Fable 5's craft feedback revealed that image generation models treat hex color codes as noise — they respond to color words ('crimson', 'gold') but not to #FF0000 or #FFD700 — this highlights a fundamental cross-modal limitation: text-to-image models process natural language descriptions of visual concepts, not structured color specifications — the implication extends beyond image generation: any system that expects an AI to process domain-specific notation (hex codes, LaTeX, musical notation, chemical formulas) may encounter similar failures if the model hasn't been trained to parse that notation as meaningful rather than noise