against the actual recording and final picture.
In this example a corner of the old patch is already lifted, and light through that opening exposes the crack beneath it. This gives the first line a visible basis without treating ordinary surface lighting as a view through an opaque patch.
If captions appear twice, identify the layers first
| Source | Repair |
|---|---|
| Text already inside the picture, plus newly burned captions | Choose the retained presentation first. Hardcoded captions are pixels; deleting a new track does not erase them. |
| Two added subtitle or packaging layers repeat a line | Disable or remove the identified duplicate in the current project while retaining other lines and necessary time/place labels. |
| Burned text plus a player’s switchable subtitle track | Compare the same file with player captions off. Establish whether duplication is playback-only or inside the export; rerecording audio is unnecessary for this diagnosis. |
| Intentional source text and translation | Review the adopted bilingual layout. Similar-looking lines alone do not justify removing a language. |
Using Pei Zhao’s boat-repair example, design a paper candidate needing repair: suppose the source already displays “Don’t patch it yet. Find the record” clearly and faithfully, and the project adds that same line again. This version needs one presentation of the dialogue while retaining both narration captions. Check actual source footage before relying on that premise.
A complete revision request could read:
Keep the existing source caption “Don’t patch it yet. Find the record.” Delete only the added duplicate dialogue item. Retain the narration captions beginning “After the tide went out…” and “The seventeen on the toolbox…,” plus existing necessary labels. Keep the selected sound and cuts. In the new render, check that the dialogue has only its chosen presentation, remains readable and leaves the toolbox number, retrieval and comparison actions visible.
On the pinned explicit subtitle_track.json path, provide every generated cue that should remain, rather than a one-line deletion patch. In adopted-packet-copy mode the track completely replaces generated subtitles; it does not merge with legacy automatic captions. Empty cues remove all generated text without erasing letters already inside the picture. Check the current audio binding and complete retained-cue list before clearing narration to repair one dialogue line.
--no-burn-subtitles stops this run’s new burn-in; it cannot remove text baked into source footage. For replacement, prefer an available clean source or an explicitly reviewed masking design. The current mask draws a rectangular backing rather than reconstructing covered picture detail. A translucent backing leaves source letters visible; a full-bottom cover can hide the number or hand actions. Measure the real region and timing, then review it and neighboring action before adoption.
This revision request produced no real project or sample frames. After repairing duplication, continue the article’s line-break, type-size and playback review. Smaller lettering does not solve duplicated content.
Check Dynamic Problems After Export
A static sample frame confirms only one moment. During full playback, also check whether:
- subtitles appear before the voice begins or linger too long after it ends;
- an old subtitle remains after a cut into the next scene;
- a moving character, raised hand, or displayed object enters the subtitle region;
- the outline, shadow, or backing stays legible as the image shifts between bright and dark;
- a long line suddenly shrinks to fit, causing the type size to jump across the video;
- source hard subtitles, narration subtitles, and titles occupy the same region at once.
Fix the layer that contains the problem. If the text is too long, remove repeated information first. If the position obstructs the image, adjust the layout for that cue or group of cues. If the timing is wrong, check it against the actual sound. If hard subtitles are already embedded in the source pixels, follow the approved editing plan for the picture or choose another region. Do not use animation or borders to conceal incorrect timing.
Judge reading allowance, sound-leading cuts, and speech across shots against this project’s choices. Do not impose identical speech and cue boundaries on every work. A wording or recording change after layout requires updated text, audio, and timing evidence, followed by review of the new export.
Common Subtitle Layout Failures and Fixes
| Failure | Fix |
|---|---|
| Splitting narration into fragments just to make two subtitle lines | Keep one continuous recording and break lines only at reading boundaries |
| Choosing one position for the whole video after checking only an empty shot | Sample bright and dark scenes, a character close-up, and the longest subtitle, then play the full export |
| Always placing subtitles at the bottom and covering hard subtitles in the source | Identify which layer each text element belongs to before deciding how to avoid or handle it |
| Automatically shrinking a long line into tiny type | Remove repeated information or rearrange cues while preserving a readable hierarchy |
| Covering a story-critical number or document | Mark story text as picture information that must remain visible, then adjust the position or shot |
| Editing subtitle text while retaining the old voice-over | When the spoken content changes, update the recording and timing as well |
Original Paper Example: Two Packaging Orders for an Observatory Recap
The following is an original paper design. No video was generated, no subtitles were rendered, no audio was auditioned, and no full-playback check was completed. This fictional AI recap follows an investigator entering an abandoned observatory. Shot A shows a wall calendar, shot B shows an astronomical clock face, and shot C shows a duty log. The first narration draft says, “He finds the alarm code on the clock face.” The designer immediately creates a bouncing decorative caption for “alarm code” and adds a glowing border around the clock.
A later narrative check finds that the real causal link is the calendar date helping the investigator locate the corresponding duty log; the clock face is not the source of a code. The shot order changes to show the calendar before the log, and the narration is re-recorded as, “He uses the blackout date on the calendar to find the duty record from the same day.” The early decorative caption now repeats a deleted claim, its animation duration no longer matches the new TTS, and its glowing border is attached to a replaced shot. Packaging did not solve the narrative error; it only made the error more prominent.
| Packaging too early | Packaging after lock |
|---|---|
| Build word-by-word animation and a border while narration and shots can still change | First confirm a clean master containing the calendar, log, and new narration |
| Enter subtitle timing by hand from the old copy | Generate cues from the new TTS and timeline actually in use |
| Adjust the effect after viewing only the clock shot | Inspect the opening, calendar, log close-up, and frame containing the longest caption |
| Keep patching animation onto an old export | Generate the complete layer through the project's existing subtitle path and overlay it on the locked master |
| Preview only the animated fragment | Plan to verify the final audio and play the actual delivery file from beginning to end |
The right column is still only the correct planned order. It does not mean that this fictional short was rendered or validated. It shows that the rework comes from changing upstream inputs, not from having too few subtitle effects.
FAQ
How many characters should a video subtitle line contain?
There is no fixed number that works for every canvas. Break lines by complete meaning first, then check the actual type size, aspect ratio, duration, and background. If viewers cannot finish reading, remove repeated content or adjust the cue instead of relying only on smaller type.
Do subtitles always have to appear at the bottom?
Content determines the position; the bottom is only one option. Keep subtitles away from facial expressions, gestures, story-critical text, and hard subtitles in the source, and account for the delivery interface. Sample real frames before settling on a position, then play the full export afterward.
Do it with the skill
In a host with Video Recap Skills installed, call /video-recap with the locked picture, confirmed narration, and subtitle requirements, and first request a line-break and sample-frame review plan. If you only need wording and layout guidance, explicitly stop at the planning stage. After confirming the recording, timing, and style, proceed to subtitle composition and final export.
Read the method: Picture/sound lock, sample frames, and packaging order · Subtitle clock and track-validation limits · Continuous narration, subtitle reading, and full playback · Subtitle burn-in controls and assembly boundaries · Actual rectangular mask geometry and opacity