Skip to content

Computer vision and AI-content detection

Models that read images and explain what they found.

We build classifiers and detectors that give a result, a confidence range and the reason behind it. It is the work behind sighting.ai.

Illustration: an image check with a marked region, a confidence range instead of a single score, and the signals behind the result. Sample data.

Illustration with sample data.

Scope

What we build.

Who it’s for

  • Trust and safety teams checking uploads.
  • Platforms that need to know whether content was made by AI.
  • Teams whose files must stay private.
  • Image understanding and classification
  • AI-content detection for text, images, video and code
  • A confidence range and the reason behind each result
  • Checks for content credentials (C2PA) and watermarks
  • Self-hosted models when files must stay private
  • Review queues for results a person should check

Approach

How it typically works. Every project is different. This is the usual shape.

  1. 01

    Define the decision

    We agree what the model must decide, and what a wrong answer costs you.

  2. 02

    Choose the model

    We test open and hosted models on your own examples, and keep the one that holds up.

  3. 03

    Explain the result

    Each result comes with a confidence range and the signal that decided it.

  4. 04

    Deploy with care

    We self-host the models when files must stay private, and route unclear cases to a person.

Models and tools

Models and tools we use.

Models

  • ModernBERT

Tools

  • ONNX
  • C2PA
  • Railway
  • Supabase
  • Next.js
  • Vercel

Names are trademarks of their owners.

Questions

Common questions.

Can the models run without sending files to an outside AI provider?

Yes. We can self-host the models, so files aren’t sent to a third-party AI service.

Can AI detection be certain?

No detector is perfect. That’s why we show a confidence range and the reason behind each result, and send unclear cases to a person.

Do you keep the files we check?

That’s your choice. sighting.ai is built to delete uploaded images and video once the analysis is done, and we can build yours the same way.

Is our data safe with you?

We don’t use client data to train shared models without your agreement. We can sign an NDA and a data processing agreement before we see your data.

Security and privacy

Careful with your data.

  1. Registered in Ireland

    Foldox Limited is an Irish company and follows the EU GDPR.

  2. Your data stays yours

    We don’t use client data to train shared models without your agreement.

  3. Agreements first

    We can sign an NDA and a data processing agreement before we see your data.

Other services

More we can build.

Start a project

Need to check images or AI content?

Tell us what you need to check, and why. We aim to reply within two working days.