From Pixels to Metal: AI-Empowered Numismatic Art

Penousal Machado, Tiago Martins, João Correia, Luís Espírito Santo, Nuno Lourenço, João Miguel Cunha, Sérgio Rebelo, Pedro Martins, João BickerIJCAI 2024

Introduction

At the IJCAI-24 conference, researchers from the University of Coimbra presented a pioneering project at the intersection of ancient numismatic tradition and modern artificial intelligence. In collaboration with the Portuguese National Press-Mint (INCM), the team designed a legal-tender 10-euro silver coin themed around the "digital world." This was achieved through a custom-built Evolutionary Computation (EC) framework, resulting in a physical artifact that serves as both a piece of art and a machine-readable record of its own algorithmic creation.

The Challenge of Digital Numismatics

Designing a coin is a high-stakes endeavor governed by strict physical, cultural, and technical constraints. Beyond aesthetic appeal, a coin must be "mintable"—it must be capable of being struck in metal with a specific relief height and must maintain durability over time.

When tasked with the theme "digital world," the authors sought to avoid the "clichéd and derivative imagery" often produced by standard generative AI tools like Midjourney or DALL-E. They observed that these models frequently generated "pastiches"—such as human faces merged with circuit board patterns—which felt unoriginal and failed to capture the native essence of the digital realm.

Example of clichéd generative AI imagery the researchers sought to avoid.

Methodology: Evolutionary Art with pyNEvAr

To create a design truly native to the digital world, the researchers developed pyNEvAr, an Evolutionary Art System. Unlike "black-box" diffusion models, pyNEvAr uses Genetic Programming (GP) to evolve mathematical expressions.

Genetic Programming and Heightmaps

Instead of manipulating pixels directly, the system evolves formulas that are evaluated across a 2D space. This process generates a heightmap, which translates mathematical values into a 3D relief suitable for the physical minting process. To ensure the designs felt "numismatic," the system utilized polar coordinates ($\alpha, r$) rather than standard Cartesian coordinates, naturally producing the circular, symmetrical patterns characteristic of coins.

CLIP-Guided Fitness and MetaPrompter

To guide the evolution toward the desired theme, the researchers employed CLIP (Contrastive Language–Image Pre-training) to score how well an evolved image matched a text prompt. To further refine this, they developed MetaPrompter, a tool that evolves the prompts themselves using a grammar derived from the Mint’s historical archives. By interrogating 910 contemporary coins and extracting 400 descriptors, the AI was "primed" with a vocabulary consistent with the Mint's heritage.

The Two Faces of the Coin: Genotype and Phenotype

The final design embodies a "two faces" philosophy, conceptualized as a dialogue between human and machine.

The final 10-euro silver coin: The reverse (left) shows the AI-generated artwork, while the obverse (right) encodes the mathematical "genotype."
  • The Reverse (Phenotype): This side displays the aesthetic result of the evolutionary process. It is an abstract representation of the digital world, favoring mathematical novelty over human-centric imitation.
  • The Obverse (Genotype): This side features a machine-readable graphical representation of the mathematical formula used to generate the reverse. Rendered as a series of concentric arcs reminiscent of a data storage disk, it acts as the "genetic code" of the coin.

Results and Impact

The project culminated in the production of 4,000 silver proof coins. The final design was selected from a massive search space, with the system evaluating approximately 43 million images across 420 evolutionary runs. The market response was overwhelmingly positive; the coins, priced at €95 each, sold out in less than two months. This success signals a strong public appetite for artifacts that bridge the gap between traditional craftsmanship and algorithmic innovation.

Conclusion

The "Digital World" coin project demonstrates that AI can be a powerful co-creative partner when guided by specific domain constraints and historical context. By moving away from data-driven imitation toward mathematical exploration, the researchers created a unique piece of numismatic history. Future work aims to refine the machine-readability of physical genotypes and apply these evolutionary methods to other traditional crafts, such as jewelry and architecture.

Summary updated

BibTeX
@inproceedings{machado2024from,
  author = {Penousal Machado and Tiago Martins and João Correia and Luís Espírito Santo and Nuno Lourenço and João Miguel Cunha and Sérgio Rebelo and Pedro Martins and João Bicker},
  title = {{From Pixels to Metal: AI-Empowered Numismatic Art}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2024},
}
Figure