How It Works

From pixel to proof —
the complete forensic pipeline.

Sniffer runs a seven-layer forensic analysis on every image submitted. Each layer operates independently. The results are weighted, scored, and assembled into a cryptographically signed evidence report — formatted for platform takedown filings, legal exhibits, and personal records.

7

Analysis layers

<60s

Report time

0

Account required

Start Investigation →

Verification Process

Four steps from submission to signed report.

01Intake

Create a verification case

Begin by providing basic context — where you found the image, the issue type, and your privacy preference. No account is required. Anonymous submissions are fully supported.

Case ID generated · no identity required

What happens

  • Choose your privacy mode: named or anonymous
  • Select the platform where the image was found
  • Describe the suspected harm or manipulation
  • A unique Case ID is generated for your session
02Upload

Upload the suspicious image

Your image is SHA-256 hashed at the boundary before anything else runs. The hash becomes the immutable identifier for the entire forensic chain. A reference image may optionally be provided for comparative analysis.

SHA-256 · hash-first architecture · file discarded post-analysis

What happens

  • JPEG, PNG, WEBP supported — up to 20 MB
  • File is hashed immediately on receipt
  • Optional: provide a reference image for comparison
  • Thumbnail cached in your session only — never retained server-side
03Analysis

Seven-layer forensic pipeline runs

All analysis layers execute concurrently. Each layer operates independently — no single layer controls the verdict. Results are weighted and aggregated into a composite authenticity score.

30–60s · 7 concurrent analysis modules

What happens

  • All layers run in parallel, typically within 30–60 seconds
  • Each layer produces a confidence value and binary flag
  • Weighted scoring algorithm combines all signals
  • C2PA provenance manifest parsed if present
04Report

Signed forensic report generated

A structured evidence document is produced containing per-layer scores, detected anomaly regions, a forensic narrative, and a cryptographically derived case hash. The report is formatted to meet platform takedown filing and legal exhibit standards.

Tamper-evident hash · printable PDF · platform-ready

What happens

  • Per-layer confidence scores with status flags
  • Annotated tamper heatmap and ELA overlay
  • Natural-language forensic explanation
  • SHA-256 report hash for chain-of-custody integrity

Analysis Engine

Seven independent forensic layers. One composite verdict.

Each layer is designed to catch a different class of manipulation. A determined forgery rarely defeats all seven simultaneously.

ELAError Level01/07

Error-Level Analysis

Re-saves the image at a known JPEG quality and measures per-pixel deviation from the original. Edited regions compress differently — their ELA residuals stand out as bright regions against a uniform noise floor.

TechJPEG re-compression differential · 95% quality baseline
FlagsIdentifies splicing, cloning, and object insertion
EXIFMetadata02/07

EXIF Metadata Forensics

Parses all embedded metadata fields including camera model, software IDs, GPS coordinates, creation timestamps, and colour profiles. Missing, forged, or inconsistent tags are primary manipulation indicators.

TechExifTool extraction · 200+ field analysis · software tag matching
FlagsDetects edited timestamps, absent GPS, conflicting software chains
GANAI Detection03/07

GAN Artifact Detection

Analyses the frequency spectrum and spatial domain for statistical patterns unique to neural network output — checkerboard artifacts, spectral peaks at regular intervals, and texture distributions inconsistent with camera optics.

TechFFT spectrum analysis · CNN classifier · spectral fingerprinting
FlagsIdentifies AI-generated faces, backgrounds, and full composites
LMKSBiometric04/07

Facial Landmark Distortion

Maps 68 facial key-points and measures geometric deviation from anatomically valid proportions. Deepfake generators and face-swap tools introduce measurable asymmetries at the bone structure level.

Techdlib 68-point predictor · Procrustes deviation scoring
FlagsFlags unnatural jaw alignment, eye asymmetry, and skin-boundary blur
CRMStructural05/07

Clone Region Mapping

Detects copy-move forgery — where a region of the image is duplicated and pasted elsewhere to cover or add content. Uses block-matching across keypoint descriptors to find near-identical regions.

TechSIFT / ORB keypoint matching · affine transform clustering
FlagsCommon in document fraud, scene manipulation, and object removal
PRNUSensor06/07

Noise Pattern Analysis

Every camera sensor has a unique photon-response non-uniformity (PRNU) noise fingerprint. Composited image regions break this field — the foreign pixels carry a different sensor signature.

TechWavelet denoising · PRNU extraction · cross-correlation mapping
FlagsDetects multi-source compositing even after JPEG recompression
DCTCompression07/07

DCT Compression Fingerprinting

JPEG images store data as discrete cosine transform (DCT) coefficients. Re-saved or edited regions have double-quantised coefficients — a statistical ghost of the editing tool's compression pass.

TechDCT histogram analysis · quantisation grid detection · double-JPEG identification
FlagsReveals local re-saves, region upscaling, and steganographic modification

Evidence Output

What the forensic report contains.

Structured for platform abuse teams, legal professionals, and survivors who need verifiable documentation.

Case Metadata

  • Case ID and platform source
  • Submission timestamp · UTC
  • Issue type and privacy mode
  • SHA-256 file hash

Forensic Verdict

  • Authenticity score (0–100)
  • High-confidence verdict label
  • Certainty level declaration
  • Flagged layer count

Layer Results

  • Per-layer confidence score
  • Binary flag (clean / anomalous)
  • Weighted signal contribution
  • Algorithm version reference

Visual Evidence

  • ELA residual heatmap
  • Tamper region bounding boxes
  • Keypoint distortion overlay
  • Comparative reference panel

C2PA Provenance

  • Manifest issuer and signer
  • Generator tool declaration
  • Cryptographic signature status
  • AI origin assertion label

Audit Trail

  • Pipeline version and hash
  • Processing timestamps
  • Module execution log
  • Report integrity hash

Chain of Custody

The report hash is derived from the forensic output — not your image.

SHA-256 is computed over the analysis results JSON. Any alteration to the report after generation changes the hash, which invalidates the document as evidence. This design means your original file is mathematically excluded from the evidence chain entirely.

hash = SHA256(analysis_results)

report_id = SNF‑{case_id}

image_ref = SHA256(file_bytes)

stored = hash only · no pixels

─────────────────────

integrity = verifiable · tamper-evident

Protection Registry

Register your originals before they're misused.

Upload original images to compute and store their cryptographic fingerprint. If the same image is ever submitted through the verification pipeline, it is automatically flagged as matching a registered original — creating an immutable paper trail of ownership.

01

Upload your original

We compute a SHA-256 hash and perceptual image fingerprint. The image itself is never retained — only the mathematical proof.

02

Fingerprint registered

Your hash is written to the protection registry. Any identical or near-identical image uploaded for verification will automatically match against your entry.

03

Match detected

When a match is found, the forensic report flags the image with your registry reference ID — giving you cryptographic proof of prior registration.

Registered6 Mar 2026 · 14:32 UTC

Reference ID

A3F8D2C0

SHA-256 Hash

a3f8d2c0…1b9e74f3

Perceptual Hash

f7e3c2b1…9a0d5e2f

Registry Match

This image was submitted for verification on 6 Mar 2026 and automatically matched your registration.

Registered pre-dispute · ownership chain established

Takedown Workflow

From report to removal — in one step.

Platform abuse forms require evidence in specific formats. Sniffer generates compliant takedown notices for all major platforms automatically.

01

Upload infringing image

Provide the image you found being misused online. We compute its hash and check the protection registry.

02

Registry check

If a registry match is found, your ownership is cryptographically confirmed and referenced in the notice.

03

Notice generated

A formal DMCA-style takedown notice is generated, pre-populated with your evidence hash and case reference.

04

Submit to platform

The notice links directly to the platform's official abuse reporting form — formatted to their requirements.

Privacy & Security

Designed so you never have to trust us with more than you intended.

Image never stored

Your file is hashed on arrival and discarded immediately after analysis. The report references the hash — not the file.

Anonymous by default

No email, account, or identity is required. Anonymous cases receive full forensic analysis — indistinguishable from attributed ones.

Tamper-evident report hash

The forensic output is itself hashed — any modification to the report after generation changes the hash, invalidating it as evidence.

No personal data retained

Session thumbnails are stored in your browser only. No IP address, device identifier, or browsing pattern is logged.

Start Now

Ready to run your first forensic analysis?

No account. No install. Upload your image and receive a signed evidence report in under 60 seconds.