AI Video Factory — Track A (Free Local Pipeline)

Automated faceless video essay generator for TikTok & YouTube. Total cost: $0/month

Pipeline Flow

script.json → Kokoro TTS → Whisper Sync → FFmpeg Render → final.mp4

Quick Start (Windows 11)

Step 1 — Install FFmpeg

1. Download from: https://www.gyan.dev/ffmpeg/builds/
2. Get ffmpeg-release-essentials.zip
3. Extract to C:\ffmpeg\
4. Add C:\ffmpeg\bin to your Windows PATH:
   • Search "Edit the system environment variables" in Start
   • Click Environment Variables → Edit PATH → New → C:\ffmpeg\bin
   • Restart your terminal

Step 2 — Run Setup

setup.bat

This installs all Python dependencies automatically.

Step 3 — Add Assets

Background videos (put in backgrounds/ folder):

• Download 3-5 free dark abstract loops from Pexels: https://www.pexels.com/search/videos/dark%20abstract/
• Recommended searches: "dark particles", "neural network", "space loop"

Drone music (put in music/ folder):

• Download from YouTube Audio Library: https://www.youtube.com/audiolibrary
• Search: "ambient drone" or "dark atmospheric"
• Pick any track under 5 minutes

Step 4 — Run Tests

python pipeline.py --test

Step 5 — Generate Video

python pipeline.py

Your video will be at output/final.mp4 — ready for TikTok/YouTube.

---

Individual Stages

Run stages independently for debugging:

# TTS only — generates voice audio
python pipeline.py --tts

# Sync only — extracts word timestamps
python pipeline.py --sync

# Render only — builds final video
python pipeline.py --render

---

File Structure

video_factory/
├── setup.bat              ← Run this first
├── pipeline.py            ← Main orchestrator
├── tts_kokoro.py          ← Kokoro-82M voice synthesis
├── sync_whisper.py        ← Whisper word timestamps
├── render_ffmpeg.py       ← FFmpeg video renderer
├── script.json            ← Your 34-scene script
├── backgrounds/           ← Drop dark loop .mp4 files here
├── music/                 ← Drop drone .mp3 track here
├── output/
│   ├── audio/             ← Generated voice WAV files
│   ├── sync/              ← Word timestamp JSON files
│   ├── scenes/            ← Individual scene MP4s
│   └── final.mp4          ← YOUR FINISHED VIDEO
└── tests/
    └── test_pipeline.py   ← Full TDD test suite

---

Output Specs

Setting	Value
Resolution	1080x1920 (9:16 vertical)
FPS	30
Format	MP4 (H.264 + AAC)
Voice	Kokoro-82M — calm, philosophical
Text	Word-by-word flash cards, white + teal highlights
Music	Drone mixed at -22dB under voice
GPU	NVIDIA NVENC auto-detected, CPU fallback

---

Customization

Change the voice

Edit tts_kokoro.py line 18:

VOICE = "af_heart"   # calm female (default)
# VOICE = "am_adam"  # deep male
# VOICE = "af_sky"   # bright female

Change flash card size

Edit sync_whisper.py line 17:

fragment_size=3   # 3 words per card (default)
# fragment_size=4 # slower, more words per flash
# fragment_size=1 # word-by-word (most aggressive)

Change emphasis color

Edit render_ffmpeg.py line 30:

EMPHASIS_COLOR = "#00e5c8"  # Tinlance teal (default)
# EMPHASIS_COLOR = "#ff4d6d"  # Tinlance pink
# EMPHASIS_COLOR = "#ffff00"  # Yellow

Add new emphasis words

Edit render_ffmpeg.py line 34:

EMPHASIS_WORDS = {
    "hunting", "survival", "predator", ...
    # Add your own keywords here
}

---

Scaling to Volume

To generate a new video:

1. Replace script.json with new 34-scene script
2. Run python pipeline.py
3. Move output/final.mp4 to your upload folder
4. Repeat

For bulk automation, run multiple scripts sequentially:

for %%f in (scripts\*.json) do (
    copy %%f script.json
    python pipeline.py
    move output\final.mp4 output\%%~nf.mp4
)

---

Troubleshooting

FFmpeg not found: Add C:\ffmpeg\bin to PATH (see Step 1)

Kokoro model not downloading: Check your internet connection — model downloads automatically on first run (~330MB)

Whisper slow on CPU: Normal — base model takes ~30s per scene on CPU. Upgrade to NVIDIA GPU for 10x speed.

Black background instead of loop: Add .mp4 files to backgrounds/ folder

No music in output: Add .mp3 file to music/ folder