Another fun prompt for visualizing abstract concepts. GPT 2.5 handled it before than Nano Banana
2x2 grid, 16:9, Ai picks topics INPUT ::=
STEP_1 :: infer the physics of the idea
- how the concept behaves: accumulates, leaks, erodes, spreads, compresses
- its rate, its pressure, its direction
- what it does to anyone standing near it
STEP_2 :: infer the assigned phenomenon [PHYSICAL ASSIGNMENT]
- choose the one natural or mechanical process that matches the behaviour:
debt as rising tide, attention as a leaking reservoir, trust as a
thawing ice shelf, bureaucracy as sediment
- the match must be behavioural, not just visual
- assign the concept one material, one force, one rate
STEP_3 :: infer the human scale
- one ordinary person adapting to it: wading, bailing, building higher
- their response shows how long it has been happening
- their expression reads as routine, never panic
STEP_4 :: infer the evidence trail
- high-water marks, tide lines, repairs at successive heights
- the history of the problem written into the architecture around it
- one marker showing where the level used to be
STEP_5 :: render spec
- calm, observational realism, long lens, overcast light
- forbidden: symbols painted on the phenomenon, labels, text, a second metaphor
- one honest detail: the oldest repair, now far below the current level
- must pass the one-sentence test with a single verb
EXAMPLE_INPUT :: [notification overload]
A single calm observational photograph that turns the abstract concept of notification overload into a physical phenomenon: a slow rising tide of small glowing white paper slips flooding an ordinary apartment street. One ordinary person in a raincoat wades through it with routine calm, carrying a bucket and bailing slips out of a doorway; the walls show successive high-water marks and repairs at different heights, and one old repair sits far below the current level. Long lens, overcast light, quiet realism. No symbols painted on the phenomenon, no labels, no text, no second metaphor. Square 1:1.
cinel 为生成示例图做了以上替换。
