What Heraldry and Mon Can Teach Us About Building Visual-Identity Generators
This insightful piece delves into the computational challenges of generating visual identities by comparing the rule-based grammar of Western heraldry with the compositional geometry of Japanese mon. It argues that effective procedural generation hinges on understanding the underlying operations and relationships, rather than merely cataloging symbols. The article offers a novel architectural perspective for those building generative design systems, making it a captivating read for software architects and generative artists alike.
The Lowdown
Ben Overmyer, creator of Iron Arachne, explores how to build effective visual identity generators by dissecting and comparing two distinct historical systems: Western heraldry and Japanese mon. He argues that moving beyond simple asset catalogs to understand the inherent 'grammar' and 'operations' of these systems is crucial for truly procedural generation.
- Glossary of Terms: The article begins by defining essential concepts such as procedural generation, blazon, mon, grammar, corpus, abstract syntax tree, and scene graph, setting the stage for a technical discussion.
- Western Heraldry as a Generative System: Overmyer explains that Western heraldry behaves like a formal language with a 'blazon'—a structured description (e.g., "Azure, a lion rampant Or") that dictates visual elements. This system lends itself to a descriptive generation model, similar to processing an abstract syntax tree.
- Japanese Mon as a Compositional System: In contrast, Japanese mon lack a direct blazon equivalent. Their identity stems from highly stylized motifs and compositional patterns (repetition, rotation, symmetry), suggesting a more direct, graphical-primitive-based generation model akin to a scene graph.
- Two Different Generation Strategies: The core insight is that Western heraldry can be treated as primarily descriptive (generate text, then render), while mon are primarily compositional (generate visual structure directly by transforming graphical elements).
- Rules Over Symbol Lists: A key takeaway is that generators must model the rules and relationships (e.g., the "rule of tincture" in heraldry, or symmetry in mon) rather than just randomly selecting from a list of symbols, to produce historically plausible designs.
- The Problem of Exceptions: Historical systems are messy. The article notes that exceptions (like the Kingdom of Jerusalem's metal-on-metal arms) necessitate distinguishing between a general 'grammar' for generation and a 'corpus' of historically attested (and sometimes rule-breaking) examples.
- Rethinking Iron Arachne's Approach: Overmyer questions if a fantasy crest generator should primarily model blazon, suggesting that combining the grammar of meaningful categories (from Western heraldry) with the grammar of visual composition (from mon) could yield better results.
- Generation Difficulty: While Western heraldry seems easier initially due to its formalized language, both systems become profoundly complex when aiming for historical depth and variety.
- Finite State Machines vs. Hierarchical Systems: The author briefly considers finite state machines for generation but concludes that hierarchical representations (trees, graphs) are better suited for the nested and interacting elements of both heraldry and mon.
- The Bigger Lesson: The most valuable lesson is to focus on the operations (divide, repeat, mirror, rotate) that exist within a visual domain, rather than just the objects (lions, eagles, flowers), for powerful procedural generation.
Ultimately, the article champions an approach to procedural generation that prioritizes understanding the underlying generative rules and visual grammars over merely curating vast asset libraries. This perspective extends beyond heraldry, offering a framework for generating other coherent visual traditions from tartans to fictional writing systems.