Seerrveil
24-layer protection, 5 modes, face-safe option, anti-deepfake mode, realtime processing.
Hidden GPS coordinates. Your device serial number. Your face — searchable by PimEyes in seconds. Your location — readable by GeoSpy from sky color alone. Seerrveil stops all of it.
Seerrveil FORTRESS v7 — Feature List
Complete list of features and what was added in the latest updates.
How to test (when you upload a pic)
- 1. Load the extension — Open Chrome → Extensions → Manage extensions → turn Developer mode ON. Load unpacked (folder with manifest.json). Pin Seerrveil to the toolbar.
- 2. Open popup & upload — Click the Seerrveil icon. Drag & drop an image onto “Drop image to protect” (recommended), or browse files. Image should appear in the canvas with Protection Score and editor.
- 3. Run AI scan — Click Run Scan (click Load Models first if needed; wait until “Ready”). Faces outlined in purple; “FACE 1” etc. in AI Detection panel. Enables face layers and “Preserve face.”
- 4. Test each protection mode — Clean → almost unchanged, downloads *_fortress.png. Fingerprint → identical look; only fingerprint changes. Stealth (L3) → slightly stronger; face stays clean. Anti-deepfake → face more altered (breaks face-swap). Maximum → strongest effect.
- 5. Verify download — Each run downloads a PNG. Optional: compare original vs protected in Google Lens/reverse search; stronger modes should reduce or break exact matches.
- 6. Other checks — Threats tab (threat list); Batch tab (drop 2–3 images, Process & Download All); Settings (noise/blur persist); New / Protect another image resets drop zone.
If image doesn’t show: use drag & drop; JPG/PNG/WEBP/TIFF. Run Scan does nothing: Load Models → wait “Ready” → Run Scan again. Protection doesn’t run: ensure image loaded and editor visible, then click FORTRESS Protect & Download once.
Tabs
| Tab | Description |
|---|---|
| Protect | Main workflow: load image, run AI scan, apply protection, download. |
| Threats | View threat assessment (what FORTRESS protects against). |
| Batch | Process up to 20 images at once (drop or browse). |
| Settings | Noise mode, blur strength, other preferences (saved to storage). |
Protection Modes (5 presets)
Which pipeline runs when you click FORTRESS Protect & Download:
| Mode | What it does | Strength |
|---|---|---|
| Clean | Geometry-only fingerprint break + 12% face layers. No DCT/JPEG/spectral. Image looks unchanged. | 12% face, 0% global noise |
| Fingerprint | Only tiny rotate/scale/translate. No face layers, no other changes. 100% invisible. | Geometry only |
| Stealth (L3) | Face + DCT + JPEG + CLIP + spectral + geometry. No visible noise on face (face-safe). Faster step delay. | 58% (L3 backend) |
| Anti-deepfake | Strong face + anti-deepfake grid + GAN latent + wavelet + hash break. Breaks face-swap/deepfake. Face intentionally perturbed. | 88% |
| Maximum | Full 24-layer pipeline + double hash break. Strongest protection; may visibly alter image. | 100% (Noise Level applies) |
Protection Layers (24 total)
Each can be toggled in the Protection Layers panel. Which layers run depends on the selected mode.
| # | Layer | Short description |
|---|---|---|
| L1 | Face Warp | Geometric face distortion |
| L2 | Embed Attack | ArcFace/FaceNet embedding perturbation |
| L3 | Eye Destroy | Eye/nose/mouth chrominance attack |
| L4 | Landmark Shift | Facial landmark coordinate poisoning |
| L5 | FGSM-PGD | Multi-pass gradient attack |
| L6 | DCT | DCT coefficient injection (JPEG-resistant) |
| L7 | CLIP Corrupt | CLIP/GPT-4V/Gemini embedding corruption |
| L8 | ViT Attack | ViT attention / patch disruption |
| L9 | Spectral | Multi-frequency spectral armor |
| L10 | Multi-Scale | 6-scale noise pyramid (1x→32x) |
| L11 | GAN Poison | GAN/VAE latent space poison |
| L12 | Wavelet | Wavelet subband noise |
| L13 | Freq Comb | Frequency comb injection |
| L14 | Color Poison | YCbCr chrominance attack |
| L15 | Anti-Deepfake | CNN grid poison (anti face-swap) |
| L16 | Texture Block | Style transfer / Gram matrix blocker |
| L17 | Depth Corrupt | Depth estimation corruption (MiDaS/DPT/Zoe) |
| L18 | OCR Poison | OCR edge poisoning |
| L19 | NeRF Poison | NeRF/3DGS reconstruction poison |
| L20 | Stego Chaff | LSB steganographic chaff |
| L21 | Hash Break | Perceptual hash (pHash/dHash/aHash) destruction |
| L22 | JPEG Armor | JPEG recompression survival |
| L23 | Face Texture | Subtle face texture injection |
| L24 | Anti-Denoise | Anti-denoising resilience patterns |
Other Protect-tab controls
| Control | Description |
|---|---|
| Preserve face | When on (Clean/Stealth/Fingerprint), noisy layers are not applied on the detected face so the face stays clean. |
| Noise Level | Low / Med / High / Max. Only affects Maximum mode. |
| Run Scan | Loads AI models (BlazeFace + COCO-SSD) and detects faces/objects. Required for face layers and Preserve face. |
| Protection Score | Meter showing estimated protection level. |
| Canvas tools | Blur, Pixelate, Background blur, Cloak (Fawkes-style), Reset, New. Brush size slider. |
| EXIF Metadata | Collapsible panel: shows EXIF status; export strips EXIF. |
| FORTRESS Protect & Download | Runs the pipeline for the selected mode and downloads PNG. |
AI / Models
| Feature | Description |
|---|---|
| BlazeFace | Face detection (TensorFlow.js, loaded in sandbox iframe). |
| COCO-SSD | Object detection for optional landmark/object blur. |
| Load Models | Downloads and loads models; progress bar and status (Not loaded / Ready). |
| Run Scan | Detects faces and objects; draws boxes; enables face layers and Preserve face. |
| AI Detection panel | Lists detected faces/objects; per-face “Protect” button to apply cloak. |
Batch & Settings
Batch tab: Drop or browse up to 20 images. Process & Download All runs protection (current mode) on each and triggers download. Clear list to start over.
Settings tab: Noise mode (same as Noise Level in Protect). Blur strength for brush blur. Settings persisted in chrome.storage.local.
Technical / Backend
- Sandbox — sandbox.html + sandbox.js load TensorFlow and run BlazeFace/COCO-SSD; communication via postMessage.
- Variance map — Built from image; used to modulate noise strength by texture.
- Face mask — Built from detected faces; used for “Preserve face” (zero or reduce noise on face).
- Geometry-only hash break — rotate + scale + translate only (no LSB, no block chrominance). Used in Clean, Fingerprint, Stealth.
- Full hash break — LSB randomization + geometry + block chrominance. Used in Maximum and Anti-deepfake when Hash Break layer is on.
- Step delay — 8 ms between pipeline steps for near realtime feel (was 60 ms).
- EXIF — Parsed for display; export is PNG with EXIF stripped.
What we added (this round)
Upload & display: Image load error handling; scroll into view after load; drop zone hint (“If the window closes when you browse, drag & drop”); Reset/New restores “Drop image to protect”; editor shown with display: block after load.
Reverse search / Google Lens: Stronger hash break (larger rotation, scale, translation, block chrominance); double hash break in Maximum; stronger face layers in Maximum (higher eps/strength for face warp, embed, eye attack, landmark shift, face texture, Fawkes-style).
Presets & modes: Clean (zero visible change; geometry + 12% face); Fingerprint (geometry only); Stealth (L3) 58% strength, CLIP + spectral, face-safe, 8 ms step delay; Anti-deepfake (new); Maximum full 24 layers + double hash break.
Face preservation: Preserve face checkbox; when on (Clean/Stealth/Fingerprint), global noisy layers multiplied by (1 − face_mask) on face; vmaskFaceSafe; StegoChaff/HashBreak skip or reduce on face when Preserve face on.
Performance & UX: Realtime 8 ms step delay; protection preset persisted; button styles (Clean green, Maximum red, Anti-deepfake amber, Stealth tooltip “L3 — powerful, realtime”).
Summary: modes vs layers
| Mode | Face layers | DCT | JPEG | CLIP | Spectral | Other globals | Hash break | Face-safe |
|---|---|---|---|---|---|---|---|---|
| Clean | ✓ 12% | ✗ | ✗ | ✗ | ✗ | — | Geometry only | ✓ |
| Fingerprint | ✗ | ✗ | ✗ | ✗ | ✗ | — | Geometry only | ✓ |
| Stealth | ✓ 58% | ✓ | ✓ | ✓ | ✓ | — | Geometry only | ✓ |
| Anti-deepfake | ✓ 88% | ✗ | ✗ | ✗ | ✗ | Deepfake, GAN, Wavelet | Full (1×) | ✗ |
| Maximum | ✓ 100% | ✓ | ✓ | ✓ | ✓ | All 24 layers | Full (2×) | If checkbox on |
Seerrveil FORTRESS v7 — 24-layer protection, 5 modes, face-safe option, anti-deepfake mode, realtime processing.
100% Local. Zero Data Sent.
All AI models run entirely inside your browser
No servers. No accounts. No analytics. No tracking.
Not even Seerror can see your photos
Who Needs Seerrveil
Part of the Seerror Suite
Seerrveil is part of the Seerror privacy suite.
Research: Publications •
Bulk delete: Seerraze