blocks for a
larger paragraph gap. Let the display wrap other lines naturally.
Use standalone [cue: Section name] markers for operator navigation.
Keep only talent-facing directions in square brackets; they are visible,
excluded from speech matching, and never execute actions automatically.
Include a spoken-word count and duration estimate with the assumed WPM,
excluding bracketed directions. Keep the operator handoff separate from
the spoken script: pronunciation, suggested display/speech settings,
Launchpad cue plan, and editor/take-note fields where relevant.
If I request notes bundled with the script, also provide single-script JSON
with title, html and notes. Do not create a workspace backup or settings API.
Check the result against the guide's handoff checklist before delivering it.
My brief: [topic, audience, purpose, target duration, speaker, language,
must-say wording, factual sources, preferred emphasis and production context]
```
## Write for a person speaking
Establish the audience, desired outcome, target duration, speaker, language and must-say wording. Ask only for missing information that would materially change the script; otherwise state reasonable assumptions. Preserve approved facts and legal or brand wording. Mark uncertainties in the handoff rather than silently inventing answers.
Use short, speakable sentences and clear transitions. Keep each paragraph focused on one thought. Read it aloud: a sentence that looks elegant can still be difficult to say. Use deliberate line breaks at meaningful breath points, not fixed screen widths. Changing the configured canvas, font size or margins changes wrapping; connected screens scale the shared canvas to preserve the operator mirror.
Write names, abbreviations, dates, units and numbers as the talent should say them when that improves clarity. Keep unfamiliar-name pronunciation guidance in the production notes, or as a brief visible bracket direction when the talent needs it. Avoid several alternative phrasings in the spoken passage. For multiple speakers, use a consistent bracket label such as `[Alex]`; the label will remain visible but will not be matched as speech.
Estimate speaking time from **spoken words ÷ words per minute**. State the assumed pace and leave room for pauses and delivery. Cue Studio's main word count and remaining estimate include bracketed directions, so they may differ from your spoken-only estimate. Timed mode and the displayed estimates do not guarantee the actual take duration.
## Formatting that survives import and editing
Use simple paragraphs and inline marks. The reliable rich-editor subset is `p`, `br`, `strong`, `em`, `u`, `s`, `span`, `h1`, `h2`, `h3`, `blockquote`, `ul`, `ol`, `li` and `hr`. Put color, background color and relative size on **inline spans**, not on paragraphs. Preserve spaces between adjacent inline elements.
| Intent | Import-ready HTML | How to use it |
| --- | --- | --- |
| Emphasis | `the main point` | The primary stress in a sentence. |
| Softer contrast | `one small change` | Use sparingly; long italic passages are harder to scan. |
| Exact wording | `by Friday` | A short phrase the talent must land precisely. |
| Text color | `a fresh start` | Check contrast on the actual talent background and editor theme. |
| Highlight | `remember this` | Translucent color with inherited text works across more backgrounds. |
| Word size | `one thing` | Relative to normal script text; combine with bold when helpful. |
| Normal line break | `First thought.
Next thought.` | Same line spacing as wrapped lines; Enter in the editor. |
| Paragraph gap | `
First thought.
Next thought.
` | Larger gap; Shift+Enter in the editor. | | Navigable section | `[cue: Opening]
` | An unspoken, visible operator marker on its own line. | **Word sizes are 50%–300%, inclusive.** Use percentage values only, normally 100%, 115%, 125% or 140%. The app rounds valid values to whole percentages; 100% restores normal size. Pixel, point, em/rem and calculated sizes are not supported in script markup. Nested sizes are flattened rather than multiplied. A few enlarged words can help emphasis; many large words change wrapping and make the script harder to follow. Bold, italic, underline, foreground color, highlight and size can be combined on the same words. Do not style every word. Keep a small consistent legend, such as **bold = stress**, **amber highlight = must land**, **underline = exact phrase**. These are editorial conventions you define, not built-in commands. Do not rely on color alone to convey meaning. Paragraph `text-align` survives editing and exports, but **Talent view uses the global alignment setting**. Set left, center or right in **Settings & tools → Text & reading**. Typeface, base size, line spacing, margins and automatic/LTR/RTL direction are also display settings. Recommend them in the handoff rather than trying to enforce them with HTML layout styles. Ordinary text should inherit the configured prompt color; explicit word colors remain explicit when that default changes. ### Ten customizable highlight presets These are the factory colors, each applied at 32% opacity. Users can change all ten swatches in the highlight menu. A script stores literal colors, not references to swatch numbers: changing a preset does not recolor existing words. Ask for the user's palette if matching an existing project matters; an AI cannot infer a browser's saved custom colors from this public guide. | Preset | Base color | Highlight style value | | --- | --- | --- | | 1 · Amber | `#e9b949` | `rgba(233,185,73,0.32)` | | 2 · Orange | `#ee8e55` | `rgba(238,142,85,0.32)` | | 3 · Coral | `#ed6f79` | `rgba(237,111,121,0.32)` | | 4 · Rose | `#d97baa` | `rgba(217,123,170,0.32)` | | 5 · Violet | `#aa86df` | `rgba(170,134,223,0.32)` | | 6 · Blue | `#7c96e5` | `rgba(124,150,229,0.32)` | | 7 · Sky | `#58b6db` | `rgba(88,182,219,0.32)` | | 8 · Teal | `#50b9af` | `rgba(80,185,175,0.32)` | | 9 · Green | `#78b86b` | `rgba(120,184,107,0.32)` | | 10 · Olive | `#b8ba62` | `rgba(184,186,98,0.32)` | The names above describe the colors for this guide; they are not script keywords. Use only the colors that serve this particular script. Avoid white text on opaque bright yellow, and do not assume a chosen foreground color will be readable in both editor themes. Preview explicit colors on the intended talent background. ### Leave these out of the script file No JavaScript, images, SVG, links, tables, forms, embeds, external stylesheets, classes, arbitrary data attributes or layout CSS. These are unsupported or removed. Use `background-color`, not the `background` shorthand. **`` and Markdown `==highlight==` are not supported highlight syntax.** SSML, animation and custom automation commands are not script features. Subscript/superscript and per-element text direction are not reliable through the current rich-editor round trip; use plain wording and the global direction setting. ## Cues, directions and what the talent sees `[cue: Opening]`, `[marker: Closing]` and `[[New section]]` create jump markers. Headings H1–H3 also create markers, but ordinary heading text is spoken text. Prefer a separate `[cue: Opening]
` when the label should not be read aloud. Putting a bracket marker inside a heading produces duplicate markers. **All single-line square-bracket directions remain visible** and are skipped by voice matching and rehearsal speech. For example, `[Take a breath]` or `[Look at camera two]` is a visible direction, not a sentence to read. Do not put intended spoken words in square brackets. Directions are already smaller and dimmer on the talent screen, so avoid styling them even smaller or darker. **A written cue never executes an action.** `[Pause]` does not pause scrolling; `[Hold 3 seconds]` does not create a timer; `[Blackout]` does not black out the screen. Pictures, pause actions, countdowns and Launchpad keys need operator configuration and activation. Put backstage material in Production notes or Editor notes when it should not appear on the prompt. Production notes are not visually shown to talent, but they travel with the shared script state; do not treat them as a separate confidential channel. ## An import-ready example This short sample demonstrates the markup, not a factual client script. Ordinary words inherit the display's text color. The green phrase is an explicit-color example; check its contrast against the chosen background. ```html[cue: Opening]
A better first take starts with one clear thought.
Give it space. Let it land.
You do not need to rush every sentence.
You need to make the important words easy to find.
[Take a breath]
Keep the promise simple: say what matters.
Then give the next idea a fresh start.
[cue: Closing]
One thought. One breath. One good take.
``` Download the [HTML example](/examples/ai-script.html) or the [JSON example with production notes](/examples/ai-script.json). The JSON file uses this structure: ```json { "title": "Your script title", "html": "[cue: Opening]
Your spoken script.
", "notes": "Pronunciation, styling legend and operator handoff; not shown on the prompt." } ``` `title`, `html` and `notes` are sufficient. Escape quotes and line breaks correctly in JSON. Do not include Markdown code fences in the actual file. Import creates a new script ID and timestamp and sets its tags to `Imported`; imported tags do not override that. A script JSON file is different from a workspace backup. Use **Import scripts** for it, not Restore backup. ## Choose settings for this delivery The app has **five scrolling modes**, **four speech input choices**, **ten editable highlight presets** and a **6–32-key Launchpad** with shared opacity and front/behind-text controls. Settings & tools has nine sections covering reading, playback, talent display, speech following, workspace, speech setup, rehearsal, recording, and recovery. The following are useful capabilities to account for; they are not a promise that every script needs every option. | Capability | What an AI should recommend or prepare | | --- | --- | | Manual scrolling | Operator advances or seeks; useful for unpredictable delivery. | | Constant scrolling | Choose a WPM pace and rehearse it. | | Timed scrolling | Choose a target duration; this does not guarantee spoken timing. | | Voice tracking | Recognized words advance the reading position while prompting is playing. Keep the script close to what will actually be said. | | Sound gated scrolling | Microphone volume starts/stops movement at the chosen pace. It does not recognize words or find the speaker's place. | | Operator mirror | Live Talent view and a separate display window share the script and reading position. Clicking a word in the preview reveals it in the editor. | | Display controls | Base size 24–180 px; line spacing 1–2.5; margins 0–35%; left/center/right alignment; automatic/LTR/RTL direction; reading line 15–65% down the canvas. | | Timing and keys | Pace 10–600 WPM; timed duration 10–7,200 seconds; countdown 0–15 seconds; arrow-key speed step 1–100 WPM, default 10. Up/down change pace; left/right move five words. | | Display presets | Elgato Original 1024×600; Elgato XL 1920×1080; optical glass 1280×720; custom canvas 320–7,680 wide by 240–4,320 high. Confirm the physical display and mirroring. Separate talent tabs have Mirror (horizontal) and Flip (vertical) beside Fullscreen. The whole talent window follows the orientation, including cues, countdowns, messages and display controls. Each axis overrides only that tab until reload; untouched axes follow the operator settings. | | Launchpad appearance | The Launchpad header sets picture/message opacity from 0–100% and places cues in front of or behind the script. Changes apply live across updated displays and persist with workspace settings. Script opacity is unchanged; Blackout still fully covers the screen. | | Rehearsal | Play / Pause / Stop beside Talent view read from the current prompter position, skipping bracket cues. They share voice, rate and pitch with Studio controls → Rehearsal. Google UK English Female (en-GB) is the default when available, otherwise the system voice is used. You can choose another voice for the current session. Available Google voices appear first; voices are not bundled or guaranteed. | | Recording | Optional microphone recording, up to 30 minutes per take. Download it before closing the tab or starting a replacement recording. | | Library and rundown | Group scripts into project folders, order them in a rundown, and use named cue points for pickups. Search titles, tags and script text within the selected project or across All scripts. | | Recovery and export | Script versions, deleted-script recovery, workspace backups and notes exports. A backup is broader than a script export. | Offer starting settings as suggestions, never as automatically applied values. Keep paragraphs uncluttered enough that live font-size and line-spacing changes remain readable. Software mirroring and the physical glass/driver setup need an actual screen check; Elgato presets start with software flipping off. ### Speech-aware handoff For English, the on-device **Moonshine streaming** engine offers incremental recognition; Small is the recommended model, with Tiny for slower devices. **Offline Whisper** supports more languages but processes audio in 3–8-second chunks. **Browser recognition** depends on browser/provider support and may send audio to that provider. **External transcript** accepts supplied text for matching/testing; it is not a bundled connection to any external recognizer. Choosing Voice or Sound gated through the operator control requests listening; microphone permission and any required model preparation still apply. Restored settings do not silently reopen the microphone. Check the indicator beside Talent view. Pausing the take keeps listening active, while choosing Manual, Constant or Timed stops it. Voice following advances only while playing, outside countdown and temporary manual takeover. Changing engine, language or microphone may require restarting through the indicator. Write words the speaker will actually say. Use unique section labels and operator seeking for large jumps; repeated passages can be ambiguous. Recognition tolerates some punctuation, number variants, fillers and skipped words, but it is not guaranteed to infer an improvised rewrite. Never promise zero lag or perfect tracking. Prepared local engines can work offline; prepare the model and test the real microphone and room before the shoot. Rehearsal stops live recognition to avoid feeding synthesized speech back into tracking. ## Plan Launchpad signals separately The Launchpad offers layouts with **6, 9, 12, 15, 16, 24 or 32 keys**, **59 button icons**, **36 visual talent signs** and a Text only option. Six actions are available: show a talent cue, blackout/restore, clear cue, play/pause, jump back to section start, and restart script. Messages can appear as full-screen cards or over-script banners. Supply a small configuration plan if the shoot needs signals; it is not executable script markup. | Label | Action | Talent message or sign | Pause | Duration | Shortcut | | --- | --- | --- | --- | --- | --- | | Please hold | Show talent cue | Please hold; choose the matching hold sign | Yes | 0: stays until cleared | Operator chooses | | Mark edit | Show talent cue | Raise your hand to mark the edit; hand sign | Yes | 0 | Operator chooses | | Pick up section | Jump back to section start | Confirm the intended section with the operator | Yes | Not applicable | Operator chooses | Launchpad visual cues keep scrolling by default. Enable Pause scrolling for this cue in Edit for any key that should pause. Jump back to section start always pauses; Blackout does not pause scrolling. Cue duration is 0–120 seconds; zero stays until cleared. Clearing or expiring a cue does not automatically resume scrolling. Button labels are limited to 36 characters and messages to 500. The default key named **Restart section** displays a direction; it only performs a real jump if configured with the jump action. Numpad and Stream Deck use configured keyboard shortcuts; the operator tab must have focus. Collapsing the Launchpad hides its keys but keeps configured shortcuts active; an active cue can still be cleared and a blacked-out display restored from the collapsed header. Icons can be exported for Stream Deck. There is no implied native background/global control. Do not include images in the script HTML to emulate Launchpad signs. ## Organize related scripts Use **Projects → New project** in the library, then move existing scripts with each script's folder button. New and imported scripts go into the selected project. Removing a project keeps its scripts in Unfiled. Workspace backups include folder names and membership; restoring a backup creates separate folders and remaps its script IDs rather than overwriting existing work. Folders organize the local script library. They do not create separate relay rooms or automatically split the Editor notes project. Single-script HTML/JSON exports do not include a folder assignment. The MCP catalog can work with scripts, but does not currently create, rename or assign these folders; include folder suggestions in the operator handoff. ## Prepare the editor's handoff Keep script-level **Production notes** for pronunciation, emphasis legend, timing assumptions and operator instructions that should not be rendered on the prompt. The separate **Editor notes** workspace can appear beneath the desk or in its own tab, with a QR code or link for another device. Editor notes organize takes by script. Available fields include talent, Good / OK / Bad / Unrated rating, scene, camera, clip/file, timecode, notes and up to 24 custom fields. Project details include name, client, shoot date, location, default talent and shared notes. Suggest relevant fields, such as preferred pickup or pronunciation correction, without inventing camera filenames, timecode or take ratings before the shoot. The app's play/pause journal is not proof of a camera take, synchronized timecode or recorded audio. A notes-sharing link grants editing access to the notes. Open it online once on each device before offline use; locally saved changes synchronize when a connection returns. Different devices cannot synchronize with no connection. Notes export to CSV or project JSON and are included in workspace backups. ## File formats and practical limits | Format | Use and limits | | --- | --- | | HTML / HTM | Preferred for precise rich formatting; simple fragment, UTF-8, supported tags/styles only. | | Single-script JSON | Carries title, HTML and production notes. Title max 200 characters; notes max 20,000. | | TXT | No rich styling. Single newline imports as `