---
name: layered-voice
description: Place an editable expanding headline behind a moving presenter, using synchronized footage and an alpha cutout.
version: 1.0.1
---

# Layered Voice

Use this parameterized composition generator for a short presenter-led headline. A title grows across the frame, eases into a steady reading interval and clears while the original shot continues. This tool does not generate footage or narration.

## Inputs

Prepare a JSON brief with `headline`, `footage`, `cutout`, and optionally `seconds` (5–20, default 7.2), `palette` (`cream`, `glacier`, `rose`), `cropScale` (1–1.6), `subjectX`, `subjectY` (0–100), and `baseline` (330–700). File paths resolve beside the brief. Use short phrases, ideally two or three words, with enough visible lettering on both sides of the person.

Use licensed continuous footage and a transparent video derived from that exact take. Match frame rate, dimensions, start time, frame count and crop. Never substitute a static portrait or unrelated cutout. Avoid implied personal endorsements in stock-footage examples. Retain media permissions with the editable project, outside this generic package.

## Build

Run `npm ci --ignore-scripts`, then `node create.mjs /absolute/brief.json /absolute/project`. The generator copies media, local fonts and GSAP into an editable HTML/SVG project. It rejects short media, mismatched rates or dimensions, missing alpha and invalid configuration. Metadata checks cannot prove frame identity: inspect facial and hand edges yourself.

Render with `npx hyperframes render /absolute/project --fps 30 --quality high --strict --workers 1 --output /absolute/movie.mp4`. HyperFrames owns media timing. Do not add imperative video playback controls to the composition.

If a cutout is missing, an optional local operation is `npx hyperframes remove-background /absolute/take.mp4 -o /absolute/cutout.webm --quality best`. It requires optional native inference dependencies and may download model weights. Review the actual alpha result before use; automatic segmentation is not guaranteed to preserve hair or hands. Respect local installation and resource permissions.

## Review

Check the initial frame before playback, title entrance, settle, reading interval, exit and final frame. Compare base and cutout during speech and hand movements. Seek backward and forward and inspect encoded frames at full resolution. Confirm loaded fonts, painted text bounds and meaningful visible letters despite intentional occlusion. Inspect horizontal and vertical crops separately if adapting the aspect ratio.

The default output is silent. Add separately licensed audio through an editable media timeline when a brief requires it; a cue list is not a mixed soundtrack. Probe the final file and keep playback, listening and user acceptance distinct.

## Dependencies

Fixed versions live in package-lock.json. Node.js, FFmpeg, a supported browser and HyperFrames are required. GSAP and font notices must accompany redistributed runtime files. Run dependency review for your deployment context; installation does not establish a security approval.