Introduction: From Textual Restoration to Visual Interpretation

The application of Artificial Intelligence in Western Classics has recently achieved paradigm-shifting results in textual criticism. Groundbreaking studies such as DeepMind’s Ithaca (Assael et al. 2022) and the newly published Aeneas (Assael et al. 2025) have demonstrated that deep neural networks can restore and attribute ancient inscriptions with superhuman accuracy. However, while these tools aid the expert researcher in reconstructing the text, a significant gap remains in reconstructing the visual imagination required to comprehend these texts. Classical literature often remains abstract to modern audiences. This poster introduces a novel project utilizing "Humanitext Antiqua," a Retrieval-Augmented Generation (RAG) system designed to support the broad spectrum of Western Classics. Our current pilot project applies this system to visualize narratives from Greek mythology—such as the Constellation of Asclepius, Arion and the Dolphin, and Midas—demonstrating a new workflow for engaging with classical texts through AI-assisted visualization.

Methodology: The Humanitext Pipeline and Evolving Workflows

The Humanitext system grounds generative AI in philological evidence using a RAG architecture indexed with data from the Perseus Digital Library. Our current workflow addresses the challenge of narrative consistency through a multi-stage process:

• Source Verification: The system retrieves and presents relevant source text chunks for verification.
• Image-to-Video Synthesis: To maintain character consistency—a notorious challenge in generative media—we first generate a high-fidelity static image using Google’s Nano Banana Pro. This serves as an "anchor." We then animate this image using Veo 3.1 to create an 8-second clip.
• Refinement: We utilize Google’s flow tools to crop clips or extend scenes, smoothing transitions to create a natural narrative flow. Given the precipitous pace of advancement in generative media, the workflow described here represents our current best practice; in our poster presentation, we intend to showcase the most optimized workflow and tools available at the time of the conference.

Pedagogical Significance 1: Expert Curation and Public Outreach

The democratization of visualization tools brings both risk and opportunity. The primary significance of this project lies in defining the role of the researcher as a "curator" of the latent space. Because AI models frequently "hallucinate" anachronisms, rigorous expert supervision is non-negotiable to ensure historical plausibility. However, once this scholarly supervision is applied, the resulting media becomes a powerful vehicle for public outreach. By converting complex textual data into dynamic, verified imagery, we can present the richness of Western Classics to a general audience in a format that is immediately engaging, potentially sparking new interest in the field among non-specialists.

Pedagogical Significance 2: The Value of Creation (Making)

Furthermore, the process of creation offers profound educational value. In the Journal of Classics Teaching, Morrice et al. (2025) argue that using AI to create "accurate illustrations" of set texts transforms the activity into a close-reading exercise. We extend this logic to video production. While generating a static image requires understanding physical details, generating video demands an understanding of temporal and narrative dynamics as well. To correct the AI’s output, the user is forced to return to the verified source text provided by Humanitext Antiqua. They must distinguish between what the text explicitly says and what must be logically inferred. This iterative cycle of prompting, checking against the source, and re-prompting turns the "hallucination" problem into a critical thinking exercise, transforming the user from a passive consumer into an active editor of the ancient world.

Conclusion

This project demonstrates that Generative AI can serve as a rigorous engine for inquiry and outreach. By integrating the source-checking capabilities of RAG systems with the creative engagement of video synthesis, we establish a framework where the act of "making" history with AI demands a rigorous, evidence-based engagement with the past.