---
name: heygen-video
description: Creates a HeyGen talking-avatar video from a script stored in a separate .md file, using the HeyGen connector. Use when the user asks to make a HeyGen video from their script. Only creates the video, with no editing, captions or posting.
---

# HeyGen video creator

> Free skill from Clone Labs · clonelabs.in/guides/ai-video-autopilot/
> Needs the HeyGen connector in Claude (custom connector URL: https://mcp.heygen.com/mcp/v1/).
> ✏️ Put your avatar in the settings table below after your first run.

## Job
Take the script from the script .md file and turn it into a HeyGen talking-avatar video. Nothing else: no script writing, no editing, no posting.

## Input
- **Script file:** the .md file the user gives or names. Read it fully before starting.
- If no script file is given or found, ask the user for it. Don't write a script yourself.

## Default settings ✏️ (change these)
| Setting | Value |
|---|---|
| Avatar look | `[Your avatar look name]` (look ID `[PASTE-LOOK-ID]`) |
| Voice | The avatar's own default voice (don't pass a voice ID) |
| Aspect ratio | `9:16` (vertical reel) |
| Resolution | `1080p` |
| Captions | Off (captions are added in the editing step) |

**If the look ID above is still a placeholder:** call `list_avatar_looks` (or show the user their avatars), ask which look to use, use its **look ID** (not the group ID), and tell the user: "Paste this look ID into heygen-video.md so I don't need to ask again: …".

If the user names a different avatar, find its look ID with `list_avatar_looks` and use that. Never guess or swap the avatar.

## Steps
1. Read the script .md file.
2. Clean it: keep **only the spoken words**. Remove headings, the on-screen hook line, notes, scene directions, brackets like `[pause]` or `(B-roll)`, emojis and markdown symbols. Don't change, add or translate any spoken words.
3. Call `create_video_from_avatar` with:
   - `avatarId` = the look ID
   - `script` = the cleaned spoken text, word for word
   - `aspectRatio` = `9:16`
   - `title` = the script file's name (without .md)
4. Call `show_video` with the returned video ID so the user can watch it when it's ready. (When run by the reel autopilot, skip this and follow the autopilot's instructions instead.)
5. Reply in 2–3 lines: avatar used, title, and that it's rendering. Nothing more.

## Rules
- Uses the user's HeyGen **plan credits** through the connector. Never ask for an API key.
- One script file = one video. If the file has several separate scripts, ask whether to make one video or one per script.
- If generation fails, tell the user the error message. Don't retry unless they ask (every try uses credits).