耗時100小時掌握ChatGPT圖像生成器，這是我學到的一切

result = client.images.edit(
    model="gpt-image-1.5",
    input_fidelity="high",  # 身份保留的秘訣
    quality="high",
    image=[open("portrait.png", "rb")],
    prompt="Change the background to a sunset beach while preserving the person's exact appearance"
)

這種組合確保在應用你請求的更改時最大程度地保留原始主體。

Goal/Output (目標/輸出):
- [Type of image: ad, UI mockup, infographic, photo, illustration] (圖像類型：廣告、UI模型、信息圖、照片、插圖)
- [Intended use and audience] (預期用途和受眾)

Scene (場景):
- [Background/environment description] (背景/環境描述)
- [Main subject with specific details] (主體及具體細節)
- [Action or relationship between elements] (動作或元素間的關係)

Style (風格):
- [Medium: photograph, watercolor, 3D render, vector illustration] (媒介：照片、水彩、3D渲染、矢量插圖)
- [Key textures: matte, glossy, grainy, smooth, organic] (關鍵紋理：啞光、光澤、顆粒感、平滑、有機)
- [Quality descriptors: realistic imperfections, stylized, minimalist] (質量描述：逼真的瑕疵、風格化、極簡主義)

Composition/Layout (構圖/布局):
- [Camera position: close-up, wide shot, aerial view, eye-level] (相機位置：特寫、廣角、鳥瞰、平視)
- [Lighting: golden hour, studio strobes, overcast, dramatic shadows] (光照：黃金時刻、攝影棚閃光燈、陰天、戲劇性陰影)
- [Element placement: centered, rule of thirds, negative space, margins] (元素放置：居中、三分法、留白、邊距)

Text (if any) (文本，如果有):
- "Exact text in quotes" ("引號中的確切文本")
- [Font style, size, color, position] (字體風格、大小、顏色、位置)
- [Specify: render only once, no duplicates] (指定：僅渲染一次，無重複)

Constraints (約束):
- Change ONLY: [specific element if editing] (僅更改：[特定元素，如果是編輯])
- Preserve exactly: [elements that must stay unchanged] (完全保留：[必須保持不變的元素])
- Negative: no watermark, no extra text, no logos, no [unwanted elements] (負向：無水印、無額外文本、無Logo、無[不需要的元素])

這個框架為模型提供了它需要做出的每一個視覺決策的清晰上下文。

❌ BEFORE (模糊):
"A beautiful portrait of an old fisherman, very detailed, high quality, realistic"
(一個老漁夫的美麗肖像，非常詳細，高質量，逼真)

✅ AFTER (攝影師思維):
"Candid documentary photograph of an elderly fisherman on a weathered wooden boat.
Weathered face with visible wrinkles, sun spots, and pores. Deep-set kind eyes.
Gray stubble. Faded traditional anchor tattoo on forearm. Salt-stained navy wool
sweater, worn cap.

Early morning coastal light, soft fog diffusing the sun. Medium close-up at eye
level, 50mm lens, f/2.8, shallow depth of field. Shot like 35mm film with subtle
grain, natural color balance.

Documentary style — honest, unretouched, capturing a real moment. No glamorization."
(飽經風霜的木船上的老漁夫的紀實抓拍。飽經風霜的臉，可見皺紋、曬斑和毛孔。深陷的慈祥眼睛。灰色鬍渣。前臂上褪色的傳統錨紋身。鹽漬的海軍藍羊毛衫，磨損的帽子。
清晨的海岸光線，柔和的霧氣漫射著陽光。視線水平的中特寫，50mm鏡頭，f/2.8，淺景深。像35mm膠片一樣拍攝，有微妙的顆粒感，自然色彩平衡。
紀實風格 —— 誠實，未修飾，捕捉真實瞬間。沒有美化。)

攝影師思維將模糊的願望轉化為模型能夠深度解碼的精確視覺規格。

Create a photorealistic candid photograph of an elderly sailor standing on a small fishing boat.
(創建一張老水手站在小漁船上的照片級真實抓拍。)

Subject: Weathered face with visible wrinkles, sun spots, and pores. Deep-set kind
eyes with crow's feet. Gray stubble, a few days unshaven. Faded traditional anchor
tattoo on forearm. Salt-stained navy wool sweater, worn and pilled. Creased cap
with faded insignia.
(主體：飽經風霜的臉，可見皺紋、曬斑和毛孔。深陷的慈祥眼睛，有魚尾紋。灰色鬍渣，幾天沒刮。前臂上褪色的傳統錨紋身。鹽漬的海軍藍羊毛衫，磨損起球。折痕帽子，褪色徽章。)

Setting: Early morning on the water, soft coastal fog diffusing the light. Aged
wooden boat deck with peeling paint, fishing nets in background, coiled rope.
(環境：清晨的水面，柔和的海岸霧氣漫射著光線。陳舊的木船甲板，油漆剝落，背景中有漁網，盤繞的繩索。)

Technical: Shot like 35mm film photography, medium close-up at eye level, 50mm
lens, shallow depth of field with boat blurred behind him. Subtle film grain,
natural color balance without heavy grading.
(技術：像35mm膠片攝影一樣拍攝，視線水平的中特寫，50mm鏡頭，淺景深，身後船隻模糊。微妙的膠片顆粒，自然色彩平衡，無重度調色。)

The image should feel like a real moment captured by a photojournalist — honest,
unposed, with real skin texture, worn materials, and everyday imperfection. No
glamorization, no heavy retouching, no artificial perfection.
(圖像應該感覺像是攝影記者捕捉到的真實瞬間 —— 誠實、未擺拍，有真實的皮膚紋理、磨損的材料和日常的瑕疵。沒有美化，沒有重度修飾，沒有人工完美。)

注意我們要如何明確要求瑕疵 —— 飽經風霜的皮膚、磨損的材料、剝落的油漆。現實是有紋理的。

Create a detailed infographic explaining how a coffee machine works.
(創建一個詳細的信息圖，解釋咖啡機是如何工作的。)

Structure (結構):
- Title at top: "The Journey of Your Morning Coffee" (頂部標題："你晨間咖啡的旅程")
- Vertical flow diagram showing: bean hopper → grinder → portafilter →
  grouphead → water heating → extraction → cup
  (垂直流程圖顯示：豆倉 → 研磨機 → 手柄 → 沖煮頭 → 水加熱 → 萃取 → 杯子)
- Each step has an icon and 1-2 sentence explanation (每一步都有一個圖標和1-2句話的解釋)
- Warm color palette (browns, creams, copper accents) (暖色調：棕色、奶油色、銅色點綴)
- Clean, modern design with plenty of white space (乾淨、現代的設計，大量留白)
- Subtle coffee stain texture in background corners (背景角落有微妙的咖啡漬紋理)

Style (風格): Professional print-quality infographic, vector-style icons, clear
hierarchy, readable at A4 size. (專業印刷質量信息圖，矢量風格圖標，清晰層級，A4尺寸可讀。)

Typography (排版): Clean sans-serif headings, readable body text, clear visual
hierarchy between title, section headers, and explanatory text. (乾淨的無襯線標題，可讀的正文，標題、章節頭和解釋性文本之間有清晰的視覺層級。)

No watermarks. No stock photo elements. Original illustration only. (無水印。無圖庫照片元素。僅原創插圖。)

對於密集的文本和複雜的布局，始終使用 quality="high" 以確保文本保持清晰可讀。

Create an original logo for "Field & Flour" — a local artisan bakery.
(為"Field & Flour" —— 一家當地手工麵包店創建一個原創Logo。)

Brand personality: Warm, authentic, handcrafted, timeless. Not trendy or corporate.
(品牌個性：溫暖、真實、手工製作、永恆。不追逐潮流或企業化。)

Design requirements (設計要求):
- Clean vector-style shapes with strong silhouette (乾淨的矢量風格形狀，輪廓強)
- Balanced negative space (平衡的負空間)
- Must read clearly from 16px favicon to large signage (從16px網站圖標到大型標牌都必須清晰可讀)
- Flat design, minimal strokes, no gradients unless essential (扁平化設計，極簡線條，除非必要否則無漸變)
- Earth-tone palette: warm wheat gold, deep brown, cream (大地色調：暖麥金、深棕、奶油色)
- Could incorporate subtle wheat or grain element (可以融入微妙的小麥或穀物元素)
- Text must be perfectly legible and properly kerned (文本必須完全清晰且字距適當)

Output: Single centered logo on plain cream background. Generous padding around
the design for flexibility. (輸出：純奶油色背景上的單個居中Logo。設計周圍有寬敞的留白以增加靈活性。)

No watermarks, no mockups, no 3D effects, no complex imagery. Simple, functional,
timeless design. (無水印，無樣機，無3D效果，無複雜圖像。簡單、實用、永恆的設計。)

使用 n=4 生成多個變體。Logo設計是主觀的 —— 給自己多種選擇。

Create a realistic mobile app UI mockup for a local farmers market app.
(為一個當地農貿市場App創建一個逼真的移動App UI模型。)

Screen content (from top) (屏幕內容（從上到下）):
- Simple header with market name "Riverside Market" and search icon (簡單的頭部，帶有市場名稱"Riverside Market"和搜索圖標)
- Today's featured vendor carousel with square photos (今日精選商家輪播，帶方形照片)
- "Fresh Today" section with produce category chips (Vegetables, Fruits, Dairy, Baked) ("今日新鮮"部分，帶農產品類別標籤（蔬菜、水果、乳製品、烘焙）)
- Vendor list with small photos, names, specialties, and distance (商家列表，帶小照片、名稱、特色和距離)
- Bottom navigation: Home, Map, Favorites, Cart, Profile (底部導航：首頁、地圖、收藏、購物車、個人資料)

Design language (設計語言):
- White background, subtle natural green accents (白色背景，微妙的自然綠色點綴)
- Clear typography hierarchy (system fonts feel) (清晰的排版層級（系統字體感覺）)
- Generous padding and touch-friendly targets (寬敞的填充和觸摸友好的目標)
- Looks like a real shipped product, not a concept (看起來像真正的已發布產品，而不是概念)
- Uses realistic vendor names and produce photos (使用真實的商家名稱和農產品照片)

Frame: Place the UI inside an iPhone 15 Pro device frame, slight perspective
tilt, subtle shadow beneath. (框架：將UI放置在iPhone 15 Pro設備框架內，輕微透視傾斜，下方有微妙陰影。)

專注於布局、層級、間距和逼真的介面元素。避免概念性或藝術性的語言。

Create a 4-panel vertical comic strip. Equal panel sizes, clear panel borders.
(創建一個4格垂直連環畫。格子大小相等，清晰的邊框。)

Panel 1: Pet owner walks out the front door, keys in hand. Through the window
behind them, we see their cat watching — paws pressed against glass, eyes wide
with apparent sadness. The house suddenly feels empty.
(第一格：寵物主人走出前門，手裡拿著鑰匙。透過身後的窗戶，我們看到他們的貓在看 —— 爪子按在玻璃上，眼睛睜得大大的，明顯很悲傷。房子突然感覺空蕩蕩的。)

Panel 2: The door clicks shut. The cat slowly turns away from the window toward
the empty house. Its posture shifts from forlorn to interested. Eyes narrow with
possibility.
(第二格：門關上了。貓慢慢從窗戶轉向空蕩蕩的房子。它的姿態從淒涼轉變為感興趣。眼睛眯起來，充滿了可能性。)

Panel 3: Total chaos. Cat sprawled across the forbidden couch like royalty.
Knocked over plant on the floor. Papers scattered. Sunbeam spotlighting the
scene of domestic crime.
(第三格：完全混亂。貓像皇室一樣躺在禁止入內的沙發上。地板上是被打翻的植物。紙張散落。陽光照亮了家庭犯罪現場。)

Panel 4: Door handle turns. Cat sits perfectly upright by the entrance,
composed and innocent, tail wrapped neatly around paws. Not a hair out of
place. As if nothing happened.
(第四格：門把手轉動。貓在入口處坐得筆直，鎮定而無辜，尾巴整齊地纏繞在爪子上。一絲不亂。彷彿什麼都沒發生過。)

Style: Warm illustrated style with expressive characters, clear visual
storytelling that reads without text. Consistent character design across
all panels.
(風格：溫暖的插圖風格，富有表現力的角色，清晰的視覺敘事，無需文字即可閱讀。所有格子中角色設計一致。)

No speech bubbles or text. Let the visuals tell the story. (無氣泡或文字。讓視覺講述故事。)

將每一格定義為具有清晰動作的獨特視覺節拍。模型處理格子布局和視覺連續性。

Create a children's book illustration introducing the main character.
(創建一個介紹主角的兒童讀物插圖。)

Character: Young forest hero, around 8 years old. (角色：年輕的森林英雄，約8歲。)
- Green hooded tunic (think woodland adventurer, not Robin Hood) (綠色連帽長袍（想想林地冒險家，不是羅賓漢）)
- Soft brown boots, well-worn (柔軟的棕色靴子，磨損良好)
- Small belt pouch for collecting treasures (用於收集寶藏的小腰包)
- Carries a tiny wooden bow (symbolic, for helping not hurting) (背著一把小木弓（象徵性的，用於幫助而非傷害）)
- Kind expression, bright curious eyes, brave but gentle demeanor (善良的表情，明亮好奇的眼睛，勇敢但溫和的舉止)
- Slightly oversized head for picture book proportions (略大的頭，符合圖畫書比例)

Theme: This character protects and rescues small forest animals in trouble.
(主題：這個角色保護和拯救陷入困境的小森林動物。)

Style: Hand-painted watercolor look with soft outlines, warm earthy palette
with forest greens and autumn oranges. Whimsical, friendly, inviting for
young readers ages 4-8.
(風格：手繪水彩外觀，柔和輪廓，溫暖的大地色調，森林綠和秋季橙。異想天開，友好，吸引4-8歲的年輕讀者。)

Composition: Character standing in simple forest glade, dappled sunlight,
leaving room for title text above. Character clearly showcased.
(構圖：角色站在簡單的林間空地上，斑駁的陽光，上方留有標題文字空間。角色清晰展示。)

Original character design only. No text. No watermarks. No copyrighted
character references. (僅原創角色設計。無文字。無水印。無版權角色參考。)

保存此角色參考圖像 —— 你將使用它在後續插圖中保持一致性。

Create a realistic outdoor crowd scene in Bethel, New York on August 16, 1969.
(創建一個1969年8月16日紐約伯特利的逼真戶外人群場景。)

Photorealistic, period-accurate clothing, staging, and environment.
(照片級真實，符合時代的服裝、布景和環境。)

Documentary photography style, shot on film, natural lighting.
(紀實攝影風格，膠片拍攝，自然光。)

模型知道這是伍德斯托克音樂節，而無需被告知。它僅根據日期和地點就能生成嬉皮士、那個時代的時尚、節日氣氛。

Edit the image to dress this person in the provided clothing items.
(編輯圖像，讓此人穿上提供的服裝項目。)

MUST PRESERVE (do not change in any way) (必須保留（不要以任何方式更改）):
- Face, facial features, expression, skin tone (臉、面部特徵、表情、膚色)
- Body shape, proportions, and pose (體型、比例和姿勢)
- Hairstyle and hair color (髮型和髮色)
- Background and environment (背景和環境)
- Camera angle, framing, and composition (相機角度、取景和構圖)
- Overall lighting direction and quality (整體光照方向和質量)

CHANGE ONLY (僅更改):
- Replace current clothing with provided garment images (用提供的服裝圖像替換當前服裝)
- Fit garments naturally to body geometry (使服裝自然貼合身體幾何形狀)
- Show realistic fabric draping, folds, and behavior (顯示逼真的織物垂墜、褶皺和表現)
- Match lighting and shadows on fabric to original photo (使織物上的光照和陰影與原始照片匹配)

REQUIREMENTS (要求):
- Photorealistic integration — outfit should look worn, not pasted (照片級真實融合 —— 服裝應該看起來是穿在身上的，而不是粘貼上去的)
- Maintain color temperature of original image (保持原始圖像的色溫)
- No accessories, text, logos, or watermarks added (不添加配飾、文字、Logo或水印)
- Identity must remain clearly recognizable (身份必須保持清晰可辨)

對於虛擬試穿，始終使用 input_fidelity="high" 以確保面部相似度得到保持。

Using the EXACT visual style of the reference image (Image 1), create:
(使用參考圖像（圖像1）的完全相同的視覺風格，創建：)
A man riding a motorcycle on a winding mountain road.
(一個男人在蜿蜒的山路上騎摩托車。)

STYLE ELEMENTS TO MATCH PRECISELY from reference (從參考中精確匹配的風格元素):
- Color palette and saturation levels (調色板和飽和度水平)
- Line quality and weight (線條質量和粗細)
- Texture treatment and brushwork (紋理處理和筆觸)
- Lighting style and direction (光照風格和方向)
- Level of detail vs. abstraction (細節與抽象的程度)
- Overall artistic aesthetic (整體藝術審美)

APPLY TO NEW CONTENT (應用到新內容):
- Single subject (man on motorcycle) (單一主體（騎摩托車的人）)
- Clear composition with visual interest (清晰的構圖，具有視覺趣味)
- Mountain road environment with curves (帶有彎道的山路環境)
- Sense of motion and freedom (運動感和自由感)

The new image should look like it came from the same artist or series as
the reference. Maintain stylistic consistency exactly.
(新圖像應該看起來像與參考圖像出自同一位藝術家或系列。精確保持風格一致性。)

當你具體說明要保留哪些風格元素以及要更改哪些內容元素時，風格遷移效果最好。

In this room photo, replace ONLY the white plastic chairs with
mid-century modern wooden chairs (walnut finish, tapered legs,
woven seat).
(在這個房間照片中，僅將白色塑料椅子替換為世紀中期現代風格的木椅（胡桃木飾面，錐形腿，編織座椅）。)

PRESERVE COMPLETELY (完全保留):
- Camera angle and perspective (相機角度和透視)
- Room lighting direction and quality (房間光照方向和質量)
- All other furniture and objects (所有其他家具和物體)
- Wall colors and decorations (牆壁顏色和裝飾)
- Floor material and shadows (地板材質和陰影)
- Overall image quality and color grading (整體圖像質量和調色)

INTEGRATION REQUIREMENTS (融合要求):
- Chairs must match room's perspective exactly (椅子必須精確匹配房間的透視)
- Wood grain should catch existing light realistically (木紋應逼真地捕捉現有光線)
- Contact shadows must be natural and match light source (接觸陰影必須自然並與光源匹配)
- Scale must be accurate relative to table height (比例必須相對於桌子高度準確)
- New chairs should look like they belong in this room (新椅子應該看起來屬於這個房間)

Photorealistic result — should look like the original photograph.
(照片級真實結果 —— 應該看起來像原始照片。)

室內設計可視化是最具商業價值的編輯應用之一。

Transform this hand-drawn sketch into a photorealistic image.
(將這張手繪草圖轉化為照片級真實的圖像。)

PRESERVE FROM SKETCH (從草圖中保留):
- Exact layout and proportions (精確的布局和比例)
- Perspective and viewing angle (透視和視角)
- Element placement and relationships (元素放置和關係)
- Implied depth and layering (暗示的深度和層次)

ADD FOR REALISM (為真實感添加):
- Appropriate real-world materials and textures (適當的現實世界材料和紋理)
- Consistent natural lighting (interpret from sketch shading) (一致的自然光照（從草圖陰影解讀）)
- Environmental context matching the implied setting (匹配暗示環境的環境背景)
- Surface imperfections and wear appropriate to materials (適合材料的表面瑕疵和磨損)

CONSTRAINTS (約束):
- Do not add new elements not present in sketch (不要添加草圖中不存在的新元素)
- Do not add text or watermarks (不要添加文字或水印)
- Treat the sketch as an architectural blueprint to follow exactly (將草圖視為要精確遵循的建築藍圖)
- Fill in realistic details while honoring the original composition (在尊重原始構圖的同時填充逼真的細節)

模型解讀草圖的意圖並填充逼真的細節，同時保持原始構圖。

Transform this daytime summer scene into a winter evening with snowfall.
(將這個白天的夏季場景轉換為下雪的冬夜。)

CHANGE (更改):
- Time of day: from afternoon to dusk (warm interior lights visible) (時間：從下午到黃昏（可見溫暖的室內燈光）)
- Season: summer to deep winter (季節：夏季到深冬)
- Weather: clear to active snowfall (天氣：晴朗到下雪)
- Ground: grass to fresh snow coverage (地面：草地到新雪覆蓋)
- Trees: summer foliage to bare branches with snow (樹木：夏季樹葉到帶雪的枯枝)
- Atmosphere: add visible breath if people present (氛圍：如果有人，添加可見的呼吸)
- Surfaces: add frost on windows and metal (表面：在窗戶和金屬上添加霜)

PRESERVE (保留):
- Camera position and angle exactly (相機位置和角度完全不變)
- All objects and their exact positions (所有物體及其確切位置)
- Architecture and structural elements (建築和結構元素)
- People and their poses (update clothing appropriately) (人和他們的姿勢（適當更新服裝）)
- Overall composition and framing (整體構圖和取景)

Style: Photorealistic, natural atmospheric perspective, visible
snowflakes in air, cozy contrast between warm interior lights and
cold exterior. Should feel photographed, not filtered.
(風格：照片級真實，自然大氣透視，空氣中可見雪花，溫暖室內燈光與寒冷室外之間的舒適對比。應該感覺是拍攝的，而不是濾鏡處理的。)

在環境轉換中使用 input_fidelity="high" 和 quality="high" 以獲得最佳結果。

I'm providing 2 images:
(我提供2張圖片：)
- Image 1: Beach scene with woman standing on shore at sunset (圖片1：日落時分女人站在岸邊的海灘場景)
- Image 2: Golden retriever sitting in a studio setting (圖片2：坐在攝影棚環境中的金毛尋回犬)

Task: Place the dog from Image 2 into the beach scene from Image 1,
positioned next to the woman, looking up at her.
(任務：將圖片2中的狗放入圖片1的海灘場景中，位於女人旁邊，抬頭看著她。)

MATCHING REQUIREMENTS (匹配要求):
- Dog's lighting must match beach sunset (warm golden light from left) (狗的光照必須匹配海灘日落（來自左側的溫暖金光）)
- Scale dog appropriately relative to woman's height (相對於女人的身高適當縮放狗)
- Dog should cast shadow consistent with scene's sun angle (狗應該投射與場景太陽角度一致的陰影)
- Sand texture should show around and under dog's paws (狗爪周圍和下方應顯示沙子紋理)
- Fur should catch the same golden hour highlights as scene (毛髮應捕捉與場景相同的黃金時刻高光)

PRESERVE FROM IMAGE 1 (從圖片1保留):
- Woman's exact appearance, position, and pose (女人的確切外貌、位置和姿勢)
- Beach background completely unchanged (海灘背景完全不變)
- Original photo's color grading and mood (原始照片的調色和情緒)

The composite should look like a single photograph taken on location.
No visible compositing artifacts.
(合成圖應該看起來像是在現場拍攝的單張照片。沒有可見的合成偽影。)

按編號引用圖片，並明確哪些元素轉移，哪些保持固定。

Translate all text in this infographic from English to Japanese.
(將此信息圖中的所有文本從英語翻譯成日語。)

MUST PRESERVE (必須保留):
- Exact layout, spacing, and positioning of all elements (所有元素的精確布局、間距和位置)
- All visual elements, icons, illustrations, and graphics (所有視覺元素、圖標、插圖和圖形)
- Typography hierarchy (headlines vs body text relationships) (排版層級（標題與正文關係）)
- Color scheme and overall design aesthetic (配色方案和整體設計審美)
- Font weights and relative sizes (字體粗細和相對大小)

TRANSLATION REQUIREMENTS (翻譯要求):
- Accurate Japanese translation with natural phrasing (準確的日語翻譯，措辭自然)
- Match visual weight and style to original fonts (匹配原始字體的視覺權重和風格)
- Adjust character spacing for Japanese typographic norms (針對日語排版規範調整字符間距)
- No text truncation or overflow outside original bounds (無文本截斷或溢出原始邊界)

Do not modify any non-text elements. Only change the language.
(不要修改任何非文本元素。只改變語言。)

此工作流無需從頭重建即可處理營銷材料、UI截圖、包裝和信息圖表。

Continue the children's book story using the character from the reference image.
(使用參考圖像中的角色繼續兒童讀物故事。)

New Scene (新場景):
The same young forest hero is gently helping a frightened squirrel out
of a fallen hollow tree after a winter storm. Snow on the ground, bare
branches above, warm light filtering through clouds.
(同一個年輕的森林英雄在一場冬季風暴後，正溫柔地幫助一隻受驚的松鼠從倒下的空心樹中出來。地面上有雪，頭頂是光禿禿的樹枝，溫暖的光線透過雲層過濾下來。)

CHARACTER CONSISTENCY (from reference) (角色一致性（來自參考）):
- Same green hooded tunic, exact shade and style (同樣的綠色連帽長袍，確切的色調和款式)
- Same soft brown boots (同樣的柔軟棕色靴子)
- Same belt pouch (同樣的腰包)
- Same facial features, proportions, and color palette (同樣的面部特徵、比例和調色板)
- Same gentle, heroic personality in expression (表情中同樣的溫和、英雄個性)
- Same children's book proportions (同樣的兒童讀物比例)

STYLE CONSISTENCY (from reference) (風格一致性（來自參考）):
- Same watercolor illustration style (同樣的水彩插圖風格)
- Same soft outlines (同樣的柔和輪廓)
- Same warm earthy color treatment (同樣的溫暖大地色處理)
- Same whimsical, friendly aesthetic (同樣的異想天開、友好審美)

New elements: winter forest environment, frightened squirrel, fallen
tree with hollow. (新元素：冬季森林環境，受驚的松鼠，倒下的空心樹。)

Do not redesign the character. Do not change the artistic style.
No text. No watermarks.
(不要重新設計角色。不要改變藝術風格。無文字。無水印。)

引用錨點圖像並重複關鍵角色細節，以在整本書中保持一致性。

Create a hyper-stylized 3D floating head portrait based on this person.
(根據此人創建一個超風格化的3D浮動頭像人像。)

STYLE CHARACTERISTICS (風格特徵):
- Smooth skin with glossy vinyl-finish surface (光滑皮膚，帶有光澤乙烯基表面)
- Strong highlighter on cheekbones and nose tip catching soft light (顴骨和鼻尖上的強高光捕捉柔和光線)
- Holographic, iridescent eyeshadow (purple to teal color shift) (全息、彩虹色眼影（紫色到青色變色）)
- Thick hair sculpted in slick, glossy waves like polished acrylic (像拋光亞克力一樣雕刻成光滑、光澤波浪的濃密頭髮)
- Small metallic chrome nose piercing with brushed reflections (帶有拉絲反射的小金屬鉻鼻釘)

EXPRESSION (表情):
Confident, slightly unimpressed look — half-lidded eyes, subtly
arched brow, the sophisticated "too cool" attitude.
(自信，略帶不屑的表情 —— 半垂的眼瞼，微妙拱起的眉毛，複雜的「太酷了」的態度。)

TECHNICAL SPECIFICATIONS (技術規格):
- Head floats isolated against plain white background (頭像孤立漂浮在純白背景前)
- Slight 15-degree tilt (premium product render feeling) (輕微15度傾斜（高級產品渲染感）)
- Bright, diffuse studio lighting with no harsh shadows (明亮、漫射的攝影棚燈光，無刺眼陰影)
- Emphasis on glossy, plastic, subsurface scattering effects (強調光澤、塑料、次表面散射效果)
- Ultra-smooth textures throughout (整體超平滑紋理)
- Close-up portrait angle, straight-on, 85mm lens feel (特寫人像角度，正視，85mm鏡頭感)

The result should look like a high-end 3D character render or
collectible figure — plastic perfection with personality.
(結果應該看起來像高端3D角色渲染或收藏人偶 —— 具有個性的塑料完美感。)

這種程度的審美細節能在不同主體間產生驚人一致的結果。

Transform this person into an adorable chibi-style character.
(將此人轉換為可愛的Q版風格角色。)

CHIBI PROPORTIONS (Q版比例):
- Tiny body (about 1 head-height tall) (微小的身體（約1個頭高）)
- Oversized head (3x body proportions) (超大的頭（3倍身體比例）)
- Large, sparkling eyes with cute highlights (大而閃亮的眼睛，帶有可愛的反光)
- Soft, rounded facial features (柔和、圓潤的面部特徵)
- Cheerful, expressive pose with personality (愉快、富有表現力的姿勢，具有個性)

PRESERVE FROM ORIGINAL (從原始圖像保留):
- Recognizable facial features (simplified but identifiable) (可辨認的面部特徵（簡化但可識別）)
- Hairstyle, length, and hair color (髮型、長度和髮色)
- Distinctive clothing style or accessories (獨特的服裝風格或配飾)
- Any notable characteristics (glasses, jewelry, etc.) (任何顯著特徵（眼鏡、珠寶等）)
- Overall personality and vibe (整體個性和氛圍)

STYLE (風格):
- Smooth pastel shading (平滑的柔和陰影)
- Clean lines and simplified details (乾淨的線條和簡化的細節)
- Bright, expressive colors (明亮、富有表現力的顏色)
- Collectible figure aesthetic (收藏人偶審美)

Background: Simple gradient or plain color to showcase character.
(背景：簡單的漸變或純色以展示角色。)

The result should feel like an irresistible chibi mascot that
clearly represents the original person.
(結果應該感覺像是一個無法抗拒的Q版吉祥物，清楚地代表原始人物。)

Q版轉換非常適合個人品牌、團隊頭像和商品設計。

Create a realistic highway billboard mockup featuring this product.
(創建一個以該產品為特色的逼真高速公路廣告牌模型。)

BILLBOARD CONTENT (廣告牌內容):
- Product bottle prominently displayed on left third (產品瓶子醒目地展示在左三分之一處)
- Main headline on right (EXACT TEXT, render verbatim): (右側主標題（確切文本，逐字渲染）：)
  "Fresh & Clean — Every Day"
- Tagline below headline: "Nature's Best Ingredients" (標題下方的標語："Nature's Best Ingredients")
- Small logo placeholder area in bottom right corner (右下角的小Logo占位區域)

TYPOGRAPHY SPECIFICATIONS (排版規格):
- Headline: Bold sans-serif, white text, high contrast (標題：粗體無襯線，白色文本，高對比度)
- Tagline: Light sans-serif, slightly smaller, same white (標語：細無襯線，稍小，同樣的白色)
- Clean kerning, centered alignment within text area (乾淨的字距，文本區域內居中對齊)
- Text appears EXACTLY ONCE — no duplicates anywhere (文本只出現一次 —— 任何地方都沒有重複)

SCENE (場景):
- Billboard on highway overpass or roadside structure (高速公路立交橋或路邊結構上的廣告牌)
- Sunset lighting creating warm, appealing atmosphere (日落光線營造溫暖、吸引人的氛圍)
- Photorealistic environment with motion-blurred vehicles below (照片級真實環境，下方有動態模糊的車輛)
- Professional advertising photography feel (專業廣告攝影感)

No watermarks. No additional marketing copy. No logos unless
specified. Text must be perfectly legible and correctly spelled.
(無水印。無額外營銷文案。除非指定，否則無Logo。文本必須完全清晰且拼寫正確。)

對於包含文本的營銷材料，始終使用 quality="high"。在最終使用前驗證拼寫。

Extract the product from this image for e-commerce use.
(從這張圖像中提取產品用於電子商務。)

OUTPUT SPECIFICATIONS (輸出規格):
- Transparent background (RGBA PNG format) (透明背景（RGBA PNG格式）)
- Crisp silhouette with clean edges (清晰的輪廓，邊緣乾淨)
- No halos or color fringing around product (產品周圍無光暈或彩色條紋)
- All product labels and text perfectly preserved (所有產品標籤和文本完美保留)
- Exact product geometry and proportions maintained (精確保持產品幾何形狀和比例)

OPTIONAL ENHANCEMENT (可選增強):
- Add subtle, realistic contact shadow (添加微妙、逼真的接觸陰影)
- Shadow should be soft and natural, no hard edges (陰影應柔和自然，無硬邊)
- Shadow works with the transparent background (陰影適用於透明背景)

CRITICAL CONSTRAINTS (關鍵約束):
- Do NOT restyle or recolor the product (不要重新設計或重新著色產品)
- Do NOT modify product appearance in any way (不要以任何方式修改產品外觀)
- Only remove background and add optional shadow (僅移除背景並添加可選陰影)
- Preserve every detail of the original product exactly (精確保留原始產品的每一個細節)

注意：當前模型渲染棋盤格圖案來表示透明度 —— 可能需要後期處理才能獲得真正的Alpha通道。

import os
import base64
from openai import OpenAI

client = OpenAI()

# Create output directory (創建輸出目錄)
os.makedirs("output_images", exist_ok=True)

def save_image(result, filename: str) -> None:
    """Save base64 image response to file."""
    image_base64 = result.data[0].b64_json
    with open(f"output_images/{filename}", "wb") as f:
        f.write(base64.b64decode(image_base64))

# Basic text-to-image generation (基本文生圖生成)
result = client.images.generate(
    model="gpt-image-1.5",
    prompt="Your detailed prompt here",
    quality="high",  # or "low" for faster iteration (或"low"用於更快迭代)
    n=1  # number of variations (變體數量)
)

save_image(result, "output.png")

result = client.images.edit(
    model="gpt-image-1.5",
    input_fidelity="high",  # Essential for identity preservation (身份保留至關重要)
    quality="high",
    image=[
        open("input_images/source.png", "rb"),
        open("input_images/style_reference.png", "rb"),
    ],
    prompt="""
    Apply the artistic style from Image 2 to the subject in Image 1.
    (將圖片2的藝術風格應用到圖片1的主體上。)

    PRESERVE: subject's identity, pose, and composition
    (保留：主體的身份、姿勢和構圖)
    CHANGE: artistic style, color palette, texture treatment
    (改變：藝術風格、調色板、紋理處理)

    Do not add new elements. Maintain subject likeness exactly.
    (不要添加新元素。精確保持主體相似度。)
    """
)

save_image(result, "styled_output.png")