Dad builds Greek alphabet cards where each object mimics its letter's shape

A parent raising trilingual kids in China designed a deck of Greek alphabet flashcards in which each illustrated object both starts with and visually resembles its letter — a known mnemonic technique for English but apparently a first for Greek. The pipeline was unusually data-driven for a craft project: he pulled the GreekLex corpus of 35,304 modern Greek words, filtered to 3–10 character words with a corpus frequency of at least 100, then batched the survivors through ChatGPT asking which referents could be drawn to echo each letter’s shape. Promising candidates went to OpenAI’s gpt-image-1.5, with the target letter attached as a reference image to keep the shape in mind during generation.

The model handled most letters but balked at some pairings — a phi-shaped snake required a hand drawing before the model would re-render it in style. Output is illustrated in an Eric Carle aesthetic across two decks (object cards and plain letter cards) used for memory matching and a physical ‘fire game’ the kids ask to replay. Omega had essentially no viable matches in the corpus, illustrating the limits of frequency-filtered candidate generation.