Back to AI Research

AI Research

From LLM-Driven Trading Card Generation to Procedur... | AI Research

Key Takeaways

  • From LLM-Driven Trading Card Generation to Procedural Relatedness: A Pokémon Case Study This paper explores a new way to combat the "stale" metagames that of...
  • Since the dawn of Trading Card Games, the genre has grown into a multi-billion-dollar industry engaging millions of analog and digital players worldwide.
  • Popular TCGs rely on regular updates, balance adjustments, and rotating constraints to sustain engagement.
  • Yet, as metagames stabilize, predictable strategies dominate and viable card options diminish, often resulting in repetitive and impaired player experiences.
  • This paper investigates the use of Large Language Models and Image Diffusion Models for Procedural Content Generation of TCG cards, addressing these challenges by enabling a personalized infinity of card designs.
Paper AbstractExpand

Since the dawn of Trading Card Games, the genre has grown into a multi-billion-dollar industry engaging millions of analog and digital players worldwide. Popular TCGs rely on regular updates, balance adjustments, and rotating constraints to sustain engagement. Yet, as metagames stabilize, predictable strategies dominate and viable card options diminish, often resulting in repetitive and impaired player experiences. This paper investigates the use of Large Language Models and Image Diffusion Models for Procedural Content Generation of TCG cards, addressing these challenges by enabling a personalized infinity of card designs. Modern generative AI not only enables large-scale content creation but could even introduce procedural relatedness, fostering unique connections between players and their cards. We present a pipeline combining player-centric co-creation, fine-tuned embeddings, local LLMs, and Diffusion Models to generate dynamic, personalized cards while potentially expanding creative range. We evaluated the pipeline in a user study with 49 participants who generated 196 Pokémon card samples. Participants rated aesthetics and representativeness of visuals and mechanics, and provided qualitative feedback. Results show high satisfaction and indicate that most participants successfully realized their own ideas through prompt adjustments. These findings lay groundwork for future content generation systems and alternatives to conventional metagame evolution through procedural relatedness.

From LLM-Driven Trading Card Generation to Procedural Relatedness: A Pokémon Case Study
This paper explores a new way to combat the "stale" metagames that often plague popular Trading Card Games (TCGs). As games mature, predictable strategies often dominate, leaving players with fewer viable options and repetitive experiences. The authors propose a system that uses generative AI to create an infinite, personalized supply of new cards. By combining Large Language Models (LLMs) and image generation, the system aims to foster "procedural relatedness"—a concept where players feel a unique, personal connection to the cards they help create, rather than just using the same powerful cards as everyone else.

How the Pipeline Works

The researchers developed a five-step technical pipeline to turn simple user prompts into fully realized trading cards. First, they mined a database of over 15,000 existing Pokémon cards to create a knowledge base of game mechanics. When a user provides a name, type, and short description, the system uses Retrieval-Augmented Generation (RAG) to find similar existing cards. This context is fed into an LLM, which generates the card's stats, such as HP, attacks, and weaknesses, in a structured format. Simultaneously, a diffusion model—fine-tuned with specific artistic styles—generates the card's visual artwork. Finally, these components are synthesized into a finished, printable card design.

User Study and Feedback

To test the system, the researchers conducted a study with 49 participants who generated 196 unique Pokémon cards. The participants were asked to evaluate the aesthetics of the generated images and the representativeness of the game mechanics compared to their original ideas. The results were largely positive, with high satisfaction ratings across the board. Most participants reported that they were able to successfully realize their creative visions by adjusting their prompts, suggesting that the system provides enough control for players to feel a sense of ownership over the content they generate.

The Future of Procedural Relatedness

The study provides early evidence that generative AI can successfully bridge the gap between player ideation and game design. By allowing players to participate in the creation of their own game pieces, the system offers a potential alternative to the traditional, developer-led model of content updates. While the current implementation focuses on Pokémon, the authors suggest that this framework could be adapted to other card games, potentially shifting the focus of TCGs from chasing the "strongest" cards to building a collection that reflects the player's personal narrative and creative choices.

Comments (0)

No comments yet

Be the first to share your thoughts!