AI Content Detection ✦ Images · Video · Audio

Was This AI Generated?
Now You Can Know.

Detection for the age of synthetic media.

FakeMind examines every image, video, and audio file through three independent detection layers — provenance credentials, pixel-level AI detection, and invisible watermark checks — and returns an explainable verdict in seconds.

Request a Demo → See How It Works

Image · Video · Audio EU AI Act Art. 50 Explainable verdicts Seconds per file

FakeMind Scanner — Image · Video · Audio

Provenance

AI Detection

Watermark

The Problem

Seeing Is No Longer Believing.

Generative AI can now produce a photorealistic face, a convincing selfie, a full video clip, or a cloned voice in seconds — of a person who never existed and an event that never happened. Every process that trusts a photo, a video, or a recording is now exposed.

🪪

Synthetic ID Portraits

Identity documents built around a generated face. The portrait matches no real person, defeats reverse-image search, and passes casual visual inspection — because there is no original to compare against.

Fuels synthetic identity fraud

🎭

Deepfake Selfies & Video

Generated "selfie with ID" submissions and injected camera feeds that defeat liveness checks. Remote onboarding, video verification calls, and face-match steps all inherit the same blind spot.

Targets KYC & remote onboarding

📸

Fabricated Proof Photos

Generated evidence: damage photos for insurance claims, proof-of-residence snapshots, delivery confirmations, receipts. Photorealistic, metadata-clean, and produced faster than any review team can keep up with.

Hits claims, disputes & moderation

How It Works

Three Layers. One Verdict.

No single detection technique survives contact with every fake. FakeMind runs three independent layers on every image, video, and audio file — each authoritative in a different situation — and combines them into one explainable verdict.

LAYER 01

PROVENANCE

Provenance Credentials

An open standard adopted by major AI generators embeds a cryptographically signed manifest into the file at the moment of creation — declaring what made it, when, and every edit since. FakeMind verifies the signature chain. When a credential is present, the answer is authoritative, not statistical.

Content Credential verification Signature & trust-chain validation Generation & edit history Stripped-metadata detection

LAYER 02

AI DETECTION

Pixel-Level AI Detection

Credentials get stripped by screenshots, re-uploads, and chat-app forwards — so this layer trusts only the content itself. Vision AI trained on real-versus-generated media reads the statistical fingerprints synthesis leaves behind: frequency-domain signatures, texture patterns, and upsampling artifacts. For audio, the same approach reads the spectral fingerprints of synthetic speech. It scores every file, even with all metadata gone.

Vision AI classification Frequency-domain analysis Noise & texture statistics Upsampling artifact detection

LAYER 03

WATERMARK

Invisible Watermark Check

Leading generators embed an imperceptible watermark directly in the pixel data — a signal designed to survive screenshots, crops, compression, and re-saves. FakeMind scans for these markers to catch generated media whose visible metadata was deliberately laundered away.

Pixel-domain watermark scan Survives crops & screenshots Robust to recompression Authoritative when detected

Aggregated Verdict

Provenance and watermark hits are authoritative when they fire and silent otherwise; AI detection scores every file. The aggregation engine weighs all three into a single verdict with a confidence score and a per-layer breakdown — so a reviewer can always see why, not just what.

● GREEN — Likely Authentic ● YELLOW — Uncertain, Review ● RED — Likely AI-Generated

Beyond Stills

Video Lies at 30 Frames a Second.

FakeMind extends all three detection layers to video: frames are sampled and scanned individually, then temporal checks look for the inconsistencies that generators leave between frames — flicker in fine detail, unstable identity, physics that almost works.

Frame sampling · Per-frame detection · Temporal consistency

📹

Deepfake Video Calls

Injected camera feeds and face-swapped live video used to impersonate customers and executives in remote verification and approval flows.

🎬

Generated Clips

Fully synthetic video fabricated from a text prompt — fake incident footage, fake testimonials, fake "evidence" that never had a camera behind it.

🔁

Face Swaps & Re-enactment

A real recording with a replaced face or puppeteered expressions. The scene is genuine; the person in it is not. Temporal analysis catches the seams.

Audio

Voices Are the Easiest Thing to Fake.

A few seconds of recorded speech is enough to clone a voice. FakeMind applies the same three layers to audio — provenance credentials, spectral analysis of the waveform, and watermark checks — to flag synthetic speech before anyone acts on it.

Spectral analysis · Voice-clone signatures · Watermark scan

📞

Cloned Voice Calls

A cloned executive voice authorizing a transfer, a cloned customer passing a call-center voice check. Real-time voice fraud is now a commodity attack.

🎙️

Fake Voice Messages

Voice notes and voicemails submitted as instructions, approvals, or evidence — a convincing recording of words that were never spoken.

🎧

Spliced & Synthetic Narration

Generated speech stitched into otherwise-real recordings: sentences added, words replaced, consent fabricated. Spectral seams give it away.

In Action

See What FakeMind Catches

A "selfie with ID" submitted to a digital onboarding flow. The document scan is genuine — but the portrait belongs to no one. Here's what FakeMind finds in seconds.

◆ Digital Onboarding

Identity Verification — Selfie With ID

⚠ SYNTHETIC PORTRAIT

Selfie Portrait SYNTHETIC SIGNATURE

⚠ AI-GENERATED

Provenance Credential METADATA STRIPPED

? NO CREDENTIAL

ID Document Scan NO SYNTHESIS SIGNAL

✓ CLEAN

File Integrity STRUCTURE VALID

✓ CONSISTENT

Layer 1 — Provenance Credentials NO SIGNAL

No Content Credential present. Metadata shows signs of deliberate stripping — a common laundering step before submission. Absence alone proves nothing; the pixel layers take over.

Layer 2 — AI Detection SYNTHETIC · 96%

Frequency-domain and texture analysis of the portrait region shows statistical signatures consistent with a generated face — no camera sensor pattern, synthetic upsampling artifacts present.

Layer 3 — Watermark Check DETECTED

An imperceptible generator watermark was detected in the pixel data of the portrait — a signal that survives the screenshot-and-recompress laundering that removed the metadata.

Final Verdict

● RED — AI-Generated · 96% Confidence

Flagged with per-layer evidence and logged to the audit trail

Capabilities

Built to Answer One Question. Reliably.

Everything in FakeMind serves the verdict: is this media authentic or generated? Fast enough for live workflows, explainable enough for auditors and courts.

🧅

Three-Layer Detection

Provenance credentials, pixel-level AI detection, and invisible watermark checks — independent layers with independent failure modes, combined into one verdict.

🎞️

Images, Video & Audio

One pipeline for all three. Stills are scanned directly, video is frame-sampled with temporal checks, and audio is analyzed spectrally for synthetic speech.

🔎

Explainable Verdicts

Every verdict ships with a per-layer breakdown and a confidence score. Reviewers see which layer fired and why — never just a black-box number.

⚖️

Compliance Reporting

EU AI Act Article 50 makes "was this generated?" an audit question. FakeMind produces the machine-verified answer and the disclosure line item to match.

🔗

API & Batch Processing

A REST API drops FakeMind into onboarding flows, claims systems, and moderation pipelines. Batch endpoints handle high-volume screening.

📝

Immutable Audit Trail

Every upload, verdict, and human decision is logged with evidence. Investigators get a defensible record, not a screenshot of a score.

👥

Human Review Workflow

Uncertain verdicts route to reviewers with the full per-layer evidence. Overrides require a reason and land in the audit trail.

🛡️

Private by Design

Your media is never used to train models and never shared. Deployment options are designed around data sovereignty from day one.

⚡

Seconds, Not Minutes

Verdicts in seconds per image and near-real-time for sampled video — fast enough to sit inline in a live onboarding or claims flow.

Regulatory Alignment

"Was This Generated?" Is Now a Legal Question.

Since August 2026, Article 50 of the EU AI Act requires AI-generated content to be disclosed and machine-readably marked. Organizations that accept photos, video, and voice recordings from the public now need a documented answer — for every file.

EU AI Act — Article 50

Synthetic Content Disclosure

Providers must mark AI-generated content machine-readably; deployers must disclose it. Any regulated workflow that ingests public media inherits the burden of checking.

✓ FakeMind reads the marks and produces the verdict

Evidence

Audit-Ready Logging

A verdict without a record is worthless in an audit. Every check FakeMind runs is logged with its per-layer evidence, timestamp, and confidence — retained for inspection.

✓ Immutable trail with per-file evidence

Beyond the EU

A Global Standard Question

Platform integrity rules, insurance regulators, courts weighing photographic evidence — the same question is arriving everywhere. A documented detection step is becoming baseline diligence.

✓ One pipeline, every jurisdiction's paper trail

Request a Demo

See FakeMind in action. In a 30-minute call, we'll scan real and generated media live, walk through the three-layer verdict, and map FakeMind onto your intake flow — onboarding, claims, moderation, or newsroom verification.

This call is for:

Fraud and risk teams facing synthetic identities
KYC and digital onboarding providers
Compliance officers preparing for EU AI Act audits
Claims, trust & safety, and verification teams

⏱ 30-minute call • No commitment required • Live detection demo

✓

Request Received

Our team will contact you within 24 hours.

About Us

Built by the Team Behind DocMind

FakeMind is a product of boxMind.ai, the AI innovation arm of AEG — Allied Engineering Group, a trusted name in financial technology and security infrastructure since 1994. It grew out of DocMind, our document fraud detection platform for banks — where customers kept asking one question: was this generated?

boxMind.ai

AI Innovation Lab

boxMind.ai is the AI research and product division behind FakeMind and DocMind. We specialize in purpose-driven AI systems that solve real-world problems for financial institutions and enterprises — from document fraud detection to synthetic media verification — all designed to respect data sovereignty.

Visit boxMind.ai →

AEG — Allied Engineering Group

30+ Years in Financial Technology

Founded in 1994, AEG is one of the top SWIFT Service Bureaus worldwide — fully accredited and compliant with SWIFT's Standard Operation Practice. For three decades, AEG has delivered enterprise-grade connectivity, compliance, and security solutions to banks and financial institutions globally.

SWIFT IBM Cisco Kaspersky Rapid7 Bottomline SafeNet Fircosoft

Visit AEG →

Was This AI Generated?Now You Can Know.