Tamil Ambigram generation by a Diffusion Model

Abstract

This research proposes a novel approach to generating ambigrams in Tamil, a non-Latin script, using diffusion models. Ambigrams are designs where a word remains visually identical even after being rotated by 180 degrees, but creating them can be a complex and time-consuming task, especially for non-Latin scripts. Previous research has focused on generating ambigrams for Latin scripts, leaving a gap in the field for non-Latin languages. This study addresses that gap by developing a Tamil ambigram generator, leveraging diffusion models to automate the process without relying on existing ambigram datasets. Tamil was chosen for this project due to its significance as the author's first language and the time constraints of an undergraduate final-year project. The goal was to create ambigrams that strike a balance between aesthetic appeal and legibility, ensuring that the generated designs maintain readability when rotated. The proposed system was evaluated using metrics adapted from similar studies on ambigram generation with diffusion models. Results show that the Tamil ambigram generator is capable of producing stylistic designs that are visually appealing while preserving a reasonable level of readability. This research contributes to the growing field of AI-driven design in non-Latin scripts, offering a solution for automating the generation of ambigrams in Tamil, with potential applications in graphic design, typography, and other visual arts.