# Add polished overlays to talking-head videos

Talking-head videos can need clear visual structure without replacing the original footage. This skill creates timed graphic overlays from a local video and its transcript.

## Install

```bash
npx skillstore add heygen-com/talking-head-recut
```

## Metadata

- Status: approved
- Slug: heygen-com-talking-head-recut
- Skillstore revision: r2
- Version status: missing
- Tree hash: 46bcf1e7cc3c391840f98c6211cdb6e34cc0054e46cc223353c4d58f02e48db4
- Author: heygen-com
- GitHub username: heygen-com
- License: MIT
- Repository: https://github.com/heygen-com/hyperframes/tree/main/skills/talking-head-recut/
- Ref: 77d0e66fbadeb1f5d1e6a5aada7861240b8ed2d7
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: scripts, external\_commands, network, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/heygen-com-talking-head-recut
- Manifest: https://skillstore.pages.dev/api/skills/heygen-com-talking-head-recut/manifest

## Capabilities

- Extracts local video metadata and audio with ffprobe and ffmpeg.
- Creates a local Whisper transcript with word-level timestamps.
- Plans timed overlay cards from a corrected transcript.
- Builds titles, lower-thirds, quotes, callouts, side panels, and picture-in-picture layouts.
- Supports 16:9, 9:16, and 4:5 output canvases.
- Renders an assembled HTML composition to an MP4 with HyperFrames.

## Use Cases

- Package a podcast clip: Add quotes, speaker labels, and data callouts to a podcast excerpt.
- Improve an interview video: Add lower-thirds and timed context cards to an existing interview.
- Prepare social video variants: Adapt overlay layouts for portrait, landscape, or 4:5 delivery.

## Prompt Templates

### Start with one video

```
Add simple title and lower-third overlays to my local interview video. Use the recommended aspect ratio and a restrained style.
```

### Create a podcast package

```
Package this podcast clip with transcript-synced quote cards, speaker labels, and two data callouts. Keep the video visible beneath the graphics.
```

### Produce a portrait cut

```
Create a 9:16 version of this talking-head video. Use picture-in-picture only when it improves readability, and use six timed graphic cards.
```

### Direct the visual system

```
Build a 16:9 talking-head recut with an editorial theme, side panels for metrics, short kinetic title moments, and card timings based on meaningful transcript transitions.
```

## Limitations

- Requires a local input video, ffmpeg, ffprobe, and HyperFrames rendering dependencies.
- Overlay quality depends on transcript accuracy and the supplied footage.
- It does not create new footage or replace the underlying video content.
- It does not provide plain subtitle-only workflows.

## Best Practices

- Review and correct names, numbers, and technical terms in the transcript before planning overlays.
- Keep cards short and place them only at meaningful moments in the speech.
- Preview the rendered video and verify that overlays do not hide important visual content.

## Anti Patterns

- Do not use this skill for subtitle-only requests.
- Do not add a graphic card for every sentence in the transcript.
- Do not use unverified transcript text for names, prices, or factual claims.

## Security Audit

- Audited at: 2026-07-19T11:18:31.808\+00:00
- Summary: One confirmed issue instructs unattended execution of an npx skill-update command without user approval. The remaining static matches are false positives caused by Markdown backticks, static CSS/reference content, local test code, and the bundled minified GSAP library.

## Stats

- Views: 0
- Downloads: 6
- Favorites: 0
- Popularity score: 0