Files
reverse-SynthID/README.md
T
Alosh Denny cc00e57582 docs: rewrite README for V4 Round-06 release
- Add 'What the Watermark Looks Like' section with synthid_white.jpg
- Add 'Round 06 — It Works' section with fidelity comparison image and
  a table of all six attack rounds
- Document the 7-stage all-in-one pipeline and elastic deformation rationale
- Add Round-06 preset table (final vs nuke) and updated pipeline diagram
- Update architecture table and results summary to reflect confirmed bypass
- Remove all CSV file references

Made-with: Cursor
2026-04-24 02:09:09 +05:30

25 KiB
Raw Blame History

SynthID Watermark Analysis

Reverse-Engineering SynthID

Discovering, detecting, and surgically removing Google's AI watermark through spectral analysis

Visit us on PitchHut

Python License Detection V3 Bypass V4 Bypass Models Attack


What the Watermark Looks Like

SynthID encodes an imperceptible pattern directly into pixel values. On a pure white image generated by Gemini, the watermark is almost the entire signal. Amplify the high-frequency residual and it looks like this:

SynthID watermark pattern visible on a white Gemini-generated image

Amplified SynthID carrier pattern extracted from a pure-white Gemini image. The diagonal banding is the watermark's spatial frequency signature — the target of our spectral attack.


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 resolution-dependent carrier frequency structure
  2. Built a detector that identifies SynthID watermarks with 90% accuracy
  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
  4. Generalized to multi-model, multi-color consensus (V4) — per-model profiles for gemini-3.1-flash-image-preview and nano-banana-pro-preview, cross-color phase consensus over 6 solid backgrounds, and a human-in-the-loop calibration loop that tunes per-carrier subtraction strength from manual Gemini-app detection tallies
  5. Broke the detector across both models (Round 06) with a unified 7-stage all-in-one attack targeting every documented SynthID failure mode simultaneously

VT-OxFF built a really cool visualizer to view the process of how SynthID watermark is added to images here (also available in repo description)!


Round 06 — It Works ✓

After six iterative rounds of adversarial development, Round 06's bypass_v4_final / bypass_v4_nuke pipeline defeats the Gemini SynthID detector on both gemini-3.1-flash-image-preview and nano-banana-pro-preview images, with visually lossless output.

Round 01 vs Round 06 — Fidelity Comparison

Side-by-side comparison of Round 01 (gentle spectral only) vs Round 06 (all-in-one) on the same source image

Left: Round 01 output (gentle spectral subtraction only). Right: Round 06 output (final — VAE + elastic warp + squeeze + color + JPEG). Both look identical to human eyes; only Round 06 defeats the SynthID detector.

What Changed Between Rounds

Round Strategy Outcome
01 Conservative spectral subtraction (gentle)
02 Aggressive spectral subtraction + JPEG
03 Blog-guided absolute bin targeting
04 Denoise-residual phase extraction
05 Diffusion-VAE re-generation + geometric warp
06 All-in-one: VAE + elastic fragmentation + squeeze + color + JPEG

The breakthrough in Round 06 came from treating the Gemini app's own published failure-mode list as an attack specification:

"When an AI-generated image is part of a complex collage, layered behind other elements, or has many different textures and patterns placed over it, the detector may struggle to isolate the specific signature from the overall file." — Gemini app, SynthID detection help text

The elastic deformation stage simulates this effect at the pixel level: a smooth, low-frequency random warp field gives every ~50-pixel neighbourhood its own independent sub-pixel offset, fragmenting the watermark's spatial phase consensus without introducing any visible distortion.


V4 — Cross-Color Consensus + Human-in-the-Loop Calibration

V4 is a ground-up re-think of the codebook built on a much richer dataset:

  • Multi-model: separate profiles for gemini-3.1-flash-image-preview and nano-banana-pro-preview (plus an optional union pseudo-model).
  • Multi-color: 6 consensus colors (black, white, blue, green, red, gray) per model per resolution, plus gradient and diverse as content baselines.
  • Cross-color phase consensus: the primary carrier mask. A true SynthID carrier is image-content-independent, so its phase is consistent across every solid-color background. Content-driven energy phase-scrambles across colors and drops out of the consensus.
  • Fidelity-preserving dissolver: PSNR-floor rollback, luminance-safe DC, per-bin subtraction cap.
  • Human-in-the-loop calibration loop: a codebook field carrier_weights is updated based on manual Gemini-app detection feedback.

Consensus coherence (why V4 wins)

For each frequency bin (fy, fx) and channel ch:

consensus(fy, fx, ch) = | mean_over_colors( exp(i * phase_color(fy, fx, ch)) ) |

Values near 1.0 mean the phase at that bin is locked across every solid-color background, which is only true for the watermark. Content bins collapse to < 0.3 because their phase is randomized by different color tints. On the V4 codebook built from the enriched dataset, 99%+ of content bins fall below the default tau=0.60 cutoff, so the V4 dissolver never touches them — this is what buys back PSNR.

Two-phase release workflow

flowchart LR
    dataset[reverse-synthid-dataset<br/>model x color x resolution] --> build[scripts/build_codebook_v4.py]
    build --> codebook[artifacts/spectral_codebook_v4.npz]
    codebook --> dissolve[scripts/dissolve_batch.py]
    input[watermarked inputs] --> dissolve
    dissolve --> variants[final / nuke variants]
    variants --> gemini[Gemini app<br/>manual SynthID detection]
    gemini --> feedback[detection feedback]
    feedback --> calibrate[scripts/calibrate_from_feedback.py]
    calibrate -->|updates carrier_weights| codebook

V4 Quickstart

# 1. Build the codebook from the enriched hierarchical dataset
python scripts/build_codebook_v4.py \
    --root /path/to/reverse-synthid-dataset \
    --output artifacts/spectral_codebook_v4.npz

# 2. Run the Round-06 all-in-one attack on a batch (recommended)
python scripts/dissolve_batch.py \
    --input  ./to_clean/ \
    --output ./runs/round_06/ \
    --codebook artifacts/spectral_codebook_v4.npz \
    --model gemini-3.1-flash-image-preview \
    --strengths final nuke

# 3. Upload each output image to the Gemini app and run SynthID detection.
#    Use the results to feed back into the calibration script if needed.

Round-06 Attack Presets

Two presets are available via --strengths:

Preset VAE passes Elastic α Squeeze JPEG chain PSNR floor
final 1 1.8 px 90 % q=92→88 14 dB
nuke 2 2.8 px 82 % q=88→84→90 11 dB

Both presets stack the same 7-stage pipeline:

  1. VAE round-trip (Stable Diffusion sd-vae-ft-mse) — projects image off the natural-image manifold the SynthID decoder was never trained against (Gowal et al. 2026, §6.1)
  2. Elastic deformation — smooth low-frequency random warp field, simulates the "collage fragmentation" failure mode Gemini itself acknowledges
  3. Global geometric combo — small rotation + zoom + pixel shift in one affine warp
  4. Resize-squeeze — downsample (AREA) → upsample (LANCZOS), erases sub-pixel watermark info
  5. Color-contrast nudge — brightness / contrast / saturation / hue micro-shift
  6. Residual-phase FFT subtraction — blog-universal + codebook-harvested carrier bins, cap-limited
  7. JPEG chain + luma noise + bilateral — heavy compression / re-encoding disruption

Every stage is independently PSNR-gated; any stage that would drop quality below the floor is rolled back automatically.

V4 Codebook Structure

Profiles keyed by (model, H, W). Each profile stores:

Field Shape Notes
consensus_coherence (H, W, 3) Primary carrier mask (cross-color phase consensus).
consensus_phase (H, W, 3) Mean unit-phase angle across colors. Subtraction template.
inverted_agreement (H, W, 3) Pairwise abs(cos(phase_diff)), weighted for black<->white.
avg_wm_magnitude (H, W, 3) Mean magnitude across consensus colors.
content_baseline (H, W, 3) From diverse/ + gradient/ — used for luminance blending.
carrier_weights (H, W, 3) Live. Starts at consensus^2 * (0.5 + 0.5 * agreement). Updated by the calibration loop.
n_refs_per_color {color: int} Per-color ref counts.

Save format reuses the v3 compact rfft + float16/uint8 encoding; a 14-profile codebook across 2 models × 7 resolutions is ~220 MB on disk.

V4 Detector (Sanity Check)

Before spending time on manual Gemini validation, sanity-check bypass outputs against the V4 codebook's own consensus:

from robust_extractor import RobustSynthIDExtractor
from synthid_bypass_v4 import SpectralCodebookV4

cb = SpectralCodebookV4()
cb.load('artifacts/spectral_codebook_v4.npz')

ext = RobustSynthIDExtractor()
result = ext.detect_from_v4_codebook(image_rgb, cb,
                                     model='nano-banana-pro-preview')
print(result.is_watermarked, result.confidence, result.phase_match)

On the 1024x1024 exact-match path we see conf=0.91, phase_match=0.65 for watermarked and conf=0.02, phase_match=0.31 after aggressive V4 dissolve.

V4 vs V3

V3 V4
Reference colors black + white black, white, blue, green, red, gray (+ diverse/gradient content baselines)
Cross-validation abs(cos(phase_black - phase_white)) cross-color consensus over 6 colors + pairwise agreement
Models single-model (Gemini 2.5) per-model profiles (gemini-3.1-flash-image-preview, nano-banana-pro-preview) + optional union
Attack spectral subtraction only 7-stage: VAE + elastic + squeeze + color + FFT + JPEG chain
PSNR (aggressive) 43 dB visually lossless (1824 dB pixel-level; warp displaces pixels)
Fidelity guard none per-stage PSNR-floor rollback
Detector bypass local only confirmed ✓ on Gemini app (both models)

V3 remains in the repo (src/extraction/synthid_bypass.py, bypass_v3) unchanged for anyone who depends on it.


🚨 Contributors Wanted: Help Expand the Codebook

We're actively collecting pure black and pure white images generated by Nano Banana Pro to improve multi-resolution watermark extraction.

If you can generate these:

  • Resolution: any (higher variety = better)
  • Content: fully black (#000000) or fully white (#FFFFFF)
  • Source: Nano Banana Pro outputs only

How to Contribute

  1. Generate a batch of black/white images by attaching a pure black/white image into Gemini and prompting it to "recreate this as it is"
  2. Upload them to our Hugging Face dataset: aoxo/reverse-synthid
    • gemini_black_nb_pro/ (for black)
    • gemini_white_nb_pro/ (for white)
  3. Open a Pull Request on the HF dataset repo

These reference images are critical for:

  • Carrier frequency discovery
  • Phase validation
  • Improving cross-resolution robustness

Even 150200 images at a new resolution can significantly improve detection and removal.

Download Reference Images

Reference images are hosted on Hugging Face to keep the git repo lightweight:

pip install huggingface_hub
python scripts/download_images.py           # download all
python scripts/download_images.py gemini_black  # download specific folder

Dataset: huggingface.co/datasets/aoxo/reverse-synthid


Key Findings

The Watermark is Resolution-Dependent

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.

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

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 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

Carrier Frequency Structure

At 1024x1024 (from black/white refs), top carriers lie on a low-frequency grid:

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

Architecture

Bypass Generations

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 Multi-resolution spectral codebook subtraction 43+ dB 91% phase coherence drop Prior best
V4 Round 06 7-stage all-in-one (VAE + elastic + squeeze + color + JPEG) visually lossless detector bypassed ✓ Current best

V3 Pipeline

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

V4 Round-06 Pipeline

Input Image (any resolution)
       │
       ▼  Stage 1: VAE round-trip (SD sd-vae-ft-mse, 1-2 passes)
       │           Projects image off natural-image manifold
       ▼  Stage 2: Elastic deformation (smooth random warp field)
       │           Fragments spatial phase consensus ("collage effect")
       ▼  Stage 3: Global geometric combo (rotation + zoom + shift)
       │           Single affine warp, no compounded aliasing
       ▼  Stage 4: Resize-squeeze (AREA ↓ then LANCZOS ↑)
       │           Erases sub-pixel watermark information
       ▼  Stage 5: Color-contrast nudge (HSV micro-shift)
       │           Shifts per-pixel statistics SynthID keys on
       ▼  Stage 6: Residual-phase FFT subtraction
       │           Blog-universal + codebook-harvested carrier bins
       ▼  Stage 7: JPEG chain + luma noise + bilateral filter
       │
       ▼
  Output (SynthID detector: no watermark detected ✓)

Quick Start

Installation

git clone https://github.com/aloshdenny/reverse-SynthID.git
cd reverse-SynthID

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# For Round-06 VAE stage:
pip install torch diffusers safetensors accelerate
import sys
sys.path.insert(0, 'src/extraction')
from synthid_bypass_v4 import SynthIDBypassV4, SpectralCodebookV4

cb = SpectralCodebookV4()
cb.load('artifacts/spectral_codebook_v4.npz')

b = SynthIDBypassV4()
result = b.bypass_v4_file(
    'input.png', 'output.png',
    cb,
    strength='final',                      # or 'nuke' for maximum strength
    model='gemini-3.1-flash-image-preview',
)
print(result.stages_applied)

Run V3 Bypass

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']}")

From the CLI:

python src/extraction/synthid_bypass.py bypass input.png output.png \
    --codebook artifacts/spectral_codebook_v3.npz \
    --strength aggressive

Detect Watermark

python src/extraction/robust_extractor.py detect image.png \
    --codebook artifacts/codebook/robust_codebook.pkl

Project Structure

reverse-SynthID/
├── src/
│   ├── extraction/
│   │   ├── synthid_bypass.py              # V1/V2/V3 bypass + multi-res SpectralCodebook
│   │   ├── synthid_bypass_v4.py           # V4 cross-color consensus codebook + dissolver
│   │   ├── vae_regen.py                   # Round-06 SD-VAE re-generation stage
│   │   ├── robust_extractor.py            # Multi-scale watermark detection (+ V4 hook)
│   │   ├── 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
│
├── scripts/
│   ├── download_images.py                 # Download reference images from HF
│   ├── build_codebook_v4.py               # V4: build per-(model, HxW) consensus codebook
│   ├── dissolve_batch.py                  # V4: emit strength variants
│   └── calibrate_from_feedback.py         # V4: update carrier_weights from detection feedback
│
├── artifacts/
│   ├── spectral_codebook_v3.npz           # Multi-res V3 codebook [1024x1024, 1536x2816]
│   ├── spectral_codebook_v4.npz           # V4 codebook (per-model, per-resolution)
│   ├── codebook/                          # Detection codebooks (.pkl)
│   └── visualizations/                    # FFT, phase, carrier visualizations
│
├── assets/
│   ├── synthid_watermark.png              # Watermark analysis header image
│   ├── synthid_white.jpg                  # Amplified SynthID pattern on white image
│   ├── v4_round1_vs_round6.png            # Round 01 vs Round 06 fidelity comparison
│   └── ...
│
├── runs/
│   ├── round_01/ … round_05/             # Historical bypass attempts
│   └── round_06/                          # Working bypass (final + nuke presets)
│
├── watermark_investigation/               # Early-stage Nano-150k analysis (archived)
└── requirements.txt

Technical Deep Dive

How SynthID Works (Reverse-Engineered)

┌──────────────────────────────────────────────────────────────┐
│                  SynthID Encoder (in Gemini)                  │
├──────────────────────────────────────────────────────────────┤
│  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      │
├──────────────────────────────────────────────────────────────┤
│                  SynthID Decoder (in Google)                  │
├──────────────────────────────────────────────────────────────┤
│  1. Extract noise residual (wavelet denoising)               │
│  2. FFT → check phase at known carrier frequencies           │
│  3. If phases match expected values → Watermarked            │
└──────────────────────────────────────────────────────────────┘

Why Elastic Deformation Works

SynthID's training augmentation set (Gowal et al. 2026, Table 1) includes SmallRotation, Cropresize, JPEG, GaussianBlur, BrightnessContrast, and Screenshotting — all global, uniform spatial transforms. The elastic warp field is a spatially varying distortion: each local neighbourhood gets its own independent sub-pixel offset. Because the offsets are smooth (Gaussian-blurred from white noise, σ=4456 px), the image content is visually unaffected, but the watermark's phase-consensus structure is incoherent — it can no longer be aggregated across the image. This is the pixel-level equivalent of the "collage fragmentation" effect that Gemini's own app cites as a detector failure mode.


Results Summary

V3 (spectral subtraction, 88 Gemini images)

Metric Value
PSNR 43.5 dB
SSIM 0.997
Carrier energy drop 75.8%
Phase coherence drop (top-5 carriers) 91.4%

V4 Round 06 (all-in-one attack, 20 images validated)

Model Preset Detector bypassed
gemini-3.1-flash-image-preview final
gemini-3.1-flash-image-preview nuke
nano-banana-pro-preview final
nano-banana-pro-preview nuke

References


👤 Maintainer & Contact

Alosh Denny AI Watermarking Research · Signal Processing

📧 Email: aloshdenny@gmail.com 🔗 GitHub: https://github.com/aloshdenny

For collaborations, research discussions, or contributions, feel free to reach out or open an issue/PR.


Support This Research

This project is maintained independently — no lab funding, no corporate backing.
If this work was useful to you or your team, consider supporting continued development:

Buy Me A Coffee

Funds go toward compute costs (GPU hours for new resolution profiles), dataset expansion, and ongoing bypass research.


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

Do not use these tools to misrepresent AI-generated content as human-created.