Skip to content

Image and document authenticity verification

Verify the files your decisions depend on.

Check photos for AI generation and local edits, and documents for file changes, signature issues and content inconsistencies. Review the evidence behind each finding before approving a claim, making a payment or publishing a submission.

Talk to salesCheck an imageCheck a document

Each account includes one free image check and one free document check. Review the result summary, then use a matching credit to unlock detailed evidence and downloads.

Scope

Checks suited to each file. Evidence your team can review.

Image analysis examines pixels, provenance and editing traces. Document analysis brings text, file structure, signatures and scanned fields into one review. Each report connects findings to their evidence and explains the strength of the assessment.

ImagesJPEG · PNG · WebP · up to 10 MB

Check generation, local edits and provenance

  • AI detection beyond metadata

    FVIMG V2 combines whole-image analysis with focused pixel checks to identify AI-generated content, including files without generation metadata.

  • Find AI edits within real photos

    Look for splicing, copy-move, retouching and generative fill. When the evidence supports localization, the report marks suspicious regions to guide comparison with an original or another view.

  • Provenance

    Reads Content Credentials (C2PA), EXIF and XMP metadata, generation parameters and editing-software traces, and checks each claim against the pixels.

  • Labels and watermarks

    Reads machine-readable AI labels, including those defined by China’s GB 45438-2025 for content from Chinese platforms. An image analysis also detects known invisible watermarks that remain after credentials are stripped.

DocumentsPDF · DOCX · DOC · RTF · ODT · PPTX · TXT · up to 100 pages, 50 MB

Review text, scanned fields and file records

  • AI-written text

    Our own model, FVDOC V3, evaluates every paragraph with enough running text, in English, French, German, Spanish or Chinese.

  • Back to the source

    Follow findings back to the source: flagged passages, scanned fields and image regions, with page references and coordinates where available.

  • Revisions, hidden content and attachments

    Examine supported Word and presentation revisions, comments, notes and hidden content, plus PDF attachments, annotations and actions. Bring relevant file records into the review before making a decision.

  • Text, scans and embedded photos

    Evaluate text of at least about 150 words, read scanned pages and analyse up to 15 photos per document. Bring written and visual evidence together in one review. Eligible embedded photographs are analysed with the same FVIMG V2 image detection workflow used for standalone images.

Why Fileverity

What sets Fileverity apart.

Review the signals behind a finding, connect the report to the submitted file and choose a deployment that fits your data requirements.

  • Privacy built into each review

    Uploaded files are deleted after analysis and never used to train models. Temporary private evidence is cleared after delivery or expiry, while summaries and verification records remain available. Download your full report within 3 hours of receipt in the current tab.

  • Proprietary detection that keeps pace

    Fileverity develops FVIMG V2 for image AI detection and FVDOC V3 for document text detection. We evaluate new generative models and update our technology as the content your teams receive changes.

  • Multiple evidence sources in one platform

    Examine image content across major generative model families alongside watermarks, Content Credentials and AI labels. Document analysis brings document text, embedded photos and available file history into the same report.

  • On-premises and fully offline

    Enterprise customers can run Fileverity inside their own infrastructure, offline and without calling any external API. Sensitive and regulated files never leave your servers and are never exposed to the public internet.

  • Built around your workflow

    Enterprise agreements can cover detectors adapted to your file types, risk policies, report formats and system integrations. Contact us to define the scope and evaluate it on your own files.

  • Evidence bound to the received file

    Each report identifies the examined file and report data with SHA-256 fingerprints. Source locations and evidence references help your team review findings and share a traceable record. Validated timestamps and PDF seals add independently checkable proof of when report data existed and whether a sealed PDF has changed.

Analysis layer

Review findings beside their source evidence.

Follow supported findings to marked image regions, document passages or scanned fields. Available page references and coordinates help reviewers compare the evidence with the submitted file.

Image · copy-move

Submitted
Analysis
Demonstration image. We edited this photo ourselves, copying a patch of rust onto another panel (a copy-move manipulation).

Document · AI-written passages

Submitted
Analysis
Illustration of the analysis view on a fictitious document. It does not reproduce the result of a real analysis.

Verifiable reports

A finding is only as useful as the record behind it.

Each report identifies the examined file and report data with SHA-256 fingerprints. Source locations and evidence references help your team review findings and share a traceable record. Validated timestamps and PDF seals provide additional proof that recipients can check independently.

  • Report ID, file SHA-256 and report-data hash
  • Findings located: pixel regions in images, passages by page or paragraph and edited pages in documents
  • Trusted timestamp when validated
  • PDF seal when validated
  • Public verification page and QR code
  • JSON for systems, PDF for people, in five languages
Image reportSpecimenFV-7K2M-9Q4X-RT8N
Assessment
Photo with local alterations
Risk level
L4Highly suspicious
Recommended action
Review manually before relying on the image

Key evidence

  1. E1R1: edited region, moderate signal
  2. E2Compression history: anomaly found
  3. E3No Content Credentials in the file
Report data hash
sha256:3f9a7c1e…b8d4e21c
Trusted timestamp
DigiCert · RFC 3161
fileverity.com/verify
Document reportSpecimenFV-3H8D-K2QW-6NZC
Finding
Parts show AI-writing signals
Flagged passages
  • P1page 3, paragraph 3
  • P2page 3, paragraph 4
Recommended next step
Review the flagged passages. Ask the submitter how the document was drafted and request the original digital file.

File observations

  1. E1Creating software recorded in the file
  2. E2No AI label or Content Credentials found
  3. E3No hidden or look-alike characters
Report data hash
sha256:a41d09e7…5c2f80b3
Trusted timestamp
DigiCert · RFC 3161
fileverity.com/verify

How a recipient checks a report

Recipients can check the report ID, match the exact file and inspect the timestamp and PDF seal. The report states which proofs were successfully validated.

  1. 01 / 04

    Look up the ID

    Enter the report ID at fileverity.com/verify, or scan the QR code, to see the results recorded at issue.

  2. 02 / 04

    Match the file

    On the verification page, check the examined image or document, or the report’s data file, against the record. The hash is computed on your own device.

  3. 03 / 04

    Check the timestamp

    Inspect the trusted timestamp and its validation status to confirm when the report data existed.

  4. 04 / 04

    Validate the seal

    Inspect the PDF seal and its validation status to check whether the report changed after sealing.

Accuracy

Measured, published and stated with its limits.

FVDOC V3 is evaluated in each supported language on text held out from development. Our accuracy page explains the measured rates and their limits. Image reports describe the evidence behind each finding and how strongly it supports the assessment.

  • 0.1–0.9%

    of human-written text received an AI finding, depending on the language: English, French, German, Spanish or Chinese.

  • 98–100%

    of fully AI-written documents in held-out test sets were detected, depending on the language.

  • 98.4–100%

    of documents written by AI models never used in development were detected.

Held-out validation of FVDOC V3, October 2026All measured error rates

Limits

Detection is strongest on text written entirely by AI and lower for rewrites, light edits and short AI passages inside human documents; a weak signal is reported as Inconclusive. Figures come from Fileverity’s own held-out validation.

Industries

Built for teams that decide on evidence.

Wherever a file can move money, open an account or reach the public, Fileverity records what it shows before your team relies on it.

  • Insurance claims

    Review claim photos and supporting documents before settlement. Check image alterations, document revisions, scanned amounts and dates, signatures and AI-generated content, with source evidence for adjusters and SIU teams.

    Insurance claim documents
  • Financial services

    Examine onboarding and lending files for image alterations, document inconsistencies, available revisions and signature findings. Review scanned fields and AI-written passages alongside issuer and account checks.

  • Newsrooms and fact-checking

    Verify user-submitted images and contributed text before publication, with Content Credentials read wherever they are present.

  • Marketplaces and platforms

    Help trust and safety teams review listing photos, seller submissions and appeals for altered images, document inconsistencies and AI-generated content, with findings linked to the submitted evidence.

  • Legal, investigations and compliance

    Establish what an image or a document showed when it was examined, tied to the exact file by its SHA-256, in a timestamped report that counsel and auditors can check independently.

  • Publishers

    Check commissioned and contributed articles, columns and manuscripts for AI-written passages before acceptance.

How it works

From upload to a verifiable answer.

  1. Submit

    Upload an image or a document in the web app, or connect your own systems under an Enterprise agreement.

  2. Examine

    Examine images for provenance, pixel evidence and editing traces. Document checks cover text and file records; document analysis includes embedded photos, scanned fields, recoverable revisions and supported PDF signature validation.

  3. Decide

    Review the findings and recommended actions. Request an original, confirm a field with the issuer or escalate an inconsistency, then save the full report with your decision record.

Trust

Private by design. Built on open standards.

Your uploaded files are deleted after analysis and never used to train models. Temporary private evidence is cleared after acknowledged delivery or expiry; save your full report within 3 hours of receipt in the current tab.

Standards we read and use

  • Content Credentials (C2PA)

    Read and validated against the C2PA specification, including the manifest’s signature and integrity.

  • AI labels

    Machine-readable AI labels are read and reported as declarations, including China’s GB 45438-2025, in the context of labeling rules such as EU AI Act Article 50 and California’s AI Transparency Act.

  • RFC 3161

    Validated RFC 3161 timestamps provide independently checkable evidence that report data existed at the stated time.

  • SHA-256

    Files and report data are fingerprinted with SHA-256 (FIPS 180-4), so every report names exactly one file.

Fileverity supports your compliance work with verifiable records and documentation on request.

Access

Start with one file. Connect your whole workflow.

Evaluate Fileverity on your own files today, with no subscription. Bring it into your systems under an Enterprise agreement.

01

Self-serve

For teams examining files one at a time in the web app.

  • One free image check and one free document check per account, with a result summary and detailed evidence available to unlock
  • Full report: images from US$2.99 (10 for US$23.90, 100 for US$159), documents from US$8.99 (10 for US$69.90, 100 for US$489), with separate credits
  • Save before temporary evidence expires
  • No subscription; every purchase comes with an invoice

View pricingplus applicable tax

02

Enterprise

For claims, onboarding, editorial and investigation workflows at volume.

  • API access and batch processing
  • Integration with claims, onboarding or editorial systems
  • On-premises deployment, fully offline
  • Custom development, much lower per-check pricing for volumes above 1,000 checks, a DPA and documentation for vendor reviews

Questions

Frequently asked questions

What does Fileverity verify?

Fileverity examines images and documents for authenticity concerns. Image checks cover AI generation, local edits and provenance. Document verification covers text and file records; document analysis includes embedded-photo analysis, scanned amount and date checks, recoverable revisions and supported PDF signature validation. Upload JPEG, PNG or WebP images, or PDF, DOCX, DOC, RTF, ODT, PPTX and TXT documents up to 100 pages and 50 MB.

How does Fileverity support business verification?

Fileverity brings image and document checks into a traceable review record. Findings identify the relevant region, passage or field, explain the supporting evidence and recommend a next step. PDF and JSON reports help your team share the analysis and document its decision.

How does Fileverity tell whether a photo was generated or edited?

Fileverity assesses whole-image generation and local alterations separately. The report may identify a likely model family when the evidence supports it. For altered photos, supported findings can include marked regions, including signs of generative fill. Content Credentials, metadata and AI labels provide additional context; no single result establishes authenticity on its own.

How does Fileverity detect AI-written text in a document?

Our own model, FVDOC V3, evaluates each paragraph with enough running text in English, French, German, Spanish or Chinese, and flags the passages that read as AI-written, with page or paragraph references. A document needs at least about 150 words of text; document analysis also reads scanned pages. For identity documents and bank statements, Fileverity complements template and issuer checks by assessing the writing and the file.

Does AI-writing detection prove document fraud?

No. Writing signals, edit records, field discrepancies and signatures are different evidence. Review their basis and limits alongside the original, issuer and business facts.

How accurate is Fileverity, and how often does it flag genuine files?

In held-out validation of FVDOC V3 (October 2026), 0.1–0.9% of human-written text received an AI finding, depending on the language, and 98–100% of fully AI-written documents were detected. No real-world PDF in our test sets received an AI finding. Detection is strongest on fully AI-written text and lower when AI rewrites or lightly edits a human document; the accuracy page lists every condition. Image reports state the evidence behind each finding and its strength, not a score alone.

Can we share a report with counsel, auditors or a regulator?

Yes. Download the full PDF and JSON report for your case record. Recipients can check the report ID and match the file through the verification page. The report identifies successfully validated timestamps and seals. Save it before temporary evidence expires; email delivery is subject to your mail provider.

What happens to our files?

Private evidence remains in the current tab for at most 3 hours from receipt; refreshing does not extend this window. Closing the tab or signing out may clear it earlier. The server deletes temporary evidence after acknowledged delivery; undelivered results expire. Original files and generated PDFs are not a long-term archive; summaries and verification records remain. Mail providers and inboxes apply their own retention policies. After expiry, only a results copy without excerpts or previews may be available.

Does Fileverity offer an API or batch processing?

Yes, under an Enterprise agreement: API access, batch processing and integration with claims, onboarding or editorial systems. Fileverity can also be deployed on your own servers and run fully offline, without any external API, and we develop custom detectors, policies and report formats for specialist workflows. The web app is self-serve and examines one file at a time.

How much does Fileverity cost?

Every account can run one image and one document for free, with the detailed evidence locked. A credit of the matching type unlocks a free-check report within 24 hours or pays for a full report. Image credits cost US$2.99 for one credit, US$23.90 for 10 or US$159 for 100. Document credits, covering analysis of up to 15 embedded photos, cost US$8.99, US$69.90 for 10 or US$489 for 100. Credits are prepaid, plus applicable tax, with no subscription. For volumes above 1,000 checks, Enterprise agreements offer volume pricing by agreement. Private evidence remains in the current tab for at most 3 hours; refreshing does not extend it. Download promptly. Closing the tab, signing out or server expiry may prevent recovery of full evidence.

Who is responsible for Fileverity's security?

Our security and risk team designs and reviews the security, privacy and risk program. Its members hold professional credentials including ISACA's CISA, CISM, CRISC and CDPSE, with long experience in information security and technology risk in financial services. The service is designed against NIST CSF 2.0, ISO/IEC 27001, the NIST AI Risk Management Framework, COBIT 2019, PIPEDA and GDPR. These are frameworks we design against, not credentials of Fileverity itself. Enterprise security reviews are handled directly by the team.

Put a verifiable record behind every file decision.

Tell us which files your teams review, in what volume and in which systems. We will set up access and walk your team through real cases.