identifAI New Feature: XAI – eXplainable AI

More than just a verdict. Discover what the AI has observed to reach its decision.

Why did Revelio classify an image or video as AI-generated? Which elements of the content captured its attention the most?

With the new XAI – eXplainable AI feature, identifAI takes another step towards greater transparency and interpretability of its classification models.

XAI is a new analysis scope that enriches Revelio's verdicts with a set of visual and descriptive tools designed to help you understand where the model focused its attention during the analysis.

From Classification to Understanding

When XAI is enabled, Revelio models do more than simply return a verdict: they also generate insights into the areas of focus that shaped the classification process.

This information is presented through three complementary tools:

• Attention Map – See where the model focuses its attention. A visual representation overlaid on the original content, highlighting the areas the model focused on most. An immediate and intuitive way to explore what Revelio has observed.
• Attention Graph – Explore the analysis in greater depth. A graph showing the distribution and peaks of the model's attention, providing a more technical and detailed perspective for those who want to gain deeper insights into the algorithm's behaviour.
• AI-Powered Explanation – From Images to Words. A natural-language description of the regions that received the most attention, available in multiple languages. This makes the information generated by the model even more accessible and easier to interpret.

Greater Transparency, More Tools to Interpret Results

With XAI, identifAI provides new insights to supExplainable AI (XAI): Bringing Transparency to AI Content Detection

More than a verdict. Understand what AI sees when analysing images and videos.

As generative AI becomes increasingly sophisticated, distinguishing authentic content from AI-generated images and videos is becoming more challenging. AI detection technology is essential, but knowing whether content is synthetic is only part of the picture. Understanding how a model reaches its decision matters too.

This is where Explainable AI (XAI) comes in.

With its new XAI functionality, identifAI takes another step towards more transparent and interpretable AI detection. Integrated into Revelio, its AI-generated content detection technology, XAI provides visual and descriptive insights into the model's analysis, helping users better understand where it focused its attention.

Beyond the Verdict: Understanding AI Decisions

Traditional AI models return a classification result. Explainable AI goes a step further, providing additional information that helps users explore how a model processes content.

With XAI enabled, Revelio enriches its classification results with three complementary tools, offering different perspectives on the model's attention during analysis.

1. Attention Map: See What AI Sees

The Attention Map provides a visual representation overlaid on the original image or video, highlighting the areas that received the most attention during analysis.

An intuitive way to explore which regions attracted the model's attention and gain a clearer visual perspective on its analysis.

2. Attention Graph: Explore the Details

For a more technical perspective, the Attention Graph visualises the distribution and peaks of the model's attention.

It allows users to explore attention patterns in greater detail, providing additional insights into the classification process.

3. AI-Powered Explanation: From Visuals to Words

The AI-Powered Explanation transforms attention-related information into natural-language descriptions, identifying the regions that received the most attention.

Available in multiple languages, it makes technical insights easier to understand, interpret and communicate.

Why Does Explainable AI Matter?

As AI-generated content becomes increasingly realistic, transparency is becoming a key element of reliable AI detection.

XAI provides additional context alongside traditional classification results, helping users:

• Understand detection results through visual and descriptive insights.
• Explore model behaviour by identifying areas of particular attention.
• Improve transparency with additional information about the analysis.
• Make more informed assessments when evaluating potentially AI-generated content.

Importantly, attention visualisations are interpretability aids, not definitive proof of how a model reached its decision. They complement the classification verdict rather than replace it.

Available on the Web Platform and via API

identifAI's XAI functionality is currently available for image and video analysis, both through the identifAI web platform and via API.

This makes it possible to explore explainability directly within the platform or integrate these capabilities into existing applications and workflows.

What about audio? We're working on it! The goal is to extend XAI capabilities to audio content analysis in the future.

The Future of AI Detection Is About Understanding

AI detection should be about more than identifying synthetic content. It should also make results easier to understand and interpret.

With XAI, identifAI continues to advance towards more transparent, understandable and accessible AI detection.

Because knowing the verdict is important. Understanding what lies behind it makes the difference.

Frequently Asked Questions

What is Explainable AI (XAI)?

Explainable AI refers to techniques that make AI model behaviour and outputs easier to understand by providing insights into how models process information.

What is an Attention Map in AI detection?

An Attention Map highlights the regions of an image or video that received the most attention from an AI model during analysis.

Does identifAI's XAI support video analysis?

Yes. XAI is currently available for both image and video analysis through the identifAI web platform and API.

Can XAI explain why an image is AI-generated?

XAI provides insights into the areas that received the most attention during classification. However, these visualisations do not constitute definitive proof of the model's reasoning.

Is XAI available through an API?

Yes. Developers and organisations can access XAI capabilities through the identifAI API and integrate them into existing workflows.

port the interpretation of results, complementing traditional classification with visual, analytical and descriptive perspectives.

A step forward towards increasingly transparent, understandable and accessible artificial intelligence.

Available on the Web Platform and via API

XAI is currently available for the analysis of images and videos, both through the identifAI web platform and via API, making it easy to integrate into existing workflows.

What about audio? We're working on it! Our goal is to extend these capabilities to audio content analysis soon.

‍