我喜欢这个，不过不一定喜欢 GPT 的输出。给 GPT Image 2.5 的可食用寿司立体模型提示词。

I like this one though not necessarily GPT's output. edible sushi diorama prompt for GPT Image 2.5. 2x2 grid, 16:9, do this for 4 days that made humans proud: -- INPUT: $ SUBJECT CREATE TEMP VIEW subject_analysis AS SELECT INFER_MAIN_SILHOUETTE(:SUBJECT) AS silhouette, INFER_INTERNAL_ZONES(:SUBJECT) AS zones, INFER_TINY_ACTORS(:SUBJECT) AS actors, INFER_SYMBOLIC_PROPS(:SUBJECT) AS props, INFER_MOVEMENT_PATHS(:SUBJECT) AS routes; CREATE TEMP VIEW plate_base AS SELECT BUILD_SERVING_STAGE( plate = 'matte black ceramic plate', ground = 'sushi rice terrain with sesame scatter', lighting = 'premium food photography, dark background, soft overhead highlights' ) AS plate FROM subject_analysis; CREATE TEMP VIEW edible_structure AS SELECT CONVERT_TO_SUSHI_ARCHITECTURE( silhouette, material_map = ARRAY[ 'pressed rice blocks for walls and mass', 'salmon and tuna sashimi for smooth surfaces', 'tamago for roofs, panels, platforms', 'nori strips for beams, supports, outlines, flags', 'cucumber for rails, wheels, columns, arcs', 'roe and sesame for small lights, stones, crowds, particles' ], construction = 'precise knife-cut edible miniature' ) AS sushi_subject FROM subject_analysis; CREATE TEMP VIEW cutaway_world AS SELECT BUILD_EDIBLE_ROOMS( zones, style = 'tiny isometric dollhouse spaces made of food', actors = 'small rice-and-nori figures', props = 'food-based mini props inferred from subject' ) AS rooms, ADD_ROUTE_SYSTEM( routes, material = 'thin cucumber paths, nori rails, sauce lines, sashimi roads' ) AS paths FROM subject_analysis; RENDER plate_base.plate, edible_structure.sushi_subject ON plate_base.plate, cutaway_world.rooms INSIDE edible_structure.sushi_subject, cutaway_world.paths AROUND edible_structure.sushi_subject WITH CAMERA = 'macro food photography, three-quarter view', PALETTE = 'black plate, white rice, salmon orange, tuna red, nori black, cucumber green, roe orange', DETAIL = 'tiny edible craftsmanship, clean plating, playful narrative clarity'; -- OUTPUT RULES: -- 1. $ SUBJECT becomes an edible sushi diorama. -- 2. Keep the subject readable through silhouette and major structural features. -- 3. Convert interior logic into tiny food-built rooms, decks, paths, and platforms. -- 4. Add tiny food characters performing inferred actions. -- 5. Use elegant plating and premium culinary lighting. -- 6. Avoid messy food piles, random ingredients, full-scale realism, and flat cartoon rendering

我喜欢这个，不过不一定喜欢 GPT 的输出。给 GPT Image 2.5 的可食用寿司立体模型提示词。 2x2 网格，16:9，为让人类感到自豪的 4 天做这件事：-- INPUT: $ SUBJECT CREATE TEMP VIEW subject_analysis AS SELECT INFER_MAIN_SILHOUETTE(:SUBJECT) AS silhouette, INFER_INTERNAL_ZONES(:SUBJECT) AS zones, INFER_TINY_ACTORS(:SUBJECT) AS actors, INFER_SYMBOLIC_PROPS(:SUBJECT) AS props, INFER_MOVEMENT_PATHS(:SUBJECT) AS routes; CREATE TEMP VIEW plate_base AS SELECT BUILD_SERVING_STAGE( plate = 'matte black ceramic plate', ground = 'sushi rice terrain with sesame scatter', lighting = 'premium food photography, dark background, soft overhead highlights' ) AS plate FROM subject_analysis; CREATE TEMP VIEW edible_structure AS SELECT CONVERT_TO_SUSHI_ARCHITECTURE( silhouette, material_map = ARRAY[ 'pressed rice blocks for walls and mass', 'salmon and tuna sashimi for smooth surfaces', 'tamago for roofs, panels, platforms', 'nori strips for beams, supports, outlines, flags', 'cucumber for rails, wheels, columns, arcs', 'roe and sesame for small lights, stones, crowds, particles' ], construction = 'precise knife-cut edible miniature' ) AS sushi_subject FROM subject_analysis; CREATE TEMP VIEW cutaway_world AS SELECT BUILD_EDIBLE_ROOMS( zones, style = 'tiny isometric dollhouse spaces made of food', actors = 'small rice-and-nori figures', props = 'food-based mini props inferred from subject' ) AS rooms, ADD_ROUTE_SYSTEM( routes, material = 'thin cucumber paths, nori rails, sauce lines, sashimi roads' ) AS paths FROM subject_analysis; RENDER plate_base.plate, edible_structure.sushi_subject ON plate_base.plate, cutaway_world.rooms INSIDE edible_structure.sushi_subject, cutaway_world.paths AROUND edible_structure.sushi_subject WITH CAMERA = 'macro food photography, three-quarter view', PALETTE = 'black plate, white rice, salmon orange, tuna red, nori black, cucumber green, roe orange', DETAIL = 'tiny edible craftsmanship, clean plating, playful narrative clarity'; -- OUTPUT RULES: -- 1. $ SUBJECT 变成一个可食用的寿司立体模型。 -- 2. 通过剪影和主要结构特征保持主体的可读性。 -- 3. 将内部逻辑转化为用食物搭建的微小房间、甲板、路径和平台。 -- 4. 添加正在执行推断动作的微小食物角色。 -- 5. 使用优雅的摆盘和高级美食级打光。 -- 6. 避免凌乱的食物堆、随机食材、全尺寸写实和扁平的卡通渲染。