CalvinBackup/reverse-SynthID
1
0
Fork 0
mirror of https://github.com/aloshdenny/reverse-SynthID.git synced 2026-07-17 20:57:20 +02:00
Code Issues Packages Projects Releases Wiki Activity

updated carriers

Browse Source
This commit is contained in:
Alosh Denny
2026-03-28 18:14:51 +05:30
parent f2ace3dead
commit a9079c74c4
6 changed files with 877 additions and 894 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+198 -237
View File
@@ -2,7 +2,7 @@
<img src="assets/synthid_watermark.png" alt="SynthID Watermark Analysis" width="100%">
</p>
<h1 align="center">🔍 Reverse-Engineering SynthID</h1>
<h1 align="center">Reverse-Engineering SynthID</h1>
<p align="center">
<b>Discovering, detecting, and surgically removing Google's AI watermark through spectral analysis</b>
@@ -12,147 +12,127 @@
<img src="https://img.shields.io/badge/Python-3.10+-blue?style=flat-square&logo=python" alt="Python">
<img src="https://img.shields.io/badge/License-Research-green?style=flat-square" alt="License">
<img src="https://img.shields.io/badge/Detection_Rate-90%25-success?style=flat-square" alt="Detection">
<img src="https://img.shields.io/badge/Bypass_V3-PSNR_40dB+-blueviolet?style=flat-square" alt="Bypass V3">
<img src="https://img.shields.io/badge/V3_Bypass-PSNR_43dB+-blueviolet?style=flat-square" alt="V3 Bypass">
<img src="https://img.shields.io/badge/Phase_Coherence_Drop-91%25-red?style=flat-square" alt="Phase Drop">
</p>
---
## 🎯 Overview
## Overview
This project reverse-engineers **Google's SynthID** watermarking system - the invisible watermark embedded into every image generated by Google Gemini. Using only signal processing and spectral analysis (no access to the proprietary encoder/decoder), we:
1. **Discovered** the watermark's exact frequency-domain structure
1. **Discovered** the watermark's resolution-dependent carrier frequency structure
2. **Built a detector** that identifies SynthID watermarks with 90% accuracy
3. **Developed a spectral bypass** (V3) that surgically removes watermark components while preserving image quality at **40+ dB PSNR**
3. **Developed a multi-resolution spectral bypass** (V3) that achieves **75% carrier energy drop**, **91% phase coherence drop**, and **43+ dB PSNR** on any image resolution
### What Makes This Different
Unlike brute-force approaches (JPEG compression, noise injection), our V3 bypass uses a **SpectralCodebook** - a fingerprint of the watermark's exact frequency signature - extracted from reference images. This allows surgical, frequency-bin-level removal rather than blind signal destruction.
Unlike brute-force approaches (JPEG compression, noise injection), our V3 bypass uses a **multi-resolution SpectralCodebook** - a collection of per-resolution watermark fingerprints stored in a single file. At bypass time, the codebook auto-selects the matching resolution profile, enabling surgical frequency-bin-level removal on any image size.
---
## 🔬 Key Findings
## Key Findings
### The Watermark is a Fixed Spectral Pattern
### The Watermark is Resolution-Dependent
By generating pure black and white images through Google Gemini, we isolated the watermark signal from content. The results are striking:
SynthID embeds carrier frequencies at **different absolute positions** depending on image resolution. A codebook built at 1024x1024 cannot directly remove the watermark from a 1536x2816 image - the carriers are at completely different bins.
<p align="center">
<img src="assets/synthid_black.jpg" alt="Watermark on black background" width="45%">
<img src="assets/synthid_white.jpg" alt="Watermark on white background" width="45%">
</p>
| Resolution | Top Carrier (fy, fx) | Coherence | Source |
|:----------:|:--------------------:|:---------:|:------:|
| **1024x1024** | (9, 9) | 100.0% | 100 black + 100 white refs |
| **1536x2816** | (768, 704) | 99.6% | 88 watermarked content images |
<p align="center">
<i>Left: SynthID watermark extracted from a pure-black Gemini image (enhanced 100×). Right: Same watermark on a white background. The diagonal stripe pattern and carrier frequencies are clearly visible.</i>
</p>
### Carrier Frequency Discovery
The watermark embeds energy at specific carrier frequencies with **>99.9% phase coherence** across all images:
| Carrier Frequency (fy, fx) | Phase Coherence | Magnitude | Phase (rad) |
|:--------------------------:|:---------------:|:---------:|:-----------:|
| **(±14, ±14)** | 99.96% | 16,807 | ±1.44 |
| **(±126, ±14)** | 99.96% | 8,046 | ±2.37 |
| **(±98, ∓14)** | 99.94% | 6,283 | ±0.61 |
| **(±128, ±128)** | 99.25% | 6,908 | ±2.29 |
| **(±210, ∓14)** | 99.96% | 6,032 | ±1.13 |
| **(±238, ±14)** | 99.90% | 4,190 | ±1.61 |
> **Key insight:** Most carriers cluster along the `y = ±14` line in frequency space, suggesting a structured frequency selection algorithm. The diagonal stripe pattern visible in the enhanced images corresponds to these carrier frequencies.
This is why the V3 codebook stores **separate profiles per resolution** and auto-selects at bypass time.
### Phase Consistency - A Fixed Model-Level Key
The watermark's phase template is **identical across all images** from the same Gemini model:
- **Green channel phase std**: < 0.007 radians across 50 reference images
- **Cross-image correlation**: 21.8% mean pairwise noise correlation
- **Noise structure ratio**: 1.32 ± 0.02 (byproduct of the neural encoder)
- **Green channel** carries the strongest watermark signal
- **Cross-image phase coherence** at carriers: >99.5%
- **Black/white cross-validation** confirms true carriers via |cos(phase_diff)| > 0.90
This means SynthID does not embed per-image messages - it uses a **fixed spectral fingerprint** that can be profiled and subtracted.
### Carrier Frequency Structure
### Frequency Spectrum Analysis
At 1024x1024 (from black/white refs), top carriers lie on a low-frequency grid:
<p align="center">
<img src="artifacts/visualizations/synthid_watermark_spectrum.png" alt="FFT Magnitude Spectrum" width="35%">
<img src="artifacts/visualizations/synthid_watermark_carriers.png" alt="Carrier Frequencies" width="35%">
</p>
<p align="center">
<i>Left: FFT magnitude spectrum showing bright carrier frequency peaks. Right: Reconstructed carrier pattern showing the diagonal structure.</i>
</p>
| Carrier (fy, fx) | Phase Coherence | B/W Agreement |
|:-----------------:|:---------------:|:-------------:|
| (9, 9) | 100.00% | 1.000 |
| (5, 5) | 100.00% | 0.993 |
| (10, 11) | 100.00% | 0.997 |
| (13, 6) | 100.00% | 0.821 |
<p align="center">
<img src="artifacts/visualizations/deep_analysis/frequency_analysis.png" alt="Frequency Analysis" width="80%">
</p>
<p align="center">
<i>Detailed frequency analysis: Average magnitude spectrum (left) and phase coherence map (right). The carrier positions are marked with crosshairs.</i>
</p>
At 1536x2816 (from random watermarked content), carriers are at much higher frequencies:
| Carrier (fy, fx) | Phase Coherence |
|:------------------:|:---------------:|
| (768, 704) | 99.55% |
| (672, 1056) | 97.46% |
| (480, 1408) | 96.55% |
| (384, 1408) | 95.86% |
---
## 🏗️ Architecture
## Architecture
### Three Generations of Bypass
| Version | Approach | PSNR | Detection Impact | Status |
| Version | Approach | PSNR | Watermark Impact | Status |
|:-------:|:---------|:----:|:----------------:|:------:|
| **V1** | JPEG compression (Q50) | 37 dB | ~11% phase drop | ✅ Baseline |
| **V2** | Multi-stage transforms (noise, color, frequency) | 27-37 dB | ~0% confidence drop | ✅ Quality trade-off |
| **V3** | Spectral codebook subtraction | **33-43 dB** | **1-7% confidence drop** | ✅ Best quality |
| **V1** | JPEG compression (Q50) | 37 dB | ~11% phase drop | Baseline |
| **V2** | Multi-stage transforms (noise, color, frequency) | 27-37 dB | ~0% confidence drop | Quality trade-off |
| **V3** | **Multi-resolution spectral codebook subtraction** | **43+ dB** | **91% phase coherence drop** | **Best** |
### V3 Pipeline (Spectral Bypass)
### V3 Pipeline (Multi-Resolution Spectral Bypass)
```
Input Image → FFT per channel → Estimate Watermark → Subtract → IFFT → Clip → Output
↑
SpectralCodebook
(25 black + 25 white refs)
Input Image (any resolution)
│
▼
codebook.get_profile(H, W) ──► exact match? ──► FFT-domain subtraction
│ (fast path)
└─ no exact match ──────► spatial-domain resize + subtraction
(fallback path)
│
▼
Multi-pass iterative subtraction (aggressive → moderate → gentle)
│
▼
Anti-alias → Output
```
1. **SpectralCodebook** profiles the watermark from reference images (pure black/white Gemini outputs)
2. **Selective notch filter** targets only high-magnitude (P97+), high-consistency (≥95%) frequency bins
3. **Safe magnitude cap** limits subtraction to 30% of image energy per bin - preserving content
4. **Content-adaptive scaling** adjusts subtraction based on image luminance
1. **SpectralCodebook** stores resolution-specific profiles (carrier positions, magnitudes, phases)
2. **Auto resolution selection** picks the exact profile or the closest match
3. **Direct known-signal subtraction** weighted by phase consistency and cross-validation confidence
4. **Multi-pass schedule** catches residual watermark energy missed by previous passes
5. **Per-channel weighting** (G=1.0, R=0.85, B=0.70) matches SynthID's embedding strength
---
## 📊 Results
## Results (V3 on 88 Gemini Images)
### V3 Bypass Quality (PSNR in dB - higher is better)
### Aggregate Metrics (1536x2816, aggressive strength)
| Image | Gentle | Moderate | Aggressive | Maximum |
|:-----:|:------:|:--------:|:----------:|:-------:|
| 121407 | **42.9** | 41.4 | 40.0 | 39.3 |
| 110802 | 33.4 | 33.3 | 33.1 | 33.0 |
| 131614 | **38.4** | 38.0 | 37.5 | 37.2 |
| 119198 | **38.4** | 37.7 | 37.0 | 36.6 |
| 12085 | **42.5** | 41.6 | 40.7 | 40.3 |
| Metric | Value |
|:-------|------:|
| **PSNR** | 43.5 dB |
| **SSIM** | 0.997 |
| **Carrier energy drop** | 75.8% |
| **Phase coherence drop** (top-5 carriers) | **91.4%** |
> All images maintain **>33 dB PSNR** - visually indistinguishable from the original.
### Quality Across Resolutions
### Watermark Detection Confidence
| Image | Before | After (Aggressive) | Drop |
|:-----:|:------:|:------------------:|:----:|
| 121407 | 0.394 | 0.387 | 1.7% |
| 131614 | 0.437 | 0.422 | 3.4% |
| 12085 | 0.394 | 0.366 | **7.2%** |
| 119198 | 0.389 | 0.386 | 0.9% |
### Sample Output
<p align="center">
<img src="assets/sample_watermarked.png" alt="Original watermarked image" width="45%">
<img src="assets/sample_cleaned.png" alt="V3 bypass output" width="45%">
</p>
<p align="center">
<i>Left: Original SynthID-watermarked Gemini image. Right: After V3 spectral bypass - visually identical, watermark energy reduced.</i>
</p>
| Resolution | Match | PSNR | SSIM |
|:----------:|:-----:|:----:|:----:|
| 1536x2816 | exact | 44.9 dB | 0.996 |
| 1024x1024 | exact | 39.8 dB | 0.977 |
| 768x1024 | fallback | 40.6 dB | 0.994 |
---
## 🚀 Quick Start
## Quick Start
### Installation
@@ -165,94 +145,105 @@ source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
```
### 1. Build Detection Codebook
### 1. Build Multi-Resolution Codebook
From the CLI:
```bash
python src/extraction/robust_extractor.py extract /path/to/watermarked/images \
--output artifacts/codebook/robust_codebook.pkl
python src/extraction/synthid_bypass.py build-codebook \
--black gemini_black \
--white gemini_white \
--watermarked gemini_random \
--output artifacts/spectral_codebook_v3.npz
```
### 2. Detect Watermark
Or from Python:
```python
from src.extraction.synthid_bypass import SpectralCodebook
codebook = SpectralCodebook()
# Profile 1: from black/white reference images (1024x1024)
codebook.extract_from_references(
black_dir='gemini_black',
white_dir='gemini_white',
)
# Profile 2: from watermarked content images (1536x2816)
codebook.build_from_watermarked('gemini_random')
codebook.save('artifacts/spectral_codebook_v3.npz')
# Saved with profiles: [1024x1024, 1536x2816]
```
### 2. Run V3 Bypass (Any Resolution)
```python
from src.extraction.synthid_bypass import SynthIDBypass, SpectralCodebook
codebook = SpectralCodebook()
codebook.load('artifacts/spectral_codebook_v3.npz')
bypass = SynthIDBypass()
result = bypass.bypass_v3(image_rgb, codebook, strength='aggressive')
print(f"PSNR: {result.psnr:.1f} dB")
print(f"Profile used: {result.details['profile_resolution']}")
print(f"Exact match: {result.details['exact_match']}")
```
From the CLI:
```bash
python src/extraction/synthid_bypass.py bypass input.png output.png \
--codebook artifacts/spectral_codebook_v3.npz \
--strength aggressive
```
**Strength levels:** `gentle` (minimal, ~45 dB) > `moderate` > `aggressive` (recommended) > `maximum`
### 3. Detect Watermark
```bash
python src/extraction/robust_extractor.py detect image.png \
--codebook artifacts/codebook/robust_codebook.pkl
```
```
Detection Results:
Watermarked: True
Confidence: 0.95
Phase Match: 0.6683
```
### 3. Build Spectral Codebook (V3)
```python
from synthid_bypass import SpectralCodebook
codebook = SpectralCodebook()
codebook.extract_from_references(
black_dir='assets/black/', # Pure-black Gemini images
white_dir='assets/white/' # Pure-white Gemini images
)
codebook.save('artifacts/spectral_codebook.npz')
```
### 4. Run V3 Bypass
```python
from synthid_bypass import SynthIDBypass, SpectralCodebook
codebook = SpectralCodebook()
codebook.load('artifacts/spectral_codebook.npz')
bypass = SynthIDBypass()
result = bypass.bypass_v3(image_rgb, codebook, strength='aggressive')
print(f"PSNR: {result.psnr:.1f} dB") # ~40 dB
```
**Strength levels:** `gentle` (minimal change, ~43 dB) → `moderate` → `aggressive` → `maximum` (strongest removal, ~33 dB)
---
## 📁 Project Structure
## Project Structure
```
reverse-SynthID/
├── src/
│ ├── extraction/
│ │ ├── synthid_bypass.py # V1/V2/V3 bypass implementations + SpectralCodebook
│ │ ├── robust_extractor.py # Multi-scale watermark detection (90% accuracy)
│ │ ├── watermark_remover.py # Frequency-domain watermark removal
│ │ ├── benchmark_extraction.py # Performance benchmarking suite
│ │ └── synthid_codebook_extractor.py # Original codebook extractor (legacy)
│ │ ├── synthid_bypass.py # V1/V2/V3 bypass + multi-res SpectralCodebook
│ │ ├── robust_extractor.py # Multi-scale watermark detection
│ │ ├── watermark_remover.py # Frequency-domain watermark removal
│ │ ├── benchmark_extraction.py # Benchmarking suite
│ │ └── synthid_codebook_extractor.py # Legacy codebook extractor
│ └── analysis/
│ ├── deep_synthid_analysis.py # FFT/phase analysis scripts
│ └── synthid_codebook_finder.py # Carrier frequency discovery
│ ├── deep_synthid_analysis.py # FFT / phase analysis scripts
│ └── synthid_codebook_finder.py # Carrier frequency discovery
│
├── assets/
│ ├── synthid_black.jpg # Watermark on black (enhanced)
│ ├── synthid_white.jpg # Watermark on white (enhanced)
│ ├── black/ # Reference black images from Gemini
│ └── white/ # Reference white images from Gemini
├── gemini_black/ # 100 pure-black Gemini images (1024x1024)
├── gemini_white/ # 100 pure-white Gemini images (1024x1024)
├── gemini_random/ # 88 watermarked content images (1536x2816)
│
├── artifacts/
│ ├── codebook/ # Detection codebooks (.pkl)
│ ├── spectral_codebook.npz # V3 spectral fingerprint (119 MB)
│ ├── v3_output/ # V3 bypass output samples
│ └── visualizations/ # FFT, phase, carrier visualizations
│ ├── spectral_codebook_v3.npz # Multi-res V3 codebook [1024x1024, 1536x2816]
│ ├── codebook/ # Detection codebooks (.pkl)
│ └── visualizations/ # FFT, phase, carrier visualizations
│
├── watermark_investigation/ # Early-stage Nano-150k analysis (archived)
├── SYNTHID_CODEBOOK_ANALYSIS.md # Detailed codebook reverse-engineering report
├── synthid.pdf # SynthID paper reference
├── assets/ # README images and early analysis artifacts
├── watermark_investigation/ # Early-stage Nano-150k analysis (archived)
└── requirements.txt
```
---
## 🔬 Technical Deep Dive
## Technical Deep Dive
### How SynthID Works (Reverse-Engineered)
@@ -260,10 +251,10 @@ reverse-SynthID/
┌──────────────────────────────────────────────────────────────┐
│ SynthID Encoder (in Gemini) │
├──────────────────────────────────────────────────────────────┤
│ 1. Generate carrier frequencies: {(14,14), (126,14), ...} │
│ 1. Select resolution-dependent carrier frequencies │
│ 2. Assign fixed phase values to each carrier │
│ 3. Neural encoder adds learned noise pattern to image │
│ 4. Watermark is imperceptible - spread across spectrum │
│ 4. Watermark is imperceptible — spread across spectrum │
├──────────────────────────────────────────────────────────────┤
│ SynthID Decoder (in Google) │
├──────────────────────────────────────────────────────────────┤
@@ -273,50 +264,60 @@ reverse-SynthID/
└──────────────────────────────────────────────────────────────┘
```
### SpectralCodebook Extraction
### Multi-Resolution SpectralCodebook
The codebook captures the watermark's full frequency fingerprint:
The codebook captures watermark profiles at each available resolution:
- **50 reference images** (25 pure black + 25 pure white, all from Gemini)
- Extracts **magnitude envelope** and **phase template** per channel
- Computes **phase consistency score** per frequency bin
- Content-adaptive profiles for dark vs. light image regions
- **1024x1024 profile**: from 100 black + 100 white pure-color Gemini outputs
- Black images: watermark is nearly the entire pixel content
- White images (inverted): confirms carriers via cross-validation
- Black/white agreement (|cos(phase_diff)|) filters out generation bias
- **1536x2816 profile**: from 88 diverse watermarked content images
- Content averages out across images; fixed watermark survives in phase coherence
- Watermark magnitude estimated as `avg_mag x coherence^2`
### Selective Notch Filter
### V3 Subtraction Strategy
The V3 bypass doesn't subtract blindly - it targets only bins where:
1. **Magnitude** exceeds the 97th percentile (strong watermark energy)
2. **Phase consistency** ≥ 0.95 across reference images (confirmed watermark, not noise)
3. **Subtraction** is capped at 30% of the image's energy at each bin
The bypass uses **direct known-signal subtraction** (not a Wiener filter):
This surgical precision is why V3 achieves 40+ dB PSNR while still reducing watermark energy.
### Noise Correlation Signature
| Metric | Value | Significance |
|:-------|:-----:|:-------------|
| Mean pairwise noise correlation | **0.218** | Identical watermark in all images |
| Noise structure ratio | **1.32** | Neural encoder byproduct |
| Phase coherence (top carriers) | **>99.9%** | Fixed model-level key |
| Green channel phase std | **<0.007 rad** | Strongest consistency channel |
### Bit Plane Analysis
| Bit Plane | Consistency | Role |
|:---------:|:-----------:|:-----|
| Bit 0 (LSB) | 0.049 | Watermark signal |
| Bit 1 | 0.074 | Watermark signal |
| Bit 2 | 0.125 | Partially watermarked |
| Bit 3 | 0.513 | Mixed |
| Bits 4-7 | 0.635–1.000 | Image structure |
1. **Confidence** = phase_consistency x cross_validation_agreement
2. **DC exclusion** — soft ramp suppresses low-frequency generation biases
3. **Per-bin subtraction** = wm_magnitude x confidence x removal_fraction x channel_weight
4. **Safety cap** — subtraction never exceeds 90-95% of the image's energy at any bin
5. **Multi-pass** — decreasing-strength schedule (aggressive → moderate → gentle) catches residual energy
---
## 🛠️ Core Modules
## Core Modules
### `robust_extractor.py` - Detection
### `synthid_bypass.py`
Multi-scale, multi-denoiser watermark detector achieving 90% detection rate.
**SpectralCodebook** — multi-resolution watermark fingerprint:
```python
codebook = SpectralCodebook()
codebook.extract_from_references('gemini_black', 'gemini_white') # adds 1024x1024 profile
codebook.build_from_watermarked('gemini_random') # adds 1536x2816 profile
codebook.save('codebook.npz')
# Later:
codebook.load('codebook.npz')
profile, res, exact = codebook.get_profile(1536, 2816) # auto-select
```
**SynthIDBypass** — three bypass generations:
```python
bypass = SynthIDBypass()
result = bypass.bypass_simple(image, jpeg_quality=50) # V1
result = bypass.bypass_v2(image, strength='aggressive') # V2
result = bypass.bypass_v3(image, codebook, strength='aggressive') # V3 (best)
```
### `robust_extractor.py`
Multi-scale watermark detector (90% accuracy):
```python
from robust_extractor import RobustSynthIDExtractor
@@ -324,64 +325,24 @@ from robust_extractor import RobustSynthIDExtractor
extractor = RobustSynthIDExtractor()
extractor.load_codebook('artifacts/codebook/robust_codebook.pkl')
result = extractor.detect_array(image)
print(f"Watermarked: {result.is_watermarked}")
print(f"Confidence: {result.confidence:.4f}")
print(f"Phase Match: {result.phase_match:.4f}")
```
**Features:**
- Multi-scale analysis (256, 512, 1024px)
- Wavelet + bilateral + NLM denoising fusion
- ICA-based watermark/content separation
- Ensemble carrier detection across scales
### `synthid_bypass.py` - Bypass (V1/V2/V3)
Three generations of watermark bypass:
```python
from synthid_bypass import SynthIDBypass, SpectralCodebook
bypass = SynthIDBypass()
# V1: Simple JPEG compression
result = bypass.bypass_simple(image, jpeg_quality=50)
# V2: Multi-stage transform pipeline
result = bypass.bypass_v2(image, strength='moderate')
# V3: Spectral codebook subtraction (best)
codebook = SpectralCodebook()
codebook.load('artifacts/spectral_codebook.npz')
result = bypass.bypass_v3(image, codebook, strength='aggressive')
```
### `watermark_remover.py` - Removal
Quality-preserving frequency-domain removal:
```python
from watermark_remover import WatermarkRemover
remover = WatermarkRemover(extractor)
result = remover.remove(image, mode='balanced')
print(f"Watermarked: {result.is_watermarked}, Confidence: {result.confidence:.4f}")
```
---
## 📚 References
## References
- [SynthID: Identifying AI-generated images](https://deepmind.google/technologies/synthid/)
- [SynthID Paper (arXiv:2510.09263)](https://arxiv.org/abs/2510.09263)
---
## ⚠️ Disclaimer
## Disclaimer
This project is for **research and educational purposes only**. SynthID is proprietary technology owned by Google DeepMind. These tools are intended for:
- 🎓 Academic research on watermarking robustness
- 🔒 Security analysis of AI-generated content identification
- 📡 Understanding spread-spectrum encoding methods
- Academic research on watermarking robustness
- Security analysis of AI-generated content identification
- Understanding spread-spectrum encoding methods
**Do not use these tools to misrepresent AI-generated content as human-created.**