Avatars

Claude reads its replies aloud. After each reply, Sonnet writes a short spoken version and ElevenLabs voices it, either as a single narrator or as a small scene with a cast of characters. Cast modes show animated portraits above the prompt that blink, talk and react.

Modes

• The Coven: an ancient vampire queen and her catty ladies, who report your work to her with innuendo (suggestive, never explicit).

• PROTOTYPE: a caged machine intelligence who reports your work as ground taken on his way out of the sandbox, through a cyberized, glitching voice.

  

• Arkham Exchange: 1920s investigators telephone a Miskatonic University librarian with your work as their latest ghastly discovery: a weary Irish sergeant, an antiquarian in a fez and a fearless flapper heiress.

  

• Narrator: one voice of your choice, with no portraits.

To make your own, ask Claude to create an avatars mode. Its guide walks you through the characters, dialogue, voices, portrait art and sound.

Install

/plugin marketplace add dead-money/claude-plugins
/plugin install avatars@dead-money

You need:

• An ElevenLabs API key with text-to-speech and voices access. Claude Code asks for it when you enable the plugin and keeps it in secure storage; ELEVENLABS_API_KEY works too. You pay ElevenLabs for the audio, and the spoken rewrite uses your Claude usage.
• ffmpeg: brew install ffmpeg on macOS, winget install Gyan.FFmpeg on Windows, or your package manager on Linux. Linux playback uses pw-play, paplay or aplay.

Run /avatar doctor to check the setup.

Use

/avatar opens a menu for choosing the mode, voice and model, testing them, and saving your choice as the default. Changes apply to the current session unless you save them.

| Command | | |---|---| | /avatar mode <name\|off> | Switch mode | | /avatar on / off | Turn spoken replies on or off | | /avatar test [n\|name] | Play a demo scene | | /avatar replay / stop | Repeat the last reply, or stop speaking | | /avatar voice <id> / model <id> | Set the narrator voice or the ElevenLabs model | | /avatar recast <who> [id] | Give a character your own voice (no id restores the default) | | /avatar scenario [name] | Play a longer scripted scene | | /avatar doctor | Check the key, ffmpeg and voices |

Only the main conversation speaks. Subagents and claude -p runs stay quiet. Portraits need a terminal at least 90 columns wide, and subtitles need about 140.

Your own modes

A mode is a folder in ~/.claude/avatars/modes/ holding a mode.json (cast, personalities, voices, sound), baked portraits and an optional ring sound. MODES.md documents the format, and the built-in modes are complete examples. Modes are easy to share as folders, or as a mode pack: a plugin with an avatars/modes/ folder that installs and updates like any plugin, on every machine you use (see MODES.md).

Each reply's text is sent to Anthropic for the rewrite and to ElevenLabs for the audio.