Gaël Lejeune is a lecturer in computer science at the Faculty of Letters, and Mathieu Marly is a historian and editorial manager of the Encyclopedia of European Digital History.

In an era of visual manipulation and AI-generated content, this project provides tools to better equip students to face the challenges of disinformation.

Could you introduce yourselves and tell us how Virapic came about?

Mathieu Marly: As an associate professor and researcher at Sorbonne University and editorial manager of the Encyclopedia of European Digital History (EHNE), I work to bridge historical research and education in both higher education and secondary schools.

Approved by the French Ministry of Education, EHNE now offers 1,200 entries and scientific articles related to school curricula, attracting 5,000 to 10,000 daily readers. As part of EHNE, we’ve previously worked on projects involving photography, such as the Colbert Project, which used AI to analyze historical photographic collections, and Chambre Noire, which enriched photographs with historical commentary and methods.

When sourcing educational images for EHNE that could serve as historical documents, I observed that many photographs online are decontextualized, lacking captions or commentary. This observation, combined with my research on photographic archives, led to the idea of a project aimed at better contextualizing and leveraging these images. That’s how Virapic was born.

Gaël Lejeune: Virapic is also part of the dynamic within the Center for Experimentation in Digital Methods for Humanities and Social Sciences Research (CERES), a unit established in 2021 within the Faculty of Letters to support digital humanities research through consulting, training, and tool development. As a computer science lecturer at the UFR for Sociology and Computer Science in the Humanities, I serve as its deputy director, with research focusing on natural language processing and, more recently, image analysis.

What are the objectives of this project?

M.M.: Virapic has three key goals. First, it aims to analyze the online circulation of photographs related to complex historical events, such as world wars, political and social struggles, genocides, and colonization. The objective is to understand why and how these images go viral and what their dissemination reveals about contemporary memories of these events.

Second, the project seeks to reintroduce academic content into photographic archives by mobilizing historians to contextualize these images. This approach highlights researchers' work, demonstrates how they analyze photographic archives, and provides internet users with reliable and rigorous academic content.

G.L.: The final goal is to directly impact users' online practices by leveraging EHNE’s significant web traffic. Given EHNE's strong search engine ranking, photographs featured on the site can appear among the top search results for users. This helps promote verified content before images are misinterpreted or shared out of context.

Why is it essential to educate students about photographic archives?

M.M.: Virapic targets both history students and secondary school students. Younger generations have grown up with digital tools, but this familiarity doesn’t always translate into critical awareness. In the digital age, photography has shifted from a tool for preservation to a conversational medium, especially on platforms like TikTok and Snapchat. This transformation profoundly influences how students perceive images, often without questioning their reliability or context.

As historians and educators, it’s crucial to encourage students to recognize that photographs don’t inherently convey truth or serve as evidence on their own—they require critical examination, including rigorous and methodical analysis.

G.L.: Today’s students are tomorrow’s educators and citizens, who will need to apply, share, and teach this knowledge. Virapic trains them to navigate a complex digital environment where search engine algorithms heavily influence content visibility. The project helps them question these mechanisms and distinguish reliable content from manipulated material.

This is particularly relevant in the era of deepfakes, isn’t it?

M.M.: Absolutely. The Ministry of Education, with whom we collaborate, is highly concerned about AI-generated content. These technologies have progressed exponentially. While AI-generated fake photographs were easy to spot three years ago, they are now much more realistic and could become nearly undetectable in a few years. This poses significant risks, particularly regarding conspiratorial reinterpretations of history or manipulations of collective memory.

Take, for instance, a country like Russia attempting to historically justify claims over Ukraine. By flooding the internet with such images, these materials could become a pseudo-historical narrative, influencing perceptions and blurring the line between truth and falsehood. Photographs, more than texts, have the power to anchor ideas in the public’s mind by appearing as visual proof.

G.L.: Virapic doesn’t simply debunk content by declaring what’s “true” or “false.” Instead, the project teaches students and pupils historical methodology—how to analyze archives, understand AI, and explore digital virality. These skills are critical in a world where history must adapt to the digital realities faced by learners.

We believe that providing methodological tools is far more effective than merely asserting whether information is correct. This approach fosters critical thinking rather than imposing an authoritative truth.

Moreover, while AI can multiply interpretations and even manipulations, it can also be a solution to these issues. In our training, we aim to show this duality: every tool has positive aspects and risks. Our goal is to teach users to leverage digital tools’ potential while staying aware of their limits and associated dangers.

How do you select the photographs used in the project?

M.M.: The first step relies on historians. We gather specialists to identify problematic photographs found online. For example, we’ve worked with historians on a 1933 photograph depicting Hitler’s rise to power—a propaganda image often reused, even by reputable outlets like Radio France.

G.L.: The second approach involves CERES teams conducting online monitoring based on keywords related to historical events. This research identifies frequently recurring images, analyzes their circulation, and complements the identification of problematic photographs by detecting those that become viral. These two approaches converge to create a relevant corpus for analysis and contextualization.

What’s the future of Virapic?

G.L.: After securing the Emergence project funding in 2022, we began concrete work in 2023, and the project will run until 2025. At this stage, we’ve laid the groundwork. The next step is to sustain the project and expand its scope by incorporating new resources and broadening our collaborations. This would include not only additional computer scientists but also researchers from other fields, such as information and communication sciences or sociology.