For Freedom Typography project, by Safiyeh Niknami (MFA 2025, Ddes).

Training the Future to Remember: A Cultural Preservation Approach to Nastaliq

By: Safiyeh Niknami, Doctor of Design in Cultural Preservation Student

The focus of my Doctor of Design research is cultural preservation at the intersection of design, technology, and visual heritage. what happens when AI becomes a dominant producer of visual culture, but today’s generative AI systems struggle to produce images that include Persian script especially Nastaliq calligraphy. I investigate why current AI image generators routinely produce “Nastaliq-like” textures that are visually persuasive yet linguistically incorrect, and what culturally responsible training resources could change that issue.

This matters because generative AI is rapidly becoming an everyday creative infrastructure, shaping what people can easily make, teach, publish, and preserve. In that environment, what AI cannot reliably “see” or generate risks becoming less visible in the digital future. When models are trained on datasets dominated by Latin scripts and English-language metadata, non-Latin visual traditions can become underrepresented—or effectively invisible, at scale, quietly pushing Persian calligraphy toward marginalization in education, media, and contemporary creative practice.

That risk is especially acute for Nastaliq. Nastaliq is a disciplined calligraphic system learned through years of rigorous training, guided by precise aesthetic rules and culturally specific conventions. Across centuries, it has deeply interwoven with Persian poetry, literature, architecture, and visual arts, it carries meaning as well as form, functioning as a living expression of Iranian cultural identity. While related scripts across the region have their own rich histories, Nastaliq is rooted specifically in Iran’s literary and artistic heritage and is inseparable from Persian language, memory, and cultural continuity.

If AI becomes the default “design assistant,” monocultural training won’t only produce technical mistakes; it will narrow what is considered legible, teachable, and worth preserving, reinforcing a single cultural default and placing culturally specific forms like Nastaliq at risk of digital erasure.To address this, I start from the premise that Nastaliq is both tangible and intangible heritage: a documented calligraphic system with rules and masters, a living embodied practice transmitted through teaching, and a national symbol invoked in cultural policy. For a practical process, I combine heritage methods with technology by auditing AI outputs to identify recurring failure patterns (connectivity, contextual forms, spacing, directionality, legibility), then translating those observations into culturally grounded requirements for data, metadata, and evaluation. I work with Nastaliq calligraphy masters in Iran to support consent, attribution, and protection of rights, treating training data not as “free raw material,” but as governed cultural knowledge.

Technically, I’m investigating how AI systems learn from and generate Nastaliq, with the goal of improving training resources, datasets, and culturally informed guidelines for image generation. I compare how diffusion models and GAN/style-transfer approaches behave with calligraphic scripts: diffusion systems often follow prompts well but struggle to form correct Persian characters, while GAN-based methods can learn stylistic coherence directly from image examples. I also examine the ethical pressure points of web-scraped datasets used in training and place this problem inside existing preservation frameworks such as UNESCO’s digital heritage thinking and the warning that AI can threaten cultural diversity through homogenization.

The outcome I’m building (still in process) is a culturally responsible toolkit: a workflow for dataset development and governance, guidelines for training or fine-tuning models with Nastaliq, and a practical evaluation rubric that distinguishes “looks like calligraphy” from readable, rule-faithful Nastaliq script. Success looks like measurable improvements in legibility and script accuracy, plus cultural integrity indicators, useful not only for designers, but also for museums, educators, and heritage institutions working with digitized manuscripts and living calligraphic practice.

Ultimately, this research aims to strengthen an AI ecosystem that is globally informed rather than culturally narrow, so generative tools support cultural continuity, expand visual diversity, and keep Persian typographic heritage living, legible, and creatively generative in the digital future, instead of drifting toward monoculture and a kind of “AI image-generation colonization.”