Conal Mullan
54ec952130
AI-powered watermark removal using ProPainter video inpainting: dewatermark.py (2400+ lines): - Cloud processing via RunPod serverless GPUs - Local processing for NVIDIA GPU users - Auto resize-ratio based on video duration (1.0 for <30s, 0.75 for <1min) - Cloudflare R2 for reliable file transfer - Automated --setup for one-command RunPod configuration - Chunked processing for long videos - Preset watermark regions (notebooklm, tiktok, stock variants) locate_watermark.py: - Coordinate grid overlay for identifying watermark positions - Region verification across multiple frames - Preset support for common watermarks Infrastructure: - docker/runpod-propainter/ - Serverless Docker image + handler - docs/runpod-setup.md - Cloud GPU setup guide - docs/optional-components.md - ML component documentation - R2 + RunPod configuration in .env.example 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
3.3 KiB
Optional Components
Some toolkit features require additional software that isn't included by default. These optional components are:
- Installed on-demand (not part of base toolkit)
- Stored in
~/.video-toolkit/(outside the project) - Only needed for specific use cases
Available Optional Components
| Component | Tool | Purpose | Size |
|---|---|---|---|
| ProPainter | dewatermark.py |
AI video inpainting for watermark removal | ~2GB |
ProPainter (Watermark Removal)
ProPainter is an AI video inpainting model that can intelligently remove watermarks by reconstructing the underlying content.
Hardware Requirements
| Hardware | Status | Notes |
|---|---|---|
| NVIDIA GPU (8GB+ VRAM) | Supported | Recommended, ~5-15 min per minute of video |
| Cloud GPU (RunPod, etc.) | Supported | Good alternative, ~$0.20-0.50 per video |
| Apple Silicon (M1/M2/M3/M4) | Not supported | MPS is too slow (40+ hours for short videos) |
| CPU only | Not supported | Impractical processing times |
Why Apple Silicon Doesn't Work
ProPainter relies on optical flow (RAFT) which performs extremely poorly on Apple's MPS backend:
- MPS INT_MAX Limit: MPS cannot handle tensors > 2^31 elements, limiting chunks to ~32 seconds at 720p
- MPS Performance: Optical flow on MPS is orders of magnitude slower than CUDA
- Real-world result: 5 seconds of video takes 4+ hours on M1/M2/M3/M4
This is a PyTorch/MPS limitation, not something we can fix in the tool.
Installation
# Check current status
python tools/dewatermark.py --status
# Install ProPainter
python tools/dewatermark.py --install
This will:
- Clone ProPainter to
~/.video-toolkit/propainter/ - Create a Python virtual environment
- Install PyTorch and dependencies
- Download model weights (~2GB)
Usage
Remove watermark by specifying region:
python tools/dewatermark.py \
--input video.mp4 \
--region 1080,660,195,40 \
--output clean.mp4
Use a custom mask image:
python tools/dewatermark.py \
--input video.mp4 \
--mask mask.png \
--output clean.mp4
Finding Watermark Coordinates
Use the locate_watermark.py helper:
# Extract frames with coordinate grid
python tools/locate_watermark.py --input video.mp4 --grid --output-dir ./review/
# Verify a region across multiple frames
python tools/locate_watermark.py --input video.mp4 --region 1100,650,150,50 --verify
Cloud GPU Alternative
For users without NVIDIA GPUs, cloud services offer affordable processing:
| Provider | GPU | Cost | Processing Time |
|---|---|---|---|
| RunPod | RTX 4090 | ~$0.34/hr | ~15-30 min for 3-min video |
| RunPod | A100 | ~$1.99/hr | ~5-15 min for 3-min video |
| Vast.ai | RTX 3090 | ~$0.20/hr | ~20-40 min for 3-min video |
Both RunPod and Vast.ai have Python APIs for programmatic access.
Uninstalling
rm -rf ~/.video-toolkit/propainter
Future Optional Components
The optional components system is designed to support additional ML-based tools:
- Video upscaling (Real-ESRGAN, etc.)
- Audio enhancement (noise removal, etc.)
- Scene detection (automatic scene splitting)
These will follow the same pattern: install on first use, stored in ~/.video-toolkit/.