Detect AI images.
Show the evidence.

Sightova checks every image three ways: a pixel model built for today's generators and AI editors, the invisible watermarks Google and OpenAI embed, and signed content credentials. You get a score, a verdict and the evidence behind them, ready for an analyst, an investigator or a review team.

Request a demoRead the API docs

Or try it free in the dashboard. No card needed.

A face swap: one AI-generated man's face placed into another AI-generated man's café photo
AI Generated
  • Pixel model95% AI
  • Invisible watermarkOpenAI
  • Content credentialsSigned by OpenAI
A face swap we made with GPT-Image 2.5. Neither man exists.
3
Independent checks in every scan
50+
Generative models in our benchmark
EEA
Images analysed in Germany and Iceland
REST
API access on every plan
// HOW IT WORKS

Three independent checks in every scan.

Each image is analysed by a pixel model, a watermark detector and a provenance check. The results are returned together, with the evidence for each finding.

01 · Pixel analysis

Pixel-level detection

Detects fully AI-generated images and AI-edited regions in real photos by analysing the pixels. Works on files with stripped metadata.

02 · Watermarks

Invisible watermark detection

Detects Google SynthID and OpenAI invisible watermarks. A detected watermark identifies the provider and sets the verdict to AI-generated.

03 · Provenance

C2PA and metadata verification

Reads C2PA content credentials and EXIF metadata. A manifest with a valid, trusted signature identifies the generator or capture device.

Output · Heatmap

Manipulation heatmap

Highlights regions likely edited with AI, so reviewers can verify them directly. Starter plan and above.

Output · Score

Probability score and verdict

AI probability from 0 to 100% and a verdict for every image. Set your own threshold for review.

Output · History

Scan history

Every scan is stored in your account with its scores and metadata for later review.

// TEST RESULTS

Detection without watermarks or metadata.

A real photo, then a ChatGPT edit of it scanned at each stage of re-sharing. The watermark and C2PA credentials are lost along the way. Pixel analysis flagged every edited copy.

  1. Original

    Authentic

    Original photo

    Real photo, before the edit

    Pixel

    Under 1%

    Watermark

    None

    C2PA

    None

  2. Step 1

    AI Generated

    ChatGPT edit

    Pixel

    96% AI

    Watermark

    Found

    C2PA

    Signed

  3. Step 2

    AI Generated

    Metadata removed

    Pixel

    96% AI

    Watermark

    Found

    C2PA

    Removed

  4. Step 3

    AI Generated

    Shared via chat app

    Pixel

    95% AI

    Watermark

    Found

    C2PA

    Removed

  5. Step 4

    AI Generated

    Cropped and re-compressed

    Pixel

    95% AI

    Watermark

    Not found

    C2PA

    Removed

Pixel analysis reads the image content, not the file's metadata, so it does not depend on a watermark or credentials being present.

Scanned on the live model, 10 November 2026. Original photo: Dollar Street by Gapminder, CC BY-SA 4.0. Edited by Sightova with GPT Image 2.5; the edited images are shared under CC BY-SA 4.0.

Testing methodology
// SECTORS

Built for government, law enforcement and enterprise.

Image verification with documented findings, in the dashboard or through the API.

Government & public sector

Verify images before they are used in statements, briefings and public communication, and assess images circulating during incidents.

  • Images analysed in Germany and Iceland (EEA)
  • DPA, MSA and subprocessor list published
  • Enterprise agreements for agencies

Law enforcement & investigations

Triage images from tips, case files and open sources. Each result shows which check flagged the image.

  • Watermark and credential findings per image
  • Heatmap of edited regions
  • Scan history in your account

Enterprise trust, fraud & risk

Screen claim photos, listings, profile images and submitted evidence through the API.

  • One REST API call per image
  • Configurable review threshold
  • API access on every plan
// WORKFLOW & API

Integrates with your existing workflow.

Use the dashboard for manual review, or the REST API to send results to case management, claims and moderation systems.

  1. 01

    Submit

    Upload via the dashboard, or send one image per API request as a file, URL or base64. JPEG, PNG, GIF and WebP, up to 50 MB.

  2. 02

    Analyse

    Pixel analysis, watermark detection and C2PA and EXIF verification run on every image. Results are returned in seconds.

  3. 03

    Review

    The response contains the score, verdict and evidence as JSON. Apply your own threshold and route flagged images to review.

Read the API documentation
POST /api/v1/detect/ai

Request

curl -X POST 'https://sightova.com/api/v1/detect/ai' \
  -H 'X-API-Key: sk_<your-key>' \
  -H 'Content-Type: application/json' \
  -d '{"image_url": "https://example.com/evidence.jpg"}'

Example response · Gemini edit shared via chat app

{
  "request_id": "req_5oQjGqU4mB2nQp9Hg7vTz",
  "endpoint": "/api/v1/detect/ai",
  "result": {
    "ai_probability": 0.96,
    "real_probability": 0.04,
    "confidence": 0.92,
    "verdict": "ai-generated",
    "watermark": {
      "detected": false,
      "provider": null,
      "openai": false,
      "google": false
    }
  },
  "metadata": { "exif": {}, "warnings": [] }
}

Detected watermarks and C2PA credentials are returned in the same response.

// DATA HANDLING

Processing, retention and contracts.

Where images are processed, what is retained, and the agreements available for review.

EEA processing

Application and database are hosted in Nuremberg, Germany. Detection also runs on a GPU server in Reykjavík, Iceland. Traffic is routed through Cloudflare over TLS 1.2 or higher.

Deleted after analysis

Uploaded images are written to a temporary file during analysis and deleted when the scan completes.

Not used for training

Scanned images are not used for model training. Images are retained only when you submit them as feedback.

Published DPA and MSA

The DPA, MSA and subprocessor list are publicly available for legal and procurement review.

Enterprise agreements

Volume pricing, contract terms and self-hosted deployment can be discussed under an enterprise agreement.

Published test results

Test methodology and results on frozen test sets are published. See the testing methodology.

Visit the Trust Center
// FAQ

Frequently asked questions.

For security, legal or procurement questions, contact us.

Three checks run on every image. The pixel model returns an AI probability. A detected Google SynthID or OpenAI watermark sets the verdict to AI-generated. C2PA content credentials with a valid, trusted signature that identify a generator or camera confirm the result. Each finding is shown next to the score.
Use the result to prioritise review. The score indicates the likelihood that an image was generated or edited with AI. Detected watermarks and signed credentials are properties of the file that can be verified independently. We recommend human review before any decision that affects a person.
The benchmark covers 50+ generative models, including Midjourney, OpenAI, Stable Diffusion, Flux and Google's image models. Testing is extended as new generators are released. When an image carries a SynthID or OpenAI watermark, or C2PA credentials, the result identifies the tool.
Images are processed on servers in Germany and on a GPU server in Iceland (EEA). Uploaded images are held in a temporary file during analysis and deleted when the scan completes. Scan records (scores, filename and metadata) are kept in your account history for the life of the account. Images submitted as feedback are retained for review.
Self-hosted deployment can be discussed under an enterprise agreement. Contact sales with your requirements.
JPEG, PNG, GIF and WebP, up to 50 MB, one image per request. Upload via the dashboard, or send to the REST API as a file, image URL or base64.
Yes. AI Video Detection scores frames across a video and returns a verdict with a frame-by-frame timeline. It is included on the Starter, Premium and Enterprise plans. See the AI Video Detection page for details.
// MORE FROM SIGHTOVA

NSFW & Violence Detection

Score images for explicit and violent content before they reach people.

Document Forgery Detection

Flag edited regions in IDs, receipts, invoices and other documents.

AI Video Detection

Detect AI-generated video frame by frame, with a timeline of the flagged moments.

Test Sightova on your own images.

Request a demo, or start a free account and run your first scans today.

Request a demoStart free