wan-video-clone

WAN 2.2 AI Video Clone Pipeline (Chunked)

Takes Kevin's video in, splits it into ~30s chunks, runs each through the AI pipeline, stitches them back together. Out the other side: Kev's Assistant talking in Kevin's chair with Kev's Assistant's voice.

Pipeline

Kevin's Video
    │
    ▼
┌─────────────────────┐
│  ffmpeg split (~5s)  │
└────────┬────────────┘
         │
    ┌────▼────┐
    │ chunk_0 │  chunk_1  chunk_2  ...
    └────┬────┘
         │  For each chunk:
         ├─ Extract audio (ffmpeg local)
         ├─ Voice clone → Kev's Assistant (ChatterboxHD / fal.ai)
         ├─ Merge cloned audio back (ffmpeg local)
         ├─ Upload merged chunk to fal.ai
         ├─ WAN 2.2 animate/replace (Kev's Assistant face + chunk)
         └─ Download output chunk
         │
    ┌────▼────────────────┐
    │  ffmpeg stitch all   │
    └────────┬─────────────┘
             │
             ▼
      Kev's Assistant's Video

Prerequisites

• FAL_KEY env var (fal.ai API key)
• ffmpeg installed
• assets/kevs-reference.png — Kev's Assistant face reference image
• assets/kevs-voice-sample.mp3 — Kev's Assistant voice sample for cloning
• Optional: REPLICATE_API_TOKEN (not needed if using local ffmpeg for audio)

Usage

# Basic — clone a video
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "https://example.com/kevin-video.mp4"

# Local file
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video /path/to/kevin.mp4

# Custom chunk size (default 5s)
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "URL" --chunk-duration 5

# Process 2 chunks at a time (parallel)
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "URL" --parallel 2

# Skip voice clone (face only)
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "URL" --skip-voice

# Skip face swap (use Kev's Assistant ref image directly with WAN)
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "URL" --skip-face-swap

# Dry run (show plan without executing)
node skills/wan-video-clone/scripts/wan-video-clone.js \
  --video "URL" --dry-run

Flags

Flag	Default	Description
--video	required	Source video URL or local path
--chunk-duration	30	Seconds per chunk
--parallel	1	Max concurrent chunk processing
--resolution	720p	WAN output resolution
--kevs-image	auto	Kev's Assistant face image URL (uses asset if not set)
--kevs-voice	auto	Kev's Assistant voice sample URL (uses asset if not set)
--prompt	auto	Face swap prompt for Nano Banana
--skip-voice	false	Skip voice cloning step
--skip-face-swap	false	Use Kev's Assistant ref image directly
--output	output/wan-clones/	Output directory
--dry-run	false	Show plan without executing

Timing

• Face swap: ~1 min (one-time)
• Per chunk: ~15-20 min (WAN animate is the bottleneck)
• A 5 min video = 10 chunks × ~20 min = ~3.5 hours sequential, ~2 hours with parallel=2
• A 30s video = 1 chunk = ~20-25 min total
• Stitching: seconds
• IMPORTANT: Source video must be H.264 codec. AV1/VP9 causes WAN to hang. Script auto-re-encodes if needed.

Output

Final video saved to output/wan-clones/<timestamp>/kevs-assistant-clone-<timestamp>.mp4

Key APIs

• fal.ai Nano Banana Pro — face swap (fal-ai/nano-banana-pro/edit)
• fal.ai ChatterboxHD — voice clone (resemble-ai/chatterboxhd/speech-to-speech)
• fal.ai WAN 2.2 — animate/replace (fal-ai/wan/v2.2-14b/animate/replace)
• ffmpeg — split, extract audio, merge audio, stitch (all local, no API)