Understand what the detector analyzes, how the models reach a result, and where their limits are. Learn more →

Inspect media.
Detect the AI signal.

Analyze images, audio and English text. Review the AI score, explore the evidence and decide what needs a closer look.

Start scanning How it works
Understand the system

Built to explain a result, not just display one.

The lower section explains what is analyzed, how the selected media moves through the system, and what limitations still apply to a model-based authenticity screening result.

How the detector works

The application validates the upload, prepares it for the selected detector, runs local inference, compares the model output against its configured decision boundary, and returns the available evidence.

What can affect the result

Compression, screenshots, transcoding, editing, new generators, and mixed human-and-AI workflows may change the signal available to a detector.

Interactive pipeline

Follow the media through the detector.

Switch between media types to see the processing path used before the result reaches the interface.

01
Upload Image enters the detector.
02
Prepare Resize, crop and normalize.
03
Infer Run the local image model.
04
Result Return score and evidence.

System walkthrough

Move through a concise overview of the project, architecture, and the correct way to interpret a detector result.

01 / PURPOSE

Turning detection into a usable workflow.

The project is more than a standalone model. It combines media selection, preprocessing, local inference, authentication, explanation and result presentation inside one application.

  • 1 Web-based scanning interface.
  • 2 Multiple media pathways.
  • 3 Human-readable analysis output.
Focus Practical media screening
Interaction Select → upload → analyze → review
Interface objective Clear and trustworthy
02 / ARCHITECTURE

Authentication and inference stay on the backend.

Django manages user authentication and sessions while the detector pipeline handles supported media without asking the browser to decide whether a user is authenticated.

  • 1 Django authentication.
  • 2 AI detector evaluation.
  • 3 Server-side detector access.
Backend Django
Image Community Forensics
Audio ArtifactNet v94
03 / INTERPRETATION

The result should be read as evidence.

A detector can still produce false positives and false negatives. Thresholds, provenance, source context, compression and human review remain relevant.

  • 1 Review the probability.
  • 2 Review the decision threshold.
  • 3 Use additional context where possible.
Output Probabilistic screening
Explanation Available model evidence
Best practice Combine evidence and context