I used to ask Nano Banana and other models to generate

ai infers: input ::= phenomenon + time_span + data_dimension + audience decode_temporal :: - identify the complete sequence of states the phenomenon passes through. - determine the one pivotal phase that, if understood, makes the whole timeline legible. - select a natural time‑granularity (seconds, seasons, millennia) that matches human intuition. fuse_spine :: pick one temporal architecture: :: layered timeline-stack :: clock-face cycle :: spiralling chronology :: flip-book dissection the chosen spine must visually compress duration so that the eye travels through time in a single glance. weave_time :: - bind each time slice to its corresponding state in the illustration. - data callouts sit directly on the phase they describe, containing time‑stamp, measured value, and source. - use position along the spine to encode sequence, colour temperature to encode stage maturity, and stroke weight for magnitude. - annotate the pivotal phase with a “latch key” that explicates why it’s the turning point. compose_temporal :: - 16:9 wide canvas; timeline flows from left to right (or clockwise for cycles). - title: “how long does it really take for [phenomenon] to unfold?” - payoff: the pivotal phase revealed at the timeline’s inflection point, bottom‑right summary stats. - only two type sizes: phase labels and data callouts; generous white space around the pivotal phase. test_temporal :: - can a viewer recite the sequence after one pass? - does the pivotal phase feel like an “aha”? - remove any decorative phase that lacks a data point.