infinishield

English | 简体中文

A command-line tool for embedding invisible watermarks into images, SVGs, and videos.

What It Does

infinishield hides a short text message (e.g., a copyright notice) inside a file. The watermark is invisible, password-protected, and survives common modifications like cropping and compression.

Supported Formats

Format	Status	Max Message
JPEG, PNG, WebP, BMP, TIFF, GIF	Supported	7 bytes (cropping-resistant) or ~40 bytes (longer, no crop protection)
SVG	Supported	2-7 bytes (depends on path complexity)
MP4, WebM, MOV, AVI, MKV	Supported (optional build)	7 bytes

Building

Requires Rust 1.70+.

make release          # Images + SVG only
make release-video    # Images + SVG + Video (see below)

Video Support (optional)

Video requires extra system packages and takes longer to build (compiles FFmpeg from source):

# Linux (Debian/Ubuntu)
sudo apt-get install nasm pkg-config gcc make libx264-dev

# macOS
brew install nasm pkg-config x264

# Then build
make release-video

Usage

Embed

infinishield embed -i photo.jpg -o watermarked.png                      # image
infinishield embed -i logo.svg -o logo_wm.svg -m "Hi"                   # SVG
infinishield embed -i clip.mp4 -o watermarked.mp4                       # video
infinishield embed -i photo.jpg -m "MyMark" -p "secret" -o out.png      # custom message + password
infinishield embed -i photo.jpg -o out.png --dry-run                    # preview only

Option	Required	Default	Description
-i	yes	—	Input file
-o	yes	—	Output file
-m	no	"Infini"	Message to embed
-p	no	"d1ng0"	Password
--intensity	no	auto	Strength 1-10 (images only, see below)
--dry-run	no	—	Preview without writing

Intensity: When omitted, infinishield automatically selects optimal strength using a logarithmic curve based on image size — smaller images get lighter watermarks to stay invisible, larger images get stronger ones. When set manually (--intensity 1..10), your value scales the auto curve (1 = 50%, 5 = ~auto, 10 = 200%), so the watermark remains size-aware.

Verify

infinishield verify -i watermarked.png
infinishield verify -i watermarked.png -p "secret"

Help

infinishield              # full help
infinishield --version    # version

Best Practices for Preparing Assets

Images (JPEG/PNG):

• Use PNG output (-o out.png) for best watermark preservation. JPEG output is lossy and degrades the watermark.
• Messages up to 7 bytes (e.g., "Infini") survive cropping. For longer messages, cropping protection is lost.
• Images should be at least 512×512 pixels. Larger images produce better results with less visible impact.
• Rotation and resizing will destroy the watermark — only cropping and compression are tolerated.
• Auto intensity adapts to image size: small images (~0.3 MP) get subtle embedding, large images (8+ MP) get stronger embedding. Override with --intensity if needed.

SVG:

• Works best on SVGs with complex paths (illustrations, icons with curves). Simple geometric shapes (plain rectangles, circles) have too few coordinates to embed.
• Use --dry-run first to check if your SVG has enough qualifying paths.
• Don't open the watermarked SVG in editors that reformat or round coordinate values — this destroys the watermark. View-only or programmatic use is safe.

Video:

• Keep messages short (≤ 7 bytes). The watermark is embedded in 1 keyframe per second.
• Output is re-encoded with H.264 (CRF 18). The original codec is not preserved.
• On some videos with complex scenes, 1-2 bits of the message may flip during re-encoding. Detection is reliable but exact message recovery is not always guaranteed.
• Streaming architecture — only 1-2 frames in memory at a time. Handles videos of any length.
• The video binary includes a statically linked FFmpeg — no runtime dependencies needed.

Web UI

A browser-based testing interface for embedding and verifying watermarks. Requires Go 1.21+.

make webapp     # Build + run on http://localhost:1983

Features:

• Upload images, SVGs, or videos for watermarking
• Configure message, password, and intensity with real-time dry-run preview
• Side-by-side comparison of original vs watermarked assets
• Verify tab for extracting watermarks from files
• Event logging to webapp/logs/

Running Tests

make test       # Images + SVG tests
make sanity     # Full check: format + lint + build + all tests (including video)
make clean      # Remove build artifacts

Troubleshooting

Visible noise or artifacts after embedding

The watermark modifies pixel values in the green channel. On some images — especially small ones or images with large flat/uniform areas (sky, solid backgrounds, simple illustrations) — this can produce faint visible noise.

Try these steps in order:

1. Use a lower intensity. The default auto mode is tuned for a good balance, but you can reduce it:

   infinishield embed -i photo.jpg -o out.png --intensity 3
   infinishield embed -i photo.jpg -o out.png --intensity 1   # minimum strength

   Lower intensity = less visible noise but weaker watermark. Use --dry-run to preview before committing.

2. Use a larger source image. The watermark is spread across more pixels in larger images, making it less visible per pixel. If possible, watermark the full-resolution original rather than a resized thumbnail. Images under 512×512 pixels are not recommended.

3. Check the output format. Always use PNG output (-o out.png). JPEG re-encodes with lossy compression, which can amplify watermark artifacts. The input can be any supported format.

4. Try --dry-run first. Preview the embedding parameters without writing a file:

   infinishield embed -i photo.jpg -o out.png --dry-run

   This shows the mode, intensity, keypoint count, and capacity — useful for understanding what the tool will do.

5. Understand the image characteristics. Watermarks are naturally less visible in textured, detailed areas (landscapes, photographs) and more visible in flat, uniform areas (logos, line art, solid backgrounds). This is inherent to all spatial-domain watermarking.

Watermark not detected after cropping

• Only messages up to 7 bytes support cropping resistance (feature-point mode). Longer messages use global DWT mode, which has no cropping resistance.
• Cropping must preserve enough keypoint regions. Very aggressive cropping (> 75%) may remove too many feature points.
• Resizing or rotating the image will destroy the watermark — only cropping and compression are tolerated.

Wrong message extracted

• Make sure you use the same password for embed and verify (-p flag).
• Very low intensity (--intensity 1) on large images may cause occasional bit errors. Use a higher intensity or auto mode for reliable extraction.
• JPEG output degrades the watermark. Always verify from the PNG output, not a re-encoded JPEG.

Video watermark issues

• Video re-encodes with H.264 (CRF 18). On complex scenes, 1-2 bits may flip — detection is reliable but exact byte recovery is not always guaranteed.
• If verifying a trimmed video, ensure at least some watermarked keyframes remain. Keyframes are watermarked at 1 per second.

Documentation

• Technical Details — architecture, algorithms, what works and what doesn't