Cm video factory

CM Video Factory v2.0 — Mass Video Production Engine (Private)

A self-hosted, batch-capable video factory built on Remotion React (code-to-video). Scripts → TTS → Subtitles → Render → Policy Check → Publish.

Dual format: 9:16 vertical (TikTok/Shorts/Reels) + 16:9 horizontal (YouTube). Dual language: Vietnamese (VieNeu-TTS) + English (ElevenLabs/OpenAI).

Spiritual twin of cm-content-factory, focused on Short-Form + Long-Form video at scale.

---

Core Philosophy

• Video is Math, Not Art. React components define every frame. JSON scripts define every word.
• VibeCoding Aesthetic. Dark SaaS visuals, cyan #00F0FF glow, Fira Code, engineering UI motifs.
• Content Mastery DNA. Every script uses SB7, Cialdini, Hook Formulas. No boring "Hello guys" intros.
• Batch-First. One command renders 10+ videos overnight. Queue, progress, retry, resume.
• Policy-Safe. Built-in guardrails prevent shadowbans, demonetization, and content strikes.

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                  📹 CM VIDEO FACTORY v2.0                    │
│            Batch Code-to-Video Production Engine             │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────┐   ┌──────────┐   ┌──────────┐   ┌────────┐  │
│  │  SCRIPT  │──▶│   TTS    │──▶│ SUBTITLE │──▶│ RENDER │  │
│  │ Generator│   │  Engine  │   │ @remotion/│   │ Engine │  │
│  │ (JSON)   │   │ VieNeu + │   │ captions  │   │ Batch  │  │
│  │ CM hooks │   │ ElevenL  │   │ SRT+JSON  │   │ Queue  │  │
│  └──────────┘   └──────────┘   └──────────┘   └────────┘  │
│       │              │              │              │        │
│  ┌────┴──────────────┴──────────────┴──────────────┴────┐  │
│  │              🎛️ FORMAT ROUTER                         │  │
│  │   ┌─────────────┐          ┌──────────────┐          │  │
│  │   │  VERTICAL   │          │  HORIZONTAL  │          │  │
│  │   │  9:16       │          │  16:9        │          │  │
│  │   │  1080×1920  │          │  1920×1080   │          │  │
│  │   │  TikTok     │          │  YouTube     │          │  │
│  │   │  Reels      │          │  Facebook    │          │  │
│  │   │  Shorts     │          │              │          │  │
│  │   └─────────────┘          └──────────────┘          │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                             │
│  ┌──────────────────────────────────────────────────────┐  │
│  │              🛡️ POLICY COMPLIANCE ENGINE              │  │
│  │  Content scan | AI labeling | Upload throttle | Hash │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                             │
│  ┌──────────────────────────────────────────────────────┐  │
│  │              📊 BATCH QUEUE MANAGER                   │  │
│  │  queue.json | progress.json | retry logic | cron     │  │
│  └──────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────┘

---

Quick Start

# 1. Setup Remotion project (one-time)
cd video-factory && npm install

# 2. Generate single video (interactive)
node scripts/pipeline.js --script scripts-input/video_01.json --format tiktok

# 3. Batch render (queue-based)
node scripts/batch-runner.js --queue queue.json --batch-size 5

# 4. Resume interrupted batch
node scripts/batch-runner.js --queue queue.json --resume

# 5. Preview in browser (hot reload)
npx remotion studio

---

7-Phase Execution Pipeline

Phase 1: 🔬 RESEARCH

Scrape documentation, GitHub repos, URLs. Build knowledge base per topic. Uses cm-cro-methodology to identify pain points for maximum hook impact.

Phase 2: ✍️ SCRIPT (JSON)

AI generates strictly-keyed JSON scripts. One JSON = one video.

{
  "id": "video_042",
  "title": "Debug mệt mỏi?",
  "hook": "99% Dev tốn thanh xuân để debug vì bỏ qua quy tắc này.",
  "hook_formula": "data_shock",
  "language": "vi",
  "duration_target": 45,
  "format": "tiktok",
  "scenes": [
    {
      "time_start": 0,
      "duration": 3.5,
      "text": "99% Dev tốn thanh xuân để debug vì...",
      "visual": "TerminalCrashView",
      "props": { "error": "SIGSEGV", "animation": "glitch" }
    },
    {
      "time_start": 3.5,
      "duration": 5,
      "text": "Với Cody Master, lỗi tự phơi bày trong 2 giây.",
      "visual": "CodeGlowView",
      "props": { "code": "cm-debugging --trace", "highlight": true }
    }
  ],
  "cta": "Click link ở Bio để cài đặt miễn phí.",
  "hashtags": ["#vibecoding", "#devtools", "#codymaster"],
  "metadata": {
    "ai_generated": true,
    "ai_disclosure": "Video created with AI-assisted animation and voice synthesis"
  }
}

Phase 3: 🎙️ TTS (Voice Synthesis)

TTS Engine Router — auto-detect language → best engine:

Language	Primary Engine	Fallback	Voice Cloning
Vietnamese	VieNeu-TTS 0.5B	OpenAI TTS	✅ (3-5s ref)
English	ElevenLabs multilingual_v2	OpenAI TTS	✅ (paid)

# Vietnamese TTS (VieNeu-TTS SDK)
from vieneu import Vieneu
tts = Vieneu()
audio = tts.infer(text="Xin chào, đây là VibeCoding Academy")
tts.save(audio, "audio_vi.wav")

# English TTS (ElevenLabs via infsh CLI)
infsh app run elevenlabs/tts --input '{
  "text": "Welcome to the VibeCoding revolution.",
  "voice": "george",
  "model": "eleven_multilingual_v2",
  "stability": 0.5,
  "similarity_boost": 0.8
}'

Outputs:

• audio.mp3 — Final audio track
• audio.json — Word-level timestamps (for subtitle sync)
• audio.srt — Standard SRT file (for YouTube upload)

Phase 4: 📝 SUBTITLE ENGINE (@remotion/captions)

Uses official Remotion captions package for frame-perfect sync:

import { parseSrt, createTikTokStyleCaptions } from "@remotion/captions";

// Parse SRT from TTS output
const captions = parseSrt({ input: srtContent });

// Create TikTok-style word-by-word highlights
const { pages } = createTikTokStyleCaptions({
  captions,
  combineTokensWithinMilliseconds: 800,
});

// Render in React component
const SubtitleOverlay: React.FC = () => {
  const frame = useCurrentFrame();
  const { fps } = useVideoConfig();
  const currentTime = (frame / fps) * 1000; // ms

  return (
    <AbsoluteFill style={{ justifyContent: "flex-end", padding: 40 }}>
      {pages.map((page) => (
        <CaptionPage key={page.startMs} page={page} currentTime={currentTime} />
      ))}
    </AbsoluteFill>
  );
};

Phase 5: 🎨 RENDER (Remotion)

Dual-format compositions:

// Root.tsx — Two compositions for two formats
export const RemotionRoot: React.FC = () => (
  <>
    <Composition
      id="TikTokVideo"
      component={TikTokComposition}
      width={1080}
      height={1920}
      fps={30}
      durationInFrames={durationFromAudio}
    />
    <Composition
      id="YouTubeVideo"
      component={YouTubeComposition}
      width={1920}
      height={1080}
      fps={30}
      durationInFrames={durationFromAudio}
    />
  </>
);

Render commands:

# TikTok vertical (9:16)
npx remotion render src/index.ts TikTokVideo out/tiktok_video_042.mp4

# YouTube horizontal (16:9)
npx remotion render src/index.ts YouTubeVideo out/youtube_video_042.mp4

# Benchmark optimal concurrency first
npx remotion benchmark --composition TikTokVideo

Remotion Performance Rules (from remotion-dev/skills):

• ✅ All animations driven by useCurrentFrame() — never useEffect/setInterval
• ✅ Memoize expensive computations with useMemo/useCallback
• ✅ Avoid GPU-heavy CSS (filter: blur(), drop-shadow()) in components
• ✅ Use npx remotion benchmark to find optimal --concurrency
• ✅ Video duration = getAudioDurationInSeconds() + 2s bumper
• ✅ Use --log=verbose to identify slow frames

Phase 6: 🛡️ POLICY CHECK

Pre-publish compliance scan (runs automatically before upload):

POLICY_GATES:
  platform_limits:
    tiktok:
      max_daily_uploads: 3
      cooldown_hours: 4
      required_labels: ["AI-generated"]
      hashtag_range: [3, 7]
    youtube:
      max_daily_uploads: 2
      cooldown_hours: 6
      required_disclosure: true    # Creator Studio AI toggle
      monetization_check: true     # Must show human creative value
    reels:
      max_daily_uploads: 3
      cooldown_hours: 4

  content_rules:
    min_visual_uniqueness: 70     # % difference between consecutive videos
    banned_content_scan: true     # Check for hate speech, misleading claims
    no_unauthorized_cloning: true # No cloning real people without consent
    no_mass_duplicate: true       # Each video visually distinct

What gets checked:

1. Script text scanned for banned keywords/topics
2. Visual uniqueness hash compared to last N renders
3. Upload rate verified against platform daily limits
4. AI disclosure metadata injected into file
5. Hashtag count validated (3-7 range)

Phase 7: 🚀 PUBLISH (Optional)

Multi-platform upload via APIs:

• YouTube Shorts/Long-form → YouTube Data API v3
• TikTok → TikTok Creator API
• Facebook/Instagram Reels → Graph API

Auto-applies: title, description, hashtags, AI disclosure labels, thumbnail.

---

Batch Queue Manager

For mass production, use the queue system:

// queue.json
{
  "batch_id": "batch_2026_03_30",
  "created_at": "2026-03-30T09:00:00Z",
  "config": {
    "formats": ["tiktok", "youtube"],
    "language": "vi",
    "tts_engine": "vieneu",
    "concurrency": 1,
    "retry_max": 3
  },
  "jobs": [
    {
      "id": "job_001",
      "script": "scripts-input/video_01.json",
      "status": "queued",
      "attempts": 0,
      "outputs": []
    },
    {
      "id": "job_002",
      "script": "scripts-input/video_02.json",
      "status": "queued",
      "attempts": 0,
      "outputs": []
    }
  ]
}

// progress.json (auto-generated)
{
  "batch_id": "batch_2026_03_30",
  "total": 10,
  "completed": 3,
  "failed": 0,
  "in_progress": 1,
  "queued": 6,
  "estimated_remaining_minutes": 42,
  "last_updated": "2026-03-30T09:25:00Z"
}

CLI:

# Start batch (processes sequentially)
node scripts/batch-runner.js --queue queue.json

# Resume after interruption
node scripts/batch-runner.js --queue queue.json --resume

# Status check
node scripts/batch-runner.js --queue queue.json --status

# Schedule via cron (every day at 2 AM)
echo "0 2 * * * cd /path/to/video-factory && node scripts/batch-runner.js --queue queue.json --resume" | crontab -

---

Component Library (VibeCoding Aesthetic)

Component	Visual Effect	Format	Use Case
DarkTerminal	MacOS-style window, typing animation, syntax colors	Both	bash commands, workflows
VsCodeGlow	Sleek IDE editor with neon-blue shadow	Both	Code demonstrations
BouncingSubtitle	TikTok-style bold text, word-by-word bounce	9:16	ALL spoken words (mandatory)
StatCounter	Rapidly spinning numbers in glassmorphic card	Both	Hook metrics ("10x", "200 articles")
SplitScreen	Top: Concept / Bottom: Code	16:9	Before/after, comparisons
CodeDiff	Side-by-side diff with green/red highlights	16:9	Showing improvements
FloatingCard	Glassmorphic card with parallax hover	Both	Feature callouts
ProgressRing	Animated circular progress indicator	Both	Stats, completion metrics

Strict Styling Rules:

• ❌ NEVER use light mode components
• Primary brand: Cyan #00F0FF
• Secondary: Hot Pink #FF007F
• Accent: Electric Purple #8B5CF6
• Background: Void Black #0A0A0A → Deep Navy #050B14
• Font: Fira Code (code), Inter (UI text), Be Vietnam Pro (Vietnamese text)
• All animations: interpolate() + spring() — NEVER CSS transitions

---

Hook & Persuasion Framework

Before writing Video JSON, select one:

Short-Form (0-60s) — TikTok/Shorts/Reels

Phase	Time	Goal
HOOK	0:00 - 0:03	Extreme curiosity gap. Fast cut. Zero intro.
AGITATE	0:03 - 0:15	Visualize common pain (red errors, spaghetti code)
SOLUTION	0:15 - 0:45	Enter CodyMaster. Drop code. Make it glow.
CTA	0:45 - 0:60	Grand Slam Offer. "Lấy miễn phí tại bio".

Long-Form (3-15min) — YouTube

Phase	Time	Goal
HOOK	0:00 - 0:30	Pattern interrupt + "Stay to the end for..."
CONTEXT	0:30 - 2:00	Problem landscape. Why this matters NOW.
DEEP DIVE	2:00 - 10:00	Step-by-step demonstration with code
PROOF	10:00 - 12:00	Results, metrics, before/after
CTA	Last 30s	Subscribe + comment + link in description

12 Hook Formulas

1. Data Shock — "99% Dev fail vì bỏ qua điều này"
2. Contrarian — "Đừng học React. Đây là lý do."
3. Future Pacing — "Trong 60 giây, bạn sẽ viết code nhanh gấp 10x"
4. Story Open — "Tháng trước, tôi mất 3 ngày debug..."
5. Authority — "Sau 10 năm code, đây là bí mật duy nhất"
6. Curiosity Gap — "Có 1 công cụ mà Google engineers dùng mà bạn chưa biết"
7. Fear — "AI sẽ thay thế developer? Chỉ nếu bạn không biết điều này"
8. Social Proof — "200+ engineers đã chuyển sang workflow này"
9. Challenge — "Thử debug trong 2 giây challenge"
10. Comparison — "Junior vs Senior: cùng 1 bug, khác nhau 10x tốc độ"
11. Reveal — "Công cụ bí mật đằng sau 50 dự án production"
12. Empathy — "Mình biết cảm giác code 12 tiếng mà không ra gì"

---

Platform Policy Summary

TikTok — What Gets You Shadowbanned

• ❌ Mass-posting identical/near-identical content
• ❌ Using 20+ generic hashtags (#fyp, #foryou spam)
• ❌ Unlabeled AI-generated realistic content
• ❌ Third-party tools for fake engagement
• ❌ Posting frequency > 5/day without variation
• ✅ Post 1-3 original videos/day with authentic engagement
• ✅ Use 3-5 niche-specific hashtags
• ✅ Label AI content using TikTok's built-in tools

YouTube — What Prevents Monetization

• ❌ Mass-produced templated AI content ("AI slop")
• ❌ Undisclosed realistic AI-generated content
• ❌ Unauthorized voice/face cloning
• ❌ Misleading titles/thumbnails
• ✅ Show human creative direction in every video
• ✅ Use AI disclosure toggle in Creator Studio
• ✅ Add unique commentary, editing, educational value
• ✅ Each video must demonstrate significant human involvement

---

Guidelines for Downstream Agents

1. Never hardcode strings in React components. All text from JSON/SRT for perfect TTS sync.
2. Animations = math. Use interpolate() and spring(). 30FPS = frame / 30 seconds.
3. Audio drives duration. Video length = getAudioDurationInSeconds() + 2s end-bumper.
4. Memoize everything. useMemo/useCallback on all expensive computations.
5. No GPU CSS. Avoid filter: blur(), drop-shadow() in rendered components.
6. Benchmark concurrency. Run npx remotion benchmark before batch renders.
7. Policy-first. Run policy check before ANY upload. No exceptions.
8. License check. VieNeu-TTS-0.5B = Apache 2.0 (commercial OK). VieNeu-TTS-0.3B = CC BY-NC 4.0 (non-commercial only).

---

Future Roadmap

• v2.1: VieNeu-TTS voice cloning integration (clone user's voice from 5s sample)
• v2.2: Remotion Lambda for parallel cloud rendering (10x speed)
• v2.3: AI Avatar mode via bytedance/omnihuman-1-5 lipsync
• v2.4: Auto A/B testing — render 2 hook variants, publish both, track engagement
• v3.0: Full inference.sh integration — text-to-video AI scenes (Veo 3, Seedance)

---

Usage Triggers

Use when the user states:

• "Start the video factory", "Create a TikTok/Shorts for this code"
• "Run the Remotion pipeline", "Make a VibeCoding video"
• "Generate a hook script for X", "Batch render videos"
• "Render YouTube video", "Create 9:16 vertical video"
• "tạo video hàng loạt", "render video tiktok"
• "video factory batch", "create video from script"