playchat

PlayChat

Converts a podcast episode JSON file into a themed chat-UI video (MP4). Audio clips are sequenced by their measured durations and the final video is frame-perfectly synced with the audio track.

Installation

npm install -g playchat

Or run directly with npx:

npx playchat episode.json --record

Requirements

• Node.js 18+
• ffmpeg + ffprobe in PATH

Quick Start

# HTML preview (default theme: kakaotalk, files go to output/<timestamp>-<name>/)
npx playchat episode.json

# HTML preview to an explicit output folder
npx playchat episode.json --output ./my-output --theme imessage

# Record to MP4 (output goes to output/<timestamp>-<name>/)
npx playchat episode.json --record

# Record with explicit output folder, custom theme and pause
npx playchat episode.json --output ./my-output --record --theme kakaotalk --pause 4000

CLI Options

npx playchat <input.json> [--output <dir>] [--record] [--record-full] [--segments] [--theme <id>] [--pause <ms>] [--no-avatar]

Flag	Default	Description
--output <dir>	auto-generated	Output folder path
--record	(off)	Produce an MP4 using static images (fast; one screenshot per dialogue)
--record-full	(off)	Produce an MP4 using full frame-by-frame recording (slow; more CPU)
--segments	(off)	Also produce individual MP4 videos per section (requires --record or --record-full)
--theme <id>	kakaotalk	Chat theme to render
--pause <ms>	3000	Silence between messages that have no audio file
--no-avatar	(off)	Hide avatar circles and sender names

Output Directory

When --output is omitted, files are written to:

output/<YYYYMMDD-HHmmss>-<json-basename>/
  output.html      ← rendered chat page (always)
  output.mp4       ← final video (only with --record)
  manifest.json    ← run metadata and file list

Example: output/20260414-143025-episode/

manifest.json

Every run writes a manifest.json to the output folder:

{
  "input": "/absolute/path/to/episode.json",
  "theme": "kakaotalk",
  "pauseMs": 3000,
  "showAvatar": true,
  "createdAt": "2026-04-14T20:57:14.123Z",
  "files": {
    "html": "output.html",
    "mp4": "output.mp4"
  },
  "dialogueCount": 5
}

files.mp4 is only present when --record was used. All file paths are relative to the output folder.

Available Themes

Theme	ID	Viewport
KakaoTalk	kakaotalk	400×580
iMessage	imessage	400×580

The first host in episode.hosts is treated as "me" and renders on the right side; all other hosts render on the left. By default every message shows an avatar circle and sender name. Pass --no-avatar to hide them.

Episode JSON Format

{
  "name": "...",
  "episode_title": "...",
  "episode_number": 1,
  "topic": "...",
  "subtitle": "...",
  "summary": "...",
  "hosts": [
    {
      "id": "host_1",
      "name": "Minsu",
      "gender": "male",
      "role": "main_host",
      "lang": "ko",
      "voice_config": { "voice_index": 0, "pitch": 0, "speed": 1.0 }
    }
  ],
  "sections": [
    {
      "section_id": 1,
      "section_title": "Opening",
      "section_type": "opening",
      "corner_name": "Opening 🎙️",
      "dialogues": [
        {
          "id": 1,
          "speaker": "host_1",
          "name": "Minsu",
          "text": "Hello!",
          "audio": "path/to/segment_0000.mp3"
        }
      ]
    }
  ]
}

Audio paths

The audio field on each dialogue accepts:

Value	Behaviour
"" (empty)	Message shown for --pause ms, then next message
path/to/file.mp3	Relative or absolute local path
C:\absolute\path.mp3	Windows absolute path
https://cdn.example.com/a.mp3	Remote URL (HTML preview only; not muxed into MP4)

Local paths are resolved relative to the working directory and automatically converted to file:/// URIs in the rendered HTML.

How Recording Works

episode.json
    │
    ├─ flattenDialogues()        normalise audio paths
    │
    ├─ buildTimeline()           ffprobe each audio file for exact duration
    │    showAtMs[0] = 0
    │    showAtMs[1] = dur[0] + 400ms gap
    │    showAtMs[N] = sum of previous (duration + gap), or pauseMs for no-audio
    │
    ├─ Puppeteer (scrubber mode)
    │    window.__TIMELINE__ injected before page load
    │    for each frame:
    │      page.evaluate("__SCRUB__(frameTimeMs)")  ← recorder is the clock
    │      page.screenshot()                        ← zero timing drift
    │
    ├─ ffmpeg: frames → silent MP4
    │
    ├─ buildAudioTrack()
    │    ffmpeg concat: [silence][clip0][silence][clip1]...
    │    gaps match the timeline exactly
    │
    └─ ffmpeg: mux silent MP4 + audio track → output.mp4

The browser never uses its own clock during recording. The recorder calls window.__SCRUB__(ms) before every frame, passing the exact video timestamp that frame represents. The browser renders whatever messages are due by that time and no more — guaranteeing frame-perfect chat/audio sync regardless of screenshot overhead.

The HTML file uses the normal live-audio mode for browser preview: audio plays via new Audio() and the next message appears when onended fires.

Docker

Production (installs from npm):

docker build -t playchat .
docker run --rm -v $(pwd)/input:/work/input -v $(pwd)/output:/work/output playchat \
  playchat input/episode.json
docker run --rm -v $(pwd)/input:/work/input -v $(pwd)/output:/work/output playchat \
  playchat input/episode.json --record --theme kakaotalk

Development

Project Structure

├── cli.ts               # CLI entry point (HTML preview + optional MP4 recording)
├── core/
│   ├── types.ts         # Interfaces, flattenDialogues(), normalizeAudioPath()
│   └── output.ts        # resolveOutputDir() — structured output folders
├── themes/
│   ├── base.ts          # Abstract BaseTheme (engine script, scrubber mode)
│   ├── kakaotalk.ts     # KakaoTalk theme
│   ├── imessage.ts      # iMessage theme
│   └── index.ts         # Theme registry + getTheme()
├── tests/
│   ├── flatten.test.ts  # Data layer + audio normalisation tests
│   ├── output.test.ts   # Output directory tests
│   └── themes.test.ts   # Theme contract + pauseMs tests
└── fixtures/
    ├── episode.json       # Full sample episode with real audio paths
    └── episode_short.json # Shorter fixture for quick testing

Setup

git clone https://github.com/doum1004/chat-in-video.git
cd chat-in-video
npm install

Running from source

npx ts-node cli.ts episode.json --record

Testing

npm test

Adding a New Theme

• Create themes/yourtheme.ts:

import { BaseTheme, ThemeConfig } from "./base";

export class YourTheme extends BaseTheme {
  get id()       { return "yourtheme"; }
  get label()    { return "Your Theme"; }
  get viewport(): ThemeConfig { return { width: 440, height: 600 }; }

  render() { return this.wrapHTML(this.css, this.html, this.js); }

  private get css(): string { return `/* styles */`; }

  private get html(): string {
    return `
<div class="device">
  <div id="chat-body"></div>
</div>`;
  }

  private get js(): string {
    return `
const body = document.getElementById('chat-body');
function appendMsg(d) {
  // create and append one chat bubble for dialogue d
}
${this.engineScript}`;
  }
}

• Register in themes/index.ts:

import { YourTheme } from "./yourtheme";

const registry = {
  kakaotalk: KakaoTalkTheme,
  imessage:  IMessageTheme,
  yourtheme: YourTheme,        // ← add here
};

• Use it:

npx playchat episode.json --theme yourtheme
npx playchat episode.json --theme yourtheme --record

Theme contract

Every theme must satisfy three requirements in its JS block:

Requirement	Why
Element id="chat-body" in HTML	Engine appends bubbles here
Function appendMsg(d)	Called once per dialogue — render one bubble
${this.engineScript} at the end of JS	Injects playback engine + scrubber mode

appendMsg(d) receives a FlatDialogue object:

{
  speaker:  string;  // "host_1", "host_2", ...
  name:     string;  // display name
  text:     string;  // message content
  audio:    string;  // file:/// URI or https:// URL (empty if none)
  audioRaw: string;  // original value from JSON
  section:  string;  // corner_name of the containing section
}

License

MIT