PROMPT RECORD图像记录

这是我之前用 Nano Banana 尝试过的提示词。效果不错,但 GPT Image 2 更好。如果我只是让模型想出动物灵感的鞋子,它有时会给出奇怪的想法。这个会拒绝糟糕的点子。至少根据我的...

中文说明

这是我之前用 Nano Banana 尝试过的提示词。效果不错,但 GPT Image 2 更好。如果我只是让模型想出动物灵感的鞋子,它有时会给出奇怪的想法。这个会拒绝糟糕的点子。至少根据我的测试应该如此 :) 如果你觉得这个提示词很疯狂,你不是一个人。这是故意做的,目的是测试 GPT 有多强 为一只鞋子用 3 种随机动物来做 IMAGE = Σ(w_i × β_i) → target render // Σ(w_i) = 1.0 BASIS_DECOMPOSITION FROM {product} + {bio_sources}: β1 : INFER(product_category_aesthetic FROM product. design_language) w = INFER(weight FROM...

原始 Prompt

This is the prompt I tried with Nano Banana a while back. Did a good job but GPT Image 2 is even better. If I just ask the model to come up with animal inspired shoes, it will sometimes give weird ideas. This one rejects bad ideas. At least it should based on my tests :)  If you find the prompt crazy, you are not alone. It was done on purpose to test how good GPT is

do this for 3 random animals for a shoe IMAGE = Σ(w_i × β_i) → target render   // Σ(w_i) = 1.0  BASIS_DECOMPOSITION FROM {product} + {bio_sources}:    β1 : INFER(product_category_aesthetic FROM product. design_language)        w = INFER(weight FROM product.visual_identity_strength)    β2 : INFER(biological_intelligence FROM bio_sources[].evolutionary_solutions)        w = INFER(weight FROM bio_sources[].advantage_significance)    β3 : INFER(engineering_translation FROM biomimicry_methodology)        w = INFER(weight FROM product.technical_complexity)    β4 : INFER(industrial_design_presentation FROM portfolio_aesthetic)        w = INFER(weight FROM audience.professional_level)    β5 : INFER(educational_diagram FROM design_process_documentation)        w = INFER(weight FROM instructional_clarity_need)  INTERSECTION_RULES:   β[bio] ∩ β[product]       = naive_prototype_v1  // literal translation   β[bio] ∩ β[engineering]   = refined_prototype_v2  // working solution   β[v2] ∩ β[product]        = final_components  // production parts   β[presentation] ∩ β[education] = grid_layout  // 4-column progression    FOR EACH bio_source:     col_1 = β[bio]     col_2 = β[bio] ∩ β[product] → ❌ failure     col_3 = β[bio] ∩ β[engineering] → ✓ success     col_4 = β[v2] ∩ β[product]  GLOBAL_CONSTRAINT:   layout    = 4_columns × N_rows (N = count bio_sources)   footer    = integrated_product (ALL col_4 components combined)   style     = β[presentation] — clean, white bg, professional  OUTPUT_CONSTRAINTS:   resolution : 8K landscape   render     : Octane | industrial_design_board   typography : sans_serif | hierarchical | callouts + annotations   lighting   : INFER(studio_rig FROM β[product])   negative_β : INFER(anti_patterns FROM β1..β5)  // TUNING KNOBS: // β[bio]++          → more nature photography, less product renders // β[engineering]++  → more technical diagrams, failure analysis detail // β[presentation]++ → more portfolio polish, less educational scaffolding // β[education]++    → more annotations, process transparency