Conversation Video

该 Skill 接收对话转写清单，调用 supertonic-tts 生成多语音音频与时间轴，再通过 ffmpeg 或 Remotion 模板渲染带说话人标签和字幕的对话视频。

pratyushchauhan

热度 33.91

最近更新 2026-06-08 12:28:23

分类：创作依赖：代码执行 / 读写文件应用场景：视频编辑

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add conversation-video

Metadata

name: conversation-video
description: Generate animated conversation videos with multi-voice TTS audio and timed text overlays. Use when the user needs to (1) turn a transcript or dialogue into a video with synced subtitles, (2) create podcast-style conversation visuals, (3) produce customer discovery interview videos, (4) animate multi-speaker content with distinct voices and styling, or (5) generate any video where spoken text appears on screen with speaker labels.

Conversation Video

Generate multi-voice conversation videos from text transcripts. Two paths: quick ffmpeg (no dependencies) or rich Remotion (React animations).

Prerequisites

Tool	Path / Notes
ffmpeg	System install or Jellyfin ffmpeg at `/usr/lib/jellyfin-ffmpeg/ffmpeg`
supertonic-tts	Python package for multi-voice TTS (see scripts/generate_audio.py for load logic)
Node.js + npm	Only needed for Remotion path

Workflow

1. Build a transcript manifest

Create a JSON file with your conversation:

[
  {"speaker": "NARRATOR",   "text": "Customer Discovery Interview", "voice": "M1", "speed": 1.0, "align": "center"},
  {"speaker": "INTERVIEWER","text": "Walk me through when you first realized...", "voice": "M5", "speed": 0.95, "align": "left"},
  {"speaker": "CUSTOMER",   "text": "I was looking for a marketer agent.", "voice": "M2", "speed": 1.0, "align": "right"}
]

Fields: speaker (label), text (spoken text), voice (supertonic voice name e.g. M1-M5, F1-F2), speed (optional playback speed), align (left/right/center for video placement).

2. Generate audio + timing manifest

python scripts/generate_audio.py manifest.json output.wav

Outputs:

output.wav — concatenated multi-voice audio
output_timings.json — per-segment start/end times for video sync

3. Render video (choose path)

Path A: ffmpeg — fast, no Node.js needed

python scripts/ffmpeg_render.py output_timings.json output.wav video.mp4

Options: --width, --height, --font-size, --bg, --font, --crf

Path B: Remotion — richer animations, React-based

Copy the boilerplate:

cp -r assets/remotion-boilerplate ./my-video
cd my-video
npm install

Edit src/Conversation.tsx:

Replace conversation array with your lines (duration in frames, 30fps)
Set SpeakerConfig colors/alignment
Uncomment <Audio src={staticFile("audio.wav")} /> and place audio in public/

Render:

npx remotion render src/index.ts Conversation out/video.mp4

Speaker Customization

Default color/alignment map (edit in either ffmpeg or Remotion):

Speaker	Color	Align
NARRATOR	#cbd5e1	center
INTERVIEWER	#60a5fa	left
CUSTOMER	#34d399	right

Add more by extending the config map in the respective renderer.

Resources

scripts/generate_audio.py — Multi-voice TTS with timing export
scripts/ffmpeg_render.py — ffmpeg drawtext video renderer
assets/remotion-boilerplate/ — Copyable Remotion project template
references/remotion-patterns.md — Advanced Remotion techniques (JSON data loading, word-by-word reveal, audio sync)
references/ffmpeg-guide.md — ffmpeg drawtext syntax and timing reference