Video Message

Generate avatar video messages from text or audio. Outputs as Telegram video notes (circular format).

Installation

npm install -g openclaw-avatarcam

Configuration

Configure in TOOLS.md:

### Video Message (avatarcam)

- avatar: default.vrm

- background: #00FF00

Settings Reference

Prerequisites

System Dependencies

Note: macOS and Windows don't need xvfb — they have native display support.

Docker Users

Add to OPENCLAW_DOCKER_APT_PACKAGES:

build-essential procps curl file git ca-certificates xvfb xauth libgbm1 libxss1 libatk1.0-0 libatk-bridge2.0-0 libgdk-pixbuf2.0-0 libgtk-3-0 libasound2 libnss3 ffmpeg

Usage

# With color background
avatarcam --audio voice.mp3 --output video.mp4 --background "#00FF00"

# With image background
avatarcam --audio voice.mp3 --output video.mp4 --background "./bg.png"

# With custom avatar
avatarcam --audio voice.mp3 --output video.mp4 --avatar "./custom.vrm"

Sending as Video Note

Use OpenClaw's message tool with asVideoNote:

message action=send filePath=/tmp/video.mp4 asVideoNote=true

Workflow

1. Read config from TOOLS.md (avatar, background)

2. Generate TTS if given text: tts text="..." → audio path

3. Run avatarcam with audio + settings → MP4 output

4. Send as video note via message action=send filePath=... asVideoNote=true

5. Return NO_REPLY after sending

Example Flow

User: "Send me a video message saying hello"

# 1. TTS
tts text="Hello! How are you today?" → /tmp/voice.mp3

# 2. Generate video
avatarcam --audio /tmp/voice.mp3 --output /tmp/video.mp4 --background "#00FF00"

# 3. Send as video note
message action=send filePath=/tmp/video.mp4 asVideoNote=true

# 4. Reply
NO_REPLY

Technical Details

Processing Pipeline

1. Electron renders VRM avatar with lip sync at 1280x720

2. WebM captured via canvas.captureStream(30)

3. FFmpeg processes: crop → fps normalize → scale → encode

4. Message tool sends via Telegram sendVideoNote API

Platform Support

Headless Rendering

Avatarcam auto-detects headless environments:

• Uses xvfb-run when $DISPLAY is not set (Linux only)

• macOS/Windows use native display

• GPU stall warnings are safe to ignore

• Generation time: ~1.5x realtime (20s audio ≈ 30s processing)

Notes

• Config is read from TOOLS.md

• Clean up temp files after sending: rm /tmp/video*.mp4

• For regular video (not circular), omit asVideoNote=true