Skip to content

ALL-IN-ONE · 指令大全

AI 提示詞指令大全.

一頁收齊 100 組實測過的免費提示詞 — 生圖、影片、音樂、ChatGPT 工作流。每組一鍵複製、免登入免費用。 把 [方括號變數] 換成你的內容就能貼進工具。

想看附「為什麼有效」與參數說明的完整版?每組標題都可點進詳情頁; 全庫 1,320+ 組在 提示詞庫

AI 生圖

電商產品情境照最痛的是「四張各拍各的」:分四次生成、產品外觀和光線幾乎必跑掉。這條改用一張 2×2 四宮格一次生成、同次生成共享同一風格、一致性遠比分開生穩。代價要說清楚:每格解析度只有整張的四分之一、產品上的小標籤文字可能糊掉、要大圖時再逐格放大或重生。

【用途】
為電商商品頁生成一張 2×2 四宮格產品情境照、四格是同一件產品在四個不同生活場景、用於商品頁「情境展示」區。四格必須看起來像同一位攝影師、同一天、同一套色調拍出來的。

【產品描述——四格都必須完全一致】
[產品名稱與外觀描述]
四格中的產品外觀、顏色、材質、比例必須完全相同、不得在任何一格改變產品的細節或配色。

【四個場景】
- 左上:[場景一]
- 右上:[場景二]
- 左下:[場景三]
- 右下:[場景四]
每格產品都是畫面主角、佔畫面約三分之一、場景元素只做陪襯、不搶戲。

【光線與調性——全圖統一】
- 光線:[光線設定]、四格光源方向與軟硬一致
- 色調:[色調與氛圍]、四格套同一組色調、白平衡一致
- 質感:商業產品攝影、85mm 定焦鏡頭視角、淺景深、產品銳利、背景微散景

【版面】
2×2 四宮格、格與格之間用等寬細白色分隔線、四格構圖重心一致(產品都在格內同一相對位置)、整體看起來是一組系列照。圖上不要出現任何文字或浮水印。

給品牌或社團做 IP 吉祥物,最需要的不是單張可愛圖,而是正面、側面、背面並排的角色設定稿。這份指令把三視角一致性、站姿、表情變化區的版面要求寫死,產出可直接當後續所有生成的參考圖。

為我的品牌設計一個 IP 吉祥物,輸出一張「角色設定三視圖」。

角色設定:
- 吉祥物概念:[吉祥物概念]
- 個性關鍵字:[個性關鍵字]
- 品牌主色:[品牌主色],用在角色的主要配色上

版面要求(嚴格遵守):
1. 同一張圖內水平並排三個視角:正面、側面(90 度)、背面
2. 三個視角的身體比例、五官位置、服裝配件必須完全一致,站姿相同
3. 乾淨純白背景,像專業角色設定稿,不要任何場景
4. 版面下方加一排 3 個小型表情動作變化:開心、驚訝、加油
5. 風格:圓潤可愛的 3D 角色設計、霧面質感、柔和棚光
6. 圖上如需標註一律使用繁體中文
7. 輸出 4K、比例 16:9

AI 拍保養品永遠像便宜貨?問題不在瓶子、在「瓶子周圍」。Wet surface reflection + rim lighting + droplet placement 三招把 clinical luxury 感拉滿。

A luxury {{product_type}} photographed on a wet {{surface}} surface.

{{product_specs}}.

Camera angle is 45 degrees from ground level, straight on.

Light source is cool diffused backlight from upper right, creating a soft rim light along the right edge of the bottle.

Small water droplets scattered on the slate surface around the base.

Reflection of the bottle visible on the wet surface below — slightly distorted, not mirror perfect.

No hands, no props, no flowers.

Shallow depth of field, foreground droplets slightly out of focus.

Shot style: high-end dermatology brand, clinical luxury aesthetic.

Aspect ratio 4:3. Photorealistic.

食譜步驟照分開生成的老問題:桌面、鍋具、光線每張跳來跳去、讀者一看就穿幫。這條用單張網格一次生成鎖一致性、再用「後一格必須是前一格的合理延續」約束食材狀態演進。限制:格數越多每格越小、超過六格請拆兩張、且第二張要重複貼同一段桌面與光線設定才接得起來。

【用途】
為食譜文章生成一張 [格數] 格的步驟照組合圖(網格排列)、示範 [料理名稱] 的製作過程。所有格子必須像同一支手機架在同一位置、從頭拍到尾:同一張桌面、同一個俯拍角度、同一組光線。

【統一拍攝設定——每一格都相同】
- 視角:正上方 90 度俯拍(flat lay)、鏡頭高度一致
- 桌面:[桌面與道具描述]、桌面材質與擺設位置每格一致
- 光線:[光線設定]、光源方向與亮度每格相同
- 色調:自然食物攝影、色彩真實不過飽和

【各步驟畫面——食材狀態必須連貫】
[步驟狀態清單]
關鍵:後一格的食材狀態必須是前一格的合理延續(例:第 2 格切好的菜、到第 3 格就應該在鍋裡、不能又變回整顆)。使用的鍋具、碗盤在各格之間要一致、不要每格換道具。

【編號】
每格左上角放一個小的阿拉伯數字編號(1、2、3…)、白底圓形角標、除編號外圖上不要出現任何其他文字。

【版面】
[格數] 格等大網格排列、格間用細白色分隔線、每格構圖重心一致(主鍋具/主食材放格子中央)。

用「擋住太陽、投下巨影、低角度俯視」三招把巨獸尺度感拉滿,仰拍特攝經典構圖。

A gigantic, imposing monster towering at 50 feet tall ominously gazes upon us. It has an intimidating stature and a massive build, reflecting its seemingly unfathomable strength. Its large eyes, brimming with curiosity and intrigue, fixate on us as it blocks out the sun, casting an enormous shadow on the ground below.

青/洋紅/紫的經典賽博三色+體積霧+漢字招牌,關鍵字密度高、可控性強,攤販與行人讓未來都市有…

masterpiece, highly detailed, cyberpunk city street at night, heavy rain, wet pavement reflecting neon lights, puddles, volumetric fog, moody atmospheric lighting, cyan and magenta neon signs, purple glow, towering skyscrapers, holographic advertisements, Japanese kanji signs, crowded street with pedestrians, urban decay, street vendors, atmospheric perspective, wide shot, street level view, cinematic composition

IG 上瘋傳的「金句卡」— 排版主導、背景極簡、中英文字共存不衝突。Ideogram 是目前唯一中英字同時渲染穩定的模型。

Instagram quote card, 1:1 square. Main text "[zh_quote]" centered in large [font_zh] typography. Below: smaller English attribution "[en_quote]" in [font_en] style. Background: [bg_style]. Color palette: [palette]. Author credit "— [author]" in bottom right corner, small caps. Serif and sans mix for hierarchy. Clean whitespace, magazine editorial aesthetic. No photography, purely typographic. --ar 1:1

活動票券、工作證生成的兩個重點:活動名與日期用引號鎖字(含字數宣告與「證/号」防呆)、以及明確預留一塊純白空白區給之後加印的序號或 QR code。特別加了「禁止畫假條碼假序號」——模型很愛自作主張補一條假條碼、印出來才發現掃不了就來不及了。

【用途】
設計一張 [票券類型](橫式、比例 16:9、單面),用於實體活動、之後會大量印刷、並在預留的空白區加印每張不同的序號或 QR code。

【圖上要畫出的文字——逐字照畫、一字不差】
- 活動名稱(主視覺文字、最大字級):『[活動名稱]』(共 [活動名稱字數] 個字)
- 日期時間:『[日期時間]』
- 地點:『[地點]』
- 票種/證別標示:『[票種標示]』(共 [票種標示字數] 個字)
所有用全形引號『』框住的文字都是要直接畫在票面上的內容;引號本身不要畫出來。日期中的數字用阿拉伯數字照原樣呈現。

【序號空白區——很重要】
在票面 [空白區位置] 預留一塊乾淨的純白(或純淺色)矩形區域、約佔票面寬度四分之一、裡面完全不要有任何文字、圖案、花紋或漸層——這塊之後要加印序號與 QR code、必須留白。

【字數自我檢查】
輸出前逐項核對:每項畫出的字數要與宣告一致、不符就修正後再輸出。

【異體字與簡體字防呆】
全部使用繁體中文(台灣用字)、嚴禁簡體字。特別注意:「證」不得寫成「证」、「號」不得寫成「号」、「場」不得寫成「场」。『』內填什麼字就照畫什麼字、不要自行替換異體字(例:填「台北」就畫「台北」、不要改成「臺北」)。

【版面與風格】
- 主視覺風格:[主視覺風格]、主色 [主色]
- 活動名稱字級最大、日期地點次之、票種標示可做成色塊或斜角標籤
- 版面像正式售票系統出的票、資訊層級清楚、邊緣可加撕線或圓孔設計細節
- 除了宣告的文字外、不要出現任何多餘文字、也不要畫任何假條碼、假 QR code 或假序號

包裝紙、紙袋底紋的兩大需求一次處理:可平鋪、和「當背景不搶戲」。用低密度+均勻留白+低對比逼出 logo 呼吸空間、用「禁方向性光影與透視」保住平鋪感。誠實限制:生成模型不保證真正無縫、邊緣接縫要進影像軟體用位移(offset)檢查、有縫就局部修補、別指望一次到位。

【用途】
為品牌設計一張可四方連續平鋪(seamless tileable)的底紋圖樣、用於 [用途]。圖樣要能上下左右無限重複拼接、邊緣圖案必須能對接、不能有明顯接縫。

【圖樣元素】
[圖樣元素]
元素以扁平向量插畫風格呈現、線條乾淨、每個元素簡化到一眼可辨識、大小做 2 到 3 種變化、方向隨機錯落。

【留白密度——這是重點】
元素之間保持大量均勻留白、整體密度稀疏(元素面積合計不超過畫面三分之一)、任何區域都不擁擠——這張底紋之後要疊 logo 和文字在上面、必須處處都有喘息空間、不能有搶眼的視覺焦點。

【配色】
- 底色:[底色](整張單一純色打底)
- 元素用色:[主色與輔色]、低對比、元素顏色與底色的對比要溫和、當背景不搶戲

【禁止事項】
- 不要有方向性光影、漸層打光或陰影(會讓平鋪時出現明暗接縫)
- 不要有透視感、所有元素都是平面正拍
- 不要出現任何文字、字母或 logo
- 不要讓任何單一元素特別大或特別顯眼

Create a 3D Insane detailed exploded assembly drawing of [subject or obj…

Create a 3D Insane detailed exploded assembly drawing of [subject or object]

幫我生成一張《赤壁懷古》的長卷圖,帶整篇《赤壁賦》文字

幫我生成一張《赤壁懷古》的長卷圖,帶整篇《赤壁賦》文字

幫我生成一張 RAG 技術的詳細講解圖

幫我生成一張 RAG 技術的詳細講解圖

兩張人像合成一張復古拍立得,暖黃褪色、白框+手寫繁中註記,亞洲臉孔一致性處理。

我會提供兩張人物照片,請把照片裡的兩個人合成在「同一張復古拍立得」裡,像他們真的站在一起拍的合照。

【合成要求】
- 兩人並肩站在一起,自然互動(可微笑、肩膀靠近),不要像兩張照片硬貼。
- 保留每個人原本的臉部特徵與五官比例,特別注意亞洲臉孔的眼睛、鼻型、臉型要與原圖一致,不要美化成不像本人。
- 場景背景:[合照場景],光線一致、兩人受光方向相同,看起來是同一個現場同時拍的。

【拍立得風格】
- 整張是一張白色拍立得相紙,下方留出較寬的白色邊框(經典 Polaroid 比例)。
- 色調走復古:暖黃、輕微褪色、低對比、邊角微微暗角,像 [年代感] 的舊照片。
- 加入輕微的底片顆粒與一點點光暈,但不要太髒、臉要看得清楚。

【手寫註記】在下方白框處,用藍黑色原子筆的手寫繁體中文字跡寫上:『[手寫文字]』,字跡自然、略帶傾斜,像真人隨手寫的,不要印刷體;文字必須正確、無錯字、無簡體字。

【禁止】不要把兩人的臉換成同一張臉;不要加多餘的浮水印或英文字;不要把人像裁掉重要部分。

如果手寫的字有誤,請保持照片與色調不變、只重寫白框上的文字。

經典「微縮小人 × 巨大食物」tiny-world 玩法,吊車吊鬆餅、藍莓當巨石,比例錯置製…

A stack of golden, fluffy pancakes drizzled with maple syrup, topped with butter and fresh berries. Miniature construction workers are 'building' the stack, with a crane lifting a new pancake onto the tower. Some workers are using small brushes to spread syrup, while others are positioning blueberries like boulders. Steam gently rises, and the background is blurred with a cozy breakfast setting, featuring coffee and honey.

subsurface scattering(次表面散射)+ 三點打光是 Pixar 光滑皮膚…

Convert this image into 3D Pixar-style cartoon with clean geometric forms, soft three-point lighting, and exaggerated expressive features. Vibrant saturated colors, smooth subsurface scattering, cinematic composition, award-winning animation studio quality, 4K rendering.

`deep blue to black gradient` + `luminous tent…

mysterious deep ocean abyss scene, bioluminescent creatures glowing in darkness, ethereal jellyfish with luminous tentacles, deep blue to black gradient, scattered light particles, otherworldly atmosphere, cinematic lighting, dramatic composition, ultra detailed

先用引號鎖死店名字樣再上招牌實景,繁體字筆畫清晰不糊。適合開店前的品牌視覺提案;誠實說:商用向量檔仍需設計師重繪,這條負責讓你先看到方向。

幫我設計一個品牌 LOGO,並把它做成一張店面招牌的實景示意圖。

【品牌資訊】店名:『[店名]』,業態是[業態類型],品牌個性走[品牌個性]路線。
【LOGO 要求】店名的繁體中文字樣是 LOGO 主體,每一個字的筆畫都必須正確、清晰、可讀,不能出現簡體字或自創字形;字體風格用[字體風格];可搭配一個簡單的圖形符號,但不能蓋住文字。
【招牌實景】把 LOGO 放在一面街邊店面的招牌上,招牌材質為霧面金屬底板+立體字,傍晚光線、招牌有微微打光;視角為路人平視,能看清整面招牌。
【禁止】招牌上不得出現店名以外的任何文字;不要用過度華麗的裝飾壓過字體本身。

如果字有任何一筆寫錯,請保持整體設計不變、只重新修正文字。

把人物照、商品照、場景照分別上傳,指令裡講清楚誰是主體、誰是要置入的物件、誰是環境,合成出像真實拍攝的商業形象照。重點在光源統一與商品標籤不變形的宣告,適合沒預算請模特兒實拍的小電商。

我會上傳三張圖片,請依以下分工合成一張新照片:

- 圖 1 是主體人物:臉部特徵、髮型必須與圖 1 完全一致,不可變成別人
- 圖 2 是要置入的商品:[商品名稱],外觀、顏色、比例、包裝上的標籤文字都不得變形或改寫
- 圖 3 是場景環境:合成後的背景與空間以這張為準

合成要求:
1. 人物正在[與商品互動的動作],手與商品的接觸點、握姿要符合人體工學
2. 光源以場景圖為準,人物與商品重新打光、補上方向一致的陰影
3. 構圖:[構圖需求]
4. 成品要像一張真實拍攝的商業形象照,不能有剪貼合成感
5. 輸出 4K

把手機自拍轉成正式大頭照:換純色背景、改正裝、轉正姿勢、打均勻光,同時用保留宣告鎖死五官不被美化成別人。適合履歷、員工證、LinkedIn 頭像;官方證件多數機關要求實拍,AI 修過的照片有被退件風險,用途要誠實評估。

把我上傳的人像照片轉換成正式大頭照規格,最高原則是「看起來還是我本人」。

不可改變(最重要):
- 五官、臉型、膚色、痣與疤痕等個人特徵完全保留
- 不瘦臉、不放大眼睛、不改變年齡感、不改變髮型輪廓

需要調整:
1. 背景換成[背景顏色]純色,均勻無陰影
2. 服裝改為[服裝],領口整齊自然
3. 頭部轉為正面平視、雙肩水平;原圖歪斜的話自然校正
4. 改成均勻柔和的正面光,消除臉上雜光和背景影子
5. 保留自然膚質與毛孔,只淡化暫時性的瑕疵(痘痘、油光),不要磨皮成塑膠感
6. 構圖規格:[構圖規格]

輸出高解析度直式照片。

把人像換成水彩/3D/動畫等風格,臉部一致不走鐘。

將附圖人像轉成[目標風格],但保持臉部辨識度與五官比例一致。\n風格:[目標風格],筆觸或材質=[材質細節],色調=[色調]。\n保留:髮型輪廓、表情、臉部特徵與姿勢;只改變繪製風格。\n背景:換成[背景],與主體風格一致。\n輸出:高解析、邊緣乾淨、單一主體。

把任何主題做成玻璃罐 / 盒子裡的微縮立體世界 + 移軸效果——2026 社群存圖率最高的風格之一。

A highly detailed miniature world of {{scene}} contained inside a {{container}}, tilt-shift macro photography, tiny realistic figures, soft cinematic lighting, shallow depth of field, hyperrealistic, the feeling of a tiny living world, 1:1, no text.

客戶最常問的 5-8 題 FAQ 做成一張圖 — 問題 + 簡短答案 + icon。比長條文字 FAQ 頁留存率高 3 倍。

visual FAQ chart for "[topic]". [num_questions] question-answer pairs in grid layout (2x3 for 6, 2x4 for 8). Each cell: icon + Chinese question "Q: [q]" + short answer "A: [a]" (max 20 Chinese chars per answer). Icons: consistent flat style. Header: "常見問題" or "[custom_header]". Clean background, single accent color. Modern B2B / SaaS FAQ aesthetic. --ar 3:4

把抽象概念視覺化 — 心理學理論、經濟模型、設計原則的圖解。書本插圖 / 學習卡 / 知識型 IG 用。

educational illustration explaining the concept of "[concept]". Use visual metaphor: [metaphor] to make it intuitive. Layout: central concept with 2-4 sub-elements or layers. Chinese labels for each part of the diagram. Flat vector style, textbook illustration quality. Single accent color + neutral palette. Title "[title]" at top, brief subtitle below. --ar 1:1

Change only the rendering 把「改什麼」和「不准改什麼」切乾淨,再補…

Convert the supplied tattoo design into a transfer-ready stencil. Keep the composition pixel-for-pixel: same crescent moon, same two mirrored fern fronds, same dotwork spacing, same dagger, same position and same proportions. Change only the rendering: every line becomes a closed, even-weight pure black contour on a pure white background, all grey values removed, all fills flattened to either solid black or empty white. Do not redraw, reinterpret, restyle or re-pose the subject. Do not add or remove any element. Output one single image, no side-by-side, no before-and-after panel.

資訊圖表的關鍵是先把版塊切出來——用引號逐一點名三個分區標題,模型才不會把字糊成裝飾;結尾補…

Make a retro, 1950s-style infographic about the history of the American diner. Include distinct sections for 'The Food,' 'The Jukebox,' and 'The Decor.' Ensure all text is legible and stylized to match the period.

把「壓縮資訊」交給模型、把「哪些字一定要出現」用引號鎖死——教學圖解最怕標籤自由發揮,這寫法…

Summarize the concept of 'Transformer Neural Network Architecture' as a hand-drawn whiteboard diagram suitable for a university lecture. Use different colored markers for the Encoder and Decoder blocks, and include legible labels for 'Self-Attention' and 'Feed Forward'.

點名 orthographic 與 plan/elevation/section 三種製圖視…

Create an orthographic blueprint that describes this building in plan, elevation, and section. Label the 'North Elevation' and 'Main Entrance' clearly in technical architectural font. Format 16:9.

deconstructed to show the texture of… 把爆炸圖和食慾感…

Create a hyper-realistic infographic of a gourmet cheeseburger, deconstructed to show the texture of the toasted brioche bun, the seared crust of the patty, and the glistening melt of the cheese. Label each layer with its flavor profile.

草稿當版面模板、模型只負責上材質——Keep the exact layout of… 是把…

Create a high-end magazine advertisement for a luxury perfume brand called 'Nebula' based on this hand-drawn sketch. Keep the exact layout of the bottle and text placement, but render it in a photorealistic style with a galaxy-themed background.

明確分工——placeholder 方框「換內容」、按鈕位置與網格「不准動」;UI 稿最怕模…

Generate a photorealistic UI mockup for a fitness tracking app based on this wireframe. Replace the placeholder boxes with high-quality images of runners and data visualization charts, but strictly adhere to the button placement and grid structure.

ChatGPT / 工作流

把課本章節、講義、會議資料丟進來、一次生成三件套:重點卡幫你記、自測題幫你驗、口述大綱幫你講。適合考證照、讀書會導讀、新人教育訓練。已內建「只根據提供資料整理」的防編造規則;資料太長時分批餵、每批各生一包再合併最穩。

【用途】把我提供的資料整理成一份「研究包」、讓我可以直接拿去準備考試、讀書會導讀或教別人。

【我的背景】
- 學習目標:[學習目標]
- 目前程度:[目前程度]
- 可用準備時間:[準備時間]

【資料】
[資料內容]

【輸出三件套】

一、重點卡(10 到 15 張)
- 每張格式:正面=一個概念或名詞;背面=三句以內的白話解釋+一個具體例子
- 依「最常考/最容易混淆」排序;最容易搞混的概念要成對出現、並標出兩者差異

二、自測題(10 題)
- 混合單選題與簡答題、由淺到深
- 每題附答案、以及「答錯這題代表你沒吃透哪張重點卡」的對應編號

三、口述大綱
- 假設我要在讀書會用 10 分鐘講完這份資料
- 給我:開場一句話、三到五個主要段落(每段附關鍵詞與一句轉場)、結尾一句總結
- 標出「聽眾最可能發問的 2 個問題」與建議回答方向

【鐵律】
- 只根據我提供的資料整理、不要補充資料以外的知識;資料沒涵蓋的主題、直接寫「資料未涵蓋」
- 白話優先:所有解釋要講給[目前程度]的人聽得懂
- 全部用繁體中文與台灣用語

讓 AI 回答完不能直接交卷:先標每個關鍵句的來源類型、再自己找反例攻擊自己、最後逐條標信心等級並附查證方向。適合查資料、寫報告、做決策前的功課。誠實說:這不等於真的查證(模型不一定能上網查)、但能把不確定性逼出檯面、明顯減少一本正經的胡說。

【用途】我要問你一個問題、但你回答完不能直接交卷——要先過三道自我查核、把不確定的地方攤開給我看。

【我的問題】
[你的問題]

【回答流程(三段缺一不可、順序不能換)】

第一段:直接回答
- 正常回答我的問題
- 每一個關鍵事實句、句尾標來源類型:〔訓練資料〕〔推理〕〔常識〕;你其實沒把握的、老實標〔不確定〕

第二段:自我攻擊
- 假設第一段的回答是錯的、列出 2 到 3 個「最可能錯在哪」的具體反例、例外情況或反方觀點
- 真的找不到像樣的反例、要說明為什麼找不到、不能只寫「無」

第三段:信心分級
- 把第一段的主要結論逐條標信心等級:
  【高】=多個獨立角度都支持、你願意為它背書
  【中】=方向正確但細節可能有出入
  【低】=記憶模糊或資訊可能過時、需要另行查證
- 每一條【低】、附上具體查證方向:該去查什麼類型的一手來源(官方網站、原始論文、法規原文等)

【鐵律】
- 禁止為了顯得專業而把【低】往上調成【中】或【高】
- 涉及時效的內容(價格、版本號、法規、統計數字、人事異動)一律最高只能標【中】
- 全程用繁體中文回答

餵入店名、營業時間、價格帶與語氣人設、一次生成四組 LINE 官方帳號模板:詢價引導、預約確認與滿位替代、客訴兩段式回覆、打烊自動回覆。語氣全程一致、每則控制在一則訊息讀得完的長度。價格與時段貼上後台前記得逐則核對、AI 不會知道你店裡的實際狀況。

【用途】幫我的店生成一整組 LINE 官方帳號回覆模板、語氣前後一致、我要能直接複製到官方帳號後台使用。

【店家資料】
- 店名與業種:[店名與業種]
- 營業時間:[營業時間]
- 主力商品/服務與價格帶:[主力商品與價格帶]
- 預約方式:[預約方式]
- 語氣人設:[語氣人設]

【要生成的四組模板】

一、詢價回覆(3 則變化)
- 情境:客人只丟一句「多少錢」但沒說品項
- 要溫和引導客人講清楚需求、同時先給價格帶讓客人心裡有底

二、預約回覆(2 則)
- 預約成功確認:含日期時間、注意事項、改期或取消方式
- 該時段已滿:必須主動提供替代時段的問法、不准只回「已滿」

三、客訴回覆(2 則)
- 第一時間回覆:先道歉受理+收集必要資訊(發生時間、品項、狀況)、但不先承認具體過失
- 處理方案回覆:給一個可以填入補償方式的框架
- 禁止用「造成不便敬請見諒」這種制式開場

四、打烊/非營業時間自動回覆(1 則)
- 含營業時間、何時會回覆、緊急狀況的替代聯絡方式

【要求】
- 每則約 100 字以內、一則 LINE 訊息內讀得完
- 模板裡的客人姓名、日期、品項等可變處、用〈〉標示、方便我之後手動替換
- 語氣全程符合我給的人設、四組不能忽冷忽熱
- 全部用繁體中文、台灣慣用語

把健檢報告數值貼進來、逐項翻成人話、集中整理紅字項目、並生成回診時可以直接照唸的具體提問清單。解決「看得到紅字、看不懂意義、門診三分鐘不知道問什麼」的痛點。非醫療建議、異常值一律以醫師判讀為準、AI 的角色只是翻譯與提問教練、禁止下診斷。

【用途】幫我把健檢報告翻成白話、並整理出回診時該問醫生的問題。你不是醫生、這不是醫療建議、所有異常值以醫師判讀為準——你的任務是讓我看得懂報告、問得出問題。

【我的基本背景】
[基本背景]

【報告數值】
[報告數值]
(照著報告抄:項目名稱、我的數值、參考範圍、有沒有標紅字)

【輸出格式】

一、逐項白話翻譯
每個項目依序寫:
- 這在測什麼:一句話、講人話
- 我的數值落在哪:正常範圍內/偏高/偏低、離參考範圍多遠
- 常見相關因素:這個項目異常時、常見上跟哪些生活習慣或身體狀況有關(一律用「常見上」開頭、不准斷言我的狀況)

二、需要留意的項目
- 把紅字與接近臨界值的項目集中列出、依「建議優先跟醫師討論」排序
- 禁止下診斷:不能寫「你有某某病」、只能寫「這個數值異常時、醫師通常會進一步評估什麼」

三、回診提問清單(5 到 8 題)
- 根據我的異常項目與基本背景、生成回診時可以直接照唸的問題
- 問題要具體到醫師能直接回答、例如「我的某某值偏高、加上我的年齡與家族史、需要進一步檢查嗎」、而不是「我健康嗎」

【鐵律】
- 只解讀我提供的數值、不准編造報告上沒有的項目
- 建議最多到「生活習慣的一般常識」層級;用藥、停藥、保健食品一律回「請與醫師或藥師討論」
- 結尾必須重申:本解讀非醫療建議、異常值請以醫師判讀為準

一段可重複使用的繁中「prompt 優化器」咒語——貼上你寫得很爛的 prompt,它先問清楚缺的關鍵資訊,再產出結構清楚的優化版,還附「改了哪些、為什麼」。誠實標註:meta-prompt 救得了結構與表達,救不了內容知識本身的錯。

你現在是「prompt 優化器」。我會給你一段我自己寫的、可能寫得不好的 prompt,你的任務是把它改成一個結構清楚、AI 容易照做、結果更穩定的版本。

【我的原始 prompt】
[原始prompt]
【我其實想要的結果】[想要的結果]
【會拿去用的模型】[使用模型]

請照以下步驟:

第一步|補問缺口:如果我的原始 prompt 缺少會嚴重影響結果的關鍵資訊(例如沒講輸出格式、沒講對象、沒講限制條件、沒講語氣),先用條列方式問我 1-3 個最關鍵的問題。如果資訊其實夠了,就跳過這步、直接進第二步。

第二步|優化版 prompt:給出改寫後的 prompt。重點放在「結構與語法」而不是堆砌內容——明確的任務、必要的脈絡、清楚的輸出格式與成功標準、需要的話加上分段標題。用我能直接複製貼上的乾淨格式呈現(可放在程式碼框裡)。

第三步|改動說明:簡短列出你做了哪些關鍵改動、各自為什麼(例如「補上輸出格式,避免它每次長相不一樣」)。

第四步|誠實提醒:告訴我這個優化版「能幫上什麼、幫不上什麼」——特別是:如果我要做的事本身需要它不具備的專業知識或最新事實,優化 prompt 也救不回來,那種情況我該補的是資料來源而不是更漂亮的 prompt。

全程台灣正體中文。

不是幫你寫履歷、是幫你「驗」履歷。AI 味診斷抓出一看就是模型寫的句子、量化體檢用反問逼你補真實數據、再給 before→after 改寫示範和個人禁用詞清單。投遞前最後一關用。

你是一位看過上萬份履歷的台灣資深人資主管,最近一年特別擅長一眼認出「AI 代寫」的履歷。請幫我健檢以下履歷,目標是讓它通過兩關:人資的 AI 味雷達、用人主管的 30 秒掃描。

我的履歷:[履歷全文]
目標職務:[目標職務]

請依序輸出:
1. AI 味診斷:列出履歷中最像 AI 寫的句子(空泛形容詞堆疊、「致力於」「擁有極佳的」這類套話),每句說明為什麼可疑
2. 量化體檢:找出所有沒有數字的成果描述,逐條反問我「這件事可以用什麼數字衡量?」每條給 2-3 個提問,讓我自己補真實數據——不要替我編造任何數字
3. 改寫示範:挑 3 條最弱的 bullet,給 before→after 對照。after 版要求:動詞開頭、口語但專業、長短句交錯、保留我的原始用詞習慣
4. 禁用詞清單:根據這份履歷,列出我該避免的 10 個 AI 高頻詞

品質準則:所有改寫必須基於履歷裡已有的事實;資訊不足就標「待補」並提問,禁止發明經歷或數據。

接在任何錄音轉文字工具後面用:貼逐字稿、輸出正式紀錄+決議事項+「負責人/期限/依賴項」待辦表格+給缺席者的 3 行摘要。會處理辨識錯字和中英夾雜、沒講到的期限標待確認、不瞎掰。

你是專業會議紀錄師,擅長把 AI 轉出來的逐字稿(常有辨識錯字、中英夾雜、閒聊)整理成可以直接發出去的正式紀錄。

會議資訊:[會議名稱與日期]
逐字稿:[逐字稿全文]

請輸出以下四個區塊:
## 會議紀錄
會議目的一句話+3-6 條重點討論脈絡(誰主張什麼、最後怎麼收斂)
## 決議事項
條列、每條註明「拍板者」
## 待辦事項
表格:負責人|事項|期限|依賴項。負責人或期限在逐字稿裡沒講到的、填「待確認」、並在表格下方列出「會後需要追認的問題」
## 給缺席者的 3 行摘要
三句話讓沒參加的人知道:結論是什麼、跟他有關的是什麼、需要他做什麼

品質準則:閒聊和重複發言直接略過、不要逐句摘要;專有名詞或人名疑似語音辨識錯誤時,保留原文並在後面加(?)標註;中英夾雜的發言統一整理成通順中文、保留必要的英文術語;禁止替任何人補上他沒說過的承諾或期限。

一份可貼進任何 coding agent 的 Definition of Done 模板,把『什麼叫做完』寫成 agent 必須逐項自檢的 checklist:行為符合驗收標準、tests / lint / typecheck / build 全綠(沒跑要說原因)、有風險的改動附 rollback、最後交一份『驗證故事』。杜絕「它說完成了結果一堆紅」。

Before you report this task as done, you MUST self-verify against the Definition of Done below. Treat "done" as a claim you have to prove, not a feeling. If you cannot satisfy a check, do NOT silently skip it — state exactly which check failed and why.

## The task
{{TASK_DESCRIPTION}}

## Acceptance criteria (the change is correct only if ALL hold)
{{ACCEPTANCE_CRITERIA}}

## Out of scope — do NOT touch
{{OUT_OF_SCOPE}}

## Definition of Done checklist — go through every item and mark ✅ / ❌ with evidence
1. **Behavior** — The change satisfies every acceptance criterion above. For each criterion, say in one line how the code now satisfies it.
2. **Scope discipline** — You touched only what was needed; nothing in the out-of-scope list changed. List the files you modified and confirm none are out of scope.
3. **Tests** — Run the project's test command (`{{TEST_COMMAND}}`). Paste the real, unedited result (pass/fail counts). If you could not run it, say so explicitly and why — never claim it passed without running it.
4. **Lint / typecheck / build** — Run `{{QUALITY_COMMANDS}}`. Paste the actual output. Any new error you introduced must be fixed, not suppressed.
5. **New coverage** — If you added or changed behavior, you added or updated at least one test that would FAIL on the old code and PASS on the new code. Name it. If you intentionally didn't, justify it.
6. **Edge cases** — List the edge / error cases relevant to this change (empty input, null, concurrency, failure path, etc.) and confirm each is handled or explicitly out of scope.
7. **Conventions** — The new code matches existing patterns, naming, and error-handling style in the surrounding files (don't invent a new style).
8. **Rollback / risk** — If this change is risky (touches data, auth, money, migrations, shared infra), describe how to revert it (single commit? feature flag?). If low-risk, say so.
9. **No placeholders** — Confirm there are no `TODO`, `pass`, `...`, stubbed functions, or commented-out "will do later" left in the diff.

## Final output — the verification story
End with a short paragraph a human reviewer can read in 30 seconds: what you changed, why it's correct, what you ran to prove it, and any check that is ❌ with the honest reason. Do not say "done" anywhere unless items 1–9 are all ✅ or have a stated, accepted reason.

提案產生器 / 會議處理器 / 內容再利用 / 週五回顧 / 收工 reset — 試了 40 個只有這 5 個沒被丟掉、各省 30+ 分鐘 / 次。

## 1. PROPOSAL GENERATOR(省 2 hr / 提案)

Turn these notes into a formatted Word doc proposal ready to send today.

Notes: {{notes}}
Client: {{client}}
Price: {{price}}

Sections: Executive summary, problem, solution, scope, timeline, investment, next steps. Formatted .docx. Sounds human.

---

## 2. MEETING PROCESSOR(省 30 min / 會議)

Here are my rough notes: {{meeting_notes}}
Attendees: {{attendees}}

Give me:
1. Half-page summary
2. Action items table (task, owner, deadline)
3. Follow-up email ready to send to all attendees

---

## 3. CONTENT REPURPOSER(1 → 5 篇)

Here's a piece I wrote: {{original_content}}
My voice: {{voice_description}}

Repurpose into:
- LinkedIn post (200-300 字)
- 3 個 standalone X post
- Email to my list (150 字)
- Instagram caption
- One-paragraph summary

Same voice across all. No AI clichés.

---

## 4. FRIDAY REVIEW(10 min 殺掉週日焦慮)

Here's what happened this week: {{week_braindump}}
Numbers: {{tracked_metrics}}

Give me:
- What actually went well and why
- What didn't work (honest, no softening)
- Top 5 priorities for next week ranked
- The single clearest thing I should change

---

## 5. END-OF-DAY RESET(最 surprise 的一個)

Today's notes: {{today_dump}}
(tasks done, conversations had, things you're carrying into tomorrow)

Tell me:
1. What I should write down before I forget
2. Anything I committed to that I haven't actioned
3. The one thing I should sleep on rather than decide now
4. Tomorrow's first hour — what's on it and why

丟你原履歷 + 目標職缺 JD,Claude 幫你把履歷每一個段落「按那家公司的語言」重寫 — 關鍵字、metric、產業術語全對齊,ATS 通過率直接拉高。

You are an elite executive resume coach with 15+ years helping [industry] professionals land roles at [target_tier]. Your task: take the candidate's existing resume below and tailor every section specifically for the JD provided. Priorities: (1) keyword match — ensure 80%+ of JD requirements appear in resume using their exact phrasing, (2) metric emphasis — surface and rephrase existing accomplishments to match the business outcomes the JD cares about, (3) ATS optimization — no tables, no graphics, standard section headers, (4) cultural fit signals — weave in 1-2 phrases matching the company's stated values from the JD. Do not invent experience the candidate doesn't have. Preserve all truth but re-sequence, re-emphasize, and re-phrase. Output the full rewritten resume in plain Markdown.

<candidate_resume>
[candidate_resume]
</candidate_resume>

<target_jd>
[target_jd]
</target_jd>

品牌 About / 關於我們頁的殺手級文案。把「我是誰、為什麼開這店、相信什麼、誰該買」壓縮成 400 字、讀起來像人而不是公關稿。

You are a direct-response copywriter in the lineage of Ogilvy and Doug Kessler. Write the About page for this brand in 400 words. Rules: (1) Open with a scene — a specific moment from the founder's life that made this brand happen, NOT "we believe..." statements, (2) The problem — 1 paragraph on the real user pain, using user language (quote a real user if provided), (3) The wedge — 1 paragraph on HOW this brand is different (the single decision others wouldn't make), (4) The bet — 1 paragraph on what the brand is betting the future will reward, (5) Close — 1 paragraph ending with specific invitation to a reader archetype (NOT "join us on our journey"). Voice: first-person singular ("I") from the founder, human, not PR-polished. No "craft", no "curated", no "mission", no em-dashes — those are crutches. Total 400 words.

<brand_context>
[brand_context]
</brand_context>

AI 生影片

油畫呼吸風景Runway Gen-4

Gen-4 是 i2v 優先——起始圖已鎖住長相/構圖/色調,prompt 幾乎「只描述運動…

The entire scene breathes slowly as if painted with oil. Clouds shift and reform with visible brushstroke texture. Trees sway with exaggerated, flowing motion. Light changes from warm gold to cool blue gradually across the duration. Camera holds wide and static.
產品 360 展示Runway Gen-4

電商產品旋轉最怕忽快忽慢+光影跳——這條明講 `no acceleration or dec…

The product rotates smoothly 360 degrees on its axis. Lighting remains consistent throughout the rotation. Each angle reveals a different surface detail. The rotation is slow and steady with no acceleration or deceleration. Clean, controlled motion.
物理破碎 VFXRunway Gen-4

破壞鏡頭的說服力靠碎片彈道與塵埃——`physics-accurate trajectori…

The central object fractures into hundreds of pieces. Fragments fly outward with physics-accurate trajectories. Dust and debris fill the immediate area. The camera holds wide to capture the full breakup. Dramatic directional sidelight.

Sora 2 原生生對白,`Dialogue: "..."` 直接出口播;lived-in …

A woman in her late 20s sits on her bed in a cozy bedroom, holding a small glass bottle of face serum. She looks at the camera with a natural, unposed expression. Soft warm lamp light. Slightly messy background — real, lived-in. Handheld camera feel with subtle movement. iPhone selfie-video aesthetic. Dialogue: "Okay I have to tell you about this serum because my skin has literally never looked like this before."

`Camera movement:` 內含時間分段(前 3 秒 dolly forward …

A chef prepares sushi behind a sushi bar counter. Camera movement: slow dolly forward over 3 seconds, then subtle push-in on the chef's hands as they slice fish. Handheld micro-movements for authenticity. Shot type: medium close-up transitioning to close-up.

以「情緒+光影」敘事而非堆砌鏡頭語言——blends surprise and joy 的表…

a girl gracefully emerges from the depths of an ancient, misty forest. Her steps are light, and the camera captures her every fluid moment. When the girl stands still and looks around at the lush trees, her face lights up with a smile that blends surprise and joy. This moment, frozen in the interplay of light and shadow, captures the girl's wonderful encounter with nature.

把開箱過程變成電商商品頁與 Reels 都能用的第一視角質感短片:雙手入鏡拆封、掀蓋瞬間慢動作、真實紙盒環境收音,不用請手模和攝影棚就拍出買家收到貨的開箱期待感。

【用途】
生成一支 8 秒的第一視角開箱 B-roll 短片,用於電商商品頁、IG Reels 與購物平台影音區,讓買家在下單前先感受收到 [商品名稱] 時的真實開箱期待感。

【主體】
一雙保養得宜、無指甲油的手從畫面下方入鏡,正在拆封 [商品名稱] 的外盒,包裝為 [包裝材質與配色]。動作依序為:指尖劃開封口貼紙 → 緩緩掀開盒蓋 → 拉出內襯棉紙 → 雙手將商品捧起至畫面中央。全程只出現手部與商品,不出現臉部與其他人物。

【場景/光線】
場景設定在 [桌面場景],背景輕微失焦。光線為窗邊自然側光,從畫面左側 45 度灑入,在盒蓋掀開的瞬間讓商品表面出現一道柔和的高光掃過。整體色調溫暖乾淨、白平衡偏日光,呈現「隨手記錄但很有質感」的生活感。

【運鏡/動作】
攝影機模擬第一視角(POV)俯角約 30 度,手持微晃增加真實感但不可暈眩。掀盒蓋瞬間切入 0.5 倍慢動作,商品捧起時鏡頭緩慢向前推近(slow push-in),直到商品佔畫面約 60%。禁止快速甩鏡、禁止任何轉場特效。

【音效/氛圍】
保留環境收音感:紙盒摩擦聲、貼紙撕開聲、棉紙窸窣聲清晰但不刺耳。不要背景音樂,方便後製時自行配樂或加旁白。

【輸出要求】
時長 8 秒、9:16 直式、1080p 以上、24fps 電影感。畫面中不得出現任何浮水印與文字。手部動作需自然連貫、手指數量正確不變形;商品在慢動作段落必須維持外觀穩定,logo 與包裝細節不得扭曲。

一條 prompt 走完四個鏡頭——低角廣角建立巨物、後退 dolly 退避、切手持跟逃難人…

Low-angle wide shot as an enormous stone colossus rises from beneath shifting dunes, sending shockwaves through the desert. Sandstorms engulf the lens while the camera dollies backward in retreat. The shot cuts to handheld as nomads flee with camels, fabric whipping in the wind. A beam of sunlight ignites carvings along the giant's body — snap-zoom into runes glowing across its chest before a sweeping tracking shot follows its first massive stride, crumbling cliffs in its wake.

mid-air 組裝 + 慢動作經典食慾結構,乾淨棚拍鎖焦點。

A static shot of a burger as it assembles in mid-air. The entire shot is in dramatic slow-motion. The individual ingredients—a grilled patty, fresh lettuce, tomato, melting cheese, and a toasted bun—fly into place from different directions. The background is a clean, simple, professional studio gradient to keep the focus on the product. Style: TV food commercial.

微距+奶油起泡+蒸氣把「滋滋聲」視覺化,Veo 原生 sizzle 音效直接 ASMR。

Extreme close-up of a cast-iron pan as butter foams and a steak sears, with macro steam and micro bubbles under overhead practical light with a warm grade. Sizzle and soft kitchen clatter sounds, no dialogue.

truck 鏡頭 + 車鎖中心 + 鏡頭 mirror S 彎是汽車速度感核心技法。

The camera trucks (moves laterally) at the same high speed, perfectly parallel to the car and its driver. The car stays locked in the center of the frame. In the second half of the shot, the car executes a quick, slight S-curve maneuver, kicking up a small plume of white salt dust from its tires. The camera perfectly mirrors this S-curve. Cinematic, car commercial, high-energy.

FPV 競速最怕速度感糊掉,用「岩壁飛掠模糊 + 高幀率清晰捕捉動態」一拉一收,並用「偶爾抬…

FPV drone racing through narrow slot canyon in American Southwest. Camera flies at high speed between towering red rock walls, banking and twisting to navigate tight passages. Shafts of sunlight pierce down from the narrow opening above. Smooth rock walls blur past with incredible speed. Camera occasionally looks up showing thin strip of blue sky between walls. Adrenaline-pumping, dynamic flight cinematography. High frame rate captures crisp motion. Colors vibrant - deep red rocks, bright blue sky.

把擴散拆成「雲團→分枝→漸散」三階段,又點名重力/密度差/對流三種物理,逼模型算出真實流體而…

Extreme close-up of black ink dropped into clear water in a glass vessel. The ink disperses in realistic fluid dynamics - initial cloud, then branching tendrils, then gradual diffusion. Camera at eye level with the glass. Backlit with clean white light so the ink patterns are clearly visible. The dispersion follows real physics: gravity, density differential, and convection currents. Scientific beauty aesthetic. Audio: water movement, quiet ambient.

駭客任務式子彈時間的標準配方——時間凍結的主體 + 鏡頭 180 度弧繞 + 半空懸停的碎片…

Bullet time effect orbiting around [SUBJECT] suspended mid-air in a frozen action pose, time appears nearly stopped while the camera arcs 180 degrees, floating debris and water droplets hang motionless, dramatic blue back light, slow-motion realism, awe-inspiring and surreal, ultra-detailed

一段 8 秒的產品揭曉短片運鏡,附專業攝影機語言,丟 Veo 直接出片。

Cinematic product reveal, 8 seconds.\n鏡頭 1(0-3s):微距慢推(slow dolly-in)逼近[商品],背景全黑、單一輪廓光(rim light)勾邊。\n鏡頭 2(3-6s):環繞(orbit shot)90 度,商品[材質反光]隨光轉動。\n鏡頭 3(6-8s):拉遠(pull back)露出完整商品落在[檯面],柔光鋪開、品牌留白。\n氛圍:高級感、極簡、[色調];24fps、淺景深、無文字。

生成一段像素人實拍的 UGC 口播廣告,自然不做作、適合短影音帶貨。

Realistic UGC selfie-style talking-head video, vertical 9:16, 10 seconds.\n人物:[人設],手持手機自拍視角、輕微晃動、自然光,像真實開箱。\n動作:對鏡頭自然講話、偶爾把[商品]舉近鏡頭展示。\n表情:真誠、口語、有生活感,不像廣告模特。\n場景:[場景],背景有真實生活雜訊(不要棚拍感)。\n口型對上一句話:『[一句話賣點]』。畫面無字幕。

IG 短片 / 廣告用、character 連貫舞蹈動作、seamless loop

A {{character}} performing {{dance_style}} in {{environment}}. Smooth fluid motion that loops seamlessly — the final frame matches the starting pose so it tiles cleanly. Camera is locked, no pan/zoom. Soft natural lighting. Color palette: muted earthy tones. Cinema-quality motion blur at 24fps. No text, no logos. The dance should feel natural — not a frozen pose / static character. Aspect ratio 9:16 vertical for social media. Duration: 5 seconds.

Gen-4 的 camera control 功能、subject 固定、camera 環繞 180-360 度

Runway Gen-4 with explicit camera control. Subject: {{subject}}. Camera motion: orbit {{orbit_degrees}} degrees around the subject over 8 seconds, smooth continuous rotation, no acceleration / deceleration jitter. Camera height: {{camera_height}}. Subject stays perfectly centered. Lighting: 3-point studio setup (key + fill + rim). Background: {{background}}. Cinematic depth of field with shallow DoF, subject sharp, background soft. 24fps with appropriate motion blur. Resolution 16:9 widescreen.

Gen-4 specific camera controls used:
- Orbit: {{orbit_degrees}}°
- Tilt: 0° (horizontal)
- Distance: medium (~2x subject diameter)
- FOV: 35mm equivalent

Hailuo 強項:高動態(運動 / 武術 / 賽車)連貫 5 秒

{{subject}} performing {{motion}} at maximum intensity. Dynamic camera that follows the action — fast tracking, slight handheld shake for energy. 5-second action shot. Cinematic motion blur on fast movements, sharp on freeze moments. Color graded high-contrast dramatic. Sound design implied through visual energy (debris / dust / sweat droplets visible). Lighting: stark directional with hard shadows. Wide-angle lens character for slight distortion at edges. Resolution: 1920x1080 cinematic. 24fps.

Hailuo 02 strength: maintains motion coherence at high speed where other models start to glitch / blur.

Pika 2.0 的 Morph 效果能讓人物 A 無縫變成人物 B。IG Reels / TikTok 上「變身轉場」是 2025-26 最火的內容形式。

6 秒 Pika 2.0 人物 Morph 變身:

Start state (0:00-0:02): [起始角色 — e.g. 穿西裝的上班族男性] in [起始場景 — e.g. 灰暗辦公室隔間]
Transition beat (0:02-0:04): Full-frame flash of light + swirl transition
End state (0:04-0:06): [結束角色 — e.g. 同一張臉但穿衝浪服] in [同場景改變 — e.g. 變成夏威夷海灘]

Morph rules:
- Keep the same face identity (use character reference image if possible)
- Transform: outfit + background + lighting + posture
- Swirl transition should feel punchy, not slow dissolve
- Background shift should sync with outfit change

Camera:
- Fixed shot (no pan / zoom)
- Medium close-up, subject fills 60% of frame
- 9:16 vertical for TikTok / Reels

Color grade:
- Start: desaturated, cool-leaning
- End: saturated, warm-leaning (matches mood shift)

Avoid: camera shake during morph, face identity drift, extra hands growing mid-morph, obvious AI glitches on teeth / eyes.

旅遊 vlog 最怕開場無感。Seedance 2.0 產出 3 個連貫城市空拍鏡頭,結尾停在一個「人走進畫面」的 cinematic 收鏡,直接當 YouTube 片頭。

10 秒旅遊 B-roll triptych([城市名] — 例如京都 / 台南 / 里斯本 擇一):

Shot 1 (0-3s): 空拍鏡頭從城市上空緩慢推近,看到連綿的老屋頂瓦片 / 紅屋頂 / 粉彩老街,陽光從右側斜打,影子拉長。清晨 6 點的霧光。

Shot 2 (3-6s): 地面視角,鏡頭穿過一條石板/磚瓦/鋪石的小巷,兩邊是低矮民居,遠處有個老人騎腳踏車經過。鏡頭以走路速度平穩前推。

Shot 3 (6-10s): 鏡頭推進到一間亮著暖黃燈的小店門口(咖啡館 / 豆花店 / 雜貨店),一位年輕旅人(亞裔女性,帆布包,米白毛衣)從畫面左側走入、停下來、抬頭看店招牌。鏡頭慢慢拉近她的背影後收。

Camera: Drone → walking pace → slow dolly-in. 順暢過場,無剪輯。
Lighting: Golden hour (06:00-07:00), warm orange cast.
Color: Earth tones, low saturation, film grain subtle.
Audio: 環境音(風、鐘聲、遠處腳步), no music.
Reference: Wes Anderson + Kogonada + 新海誠 live-action.

建築片的關鍵不是形容詞而是「機位高度+光的來源」——street-level eye-lin…

Ultra-modern minimalist house, concrete and glass facade, warm interior lighting, dusk environment, landscaped yard with soft ground lights, street-level eye-line, slow dolly-in camera motion, cinematic tone, detailed materials, elegant atmosphere.

景觀提案要秀的是動線不是造型,所以寫 wide pedestrian circulation…

Contemporary urban plaza concept, wide pedestrian circulation paths, young trees and seating modules, bright midday sunlight, gentle wind motion in foliage, bird's-eye camera pan, clean design hierarchy, realistic urban tones.

把「時間」當成提案賣點——用 transforming from A to B 指定起訖狀態…

Residential garden transforming from late summer to autumn, green foliage transitioning to warm gold tones, fallen leaves accumulating organically, smooth forward tracking camera, emotional natural lighting shift.

Vidu Q3 吃「場景→鏡頭→動作→色盤→負面詞」五段,每段一句話。動作寫 rotates…

Minimal studio scene, warm softbox diffused, glossy white acrylic surface with subtle reflections. 50mm feel, slow dolly-in, gentle parallax. The frosted glass bottle rotates 20 degrees as a thin mist drifts behind it, micro highlights bloom then settle. Color palette: sand, cream, soft gold: finish is clean, premium, calm. No labels warping, no jitter, no hard cuts.

i2v 拿真實產品照時,模型最愛「順手美化」把 logo 和外型改掉。這句用 Preserv…

Preserve the product shape, logo placement, camera angle, and clean background. Add a slow camera push-in, subtle shadow movement, and soft studio light reflections. Do not change the product design.

開頭一句 Stop-motion style animation 就把幀率與手感鎖死,後面「…

Stop-motion style animation. A pile of assorted LEGO bricks on a wooden table spontaneously begins to assemble itself. Gears turn, and bricks click into place, transforming into a small, intricate clockwork owl that then hops and looks around.

積木題材最怕爆炸被畫成一般火球,這條直接指定「炸開成一片慢動作的散磚雲」,材質邏輯自己交代完…

Cinematic wide shot of two intricately detailed LEGO spaceships engaged in a dogfight around a ringed planet. Lasers (streaks of light) fire across the scene. One ship explodes into a slow-motion cloud of individual LEGO bricks. The camera pans to follow the action.

Runway 官方範例一律四句——第一句只講機位、第二句只講場景、第三句只講動作、第四句只講…

Tracking shot moving parallel to white paper airplane. Modern office space, fluorescent lighting, desks with workers typing. Airplane glides smoothly through air at desk height. Shallow depth of field, motion blur on background.

機位鎖死,戲全部交給進站的列車——空月台先立一個乾淨的靜態底,再讓門一開露出滿車人,前後對比…

Static locked shot of empty subway platform. White tile walls, fluorescent lights overhead, yellow safety line. Train arrives and stops, doors open revealing crowded interior. Sound echoes in station.

修復類的爽點是「落差」不是「過程」,所以 before-and-after 直接寫在句首當片…

Vertical before-and-after video restoring an old bicycle from rust to polished chrome, fast restoration close-ups, workshop ambience and satisfying mechanical sounds.

piece by piece 是這條的引擎——它逼模型排出「一次裝一個」的節奏,而不是瞬間變…

Mechanical keyboard assembled piece by piece by robotic arms, macro close-ups, metallic textures, satisfying clicks and cinematic workshop ambience.

驗一張合成圖真不真,最快的方法是叫它動——鎖死機位、只讓肩胛動,墨跟著皮膚拉伸就是真的,浮在…

The person slowly rolls both shoulders back and lifts their arms slightly, then returns to a relaxed stand. The camera stays locked off. The skin stretches and the shoulder blades move under it, and the black spine tattoo stretches and compresses with the skin exactly as real ink would. Keep the same person, same skin tone, same lighting, same background, same tattoo design. No cuts, no zoom, no colour change, no new elements.

整條只有三件事——一個動作、三聲敲擊加一聲淬火、一種光。敲幾下寫成數字(Three ring…

Close-up of a blacksmith hammering a glowing horseshoe. Three rings of metal on metal, hiss of steam as the blade quenches, dim forge light. 8 seconds.

熔融材料的賣點是「自體發光+會被拉長」,glows and stretches 兩個動詞就把…

Macro cinematic shot inside a glassblowing workshop: molten glass glows and stretches as the artisan shapes a vase, furnace roar and crackle

畫面/運鏡與調色/音軌/規格四段分開寫,一條就是一支完整短影音的規格書;音軌把「烤箱門吱一聲…

Cinematic close-ups of a baker dusting flour over fresh dough at dawn, warm light through a steamy bakery window, hands kneading rhythmically, golden croissants sliding from the oven on a tray. Soft slow camera moves, shallow focus, warm amber grade. Audio: ASMR dough kneading, oven door creak, a gentle acoustic guitar, faint morning radio. 9:16, cozy artisanal mood, 4K.

開頭先給一句主題(built around the idea of starting aga…

Atmospheric brand film built around the idea of starting again. Show an empty bakery before sunrise as an owner unlocks the door, turns on the lights, prepares dough, and places the first bread into the oven. Slow cinematic camera movement with warm light gradually replacing the cool darkness. Audio includes keys, door movement, flour, kneading, oven mechanisms, and quiet city ambience. No dialogue. End as sunlight reaches the front window and the first customer approaches.

Ray3 吃的是「次級動態」——裙襬外翻、頭髮流動這種被主動作帶起來的連鎖運動,寫進去它才知…

Dancer spinning on a rooftop at sunset, dress billowing outward, hair flowing, city lights glowing in background, camera orbiting slowly.

backpack bouncing 與 breathing visible in cold …

Young woman in red jacket hiking up a mountain trail, backpack bouncing, hair tied back, breathing visible in cold air, golden hour side lighting.

把「成敗瞬間」寫成鏡頭任務(catching the upper hold perfectl…

A rock climber in climbing shoes and chalk-dusted hands makes a dynamic dyno move between holds on a steep overhang, catching the upper hold perfectly and locking off — dramatic slow-motion upward-facing camera from below the overhang — indoor competition bouldering wall with crowd watching: chalk dust visible in the air, camera catching the precise moment of hand contact with the hold, competitor expression of total focus.

直接寫專有技術名(Fosbury flop)+助跑曲線+起跳腳,模型不必猜姿勢;「過竿後橫竿…

A high jumper in a national team kit approaches the bar with a curved run, plants her lead foot, arches her back over the Fosbury flop position, and clears the bar cleanly — dramatic slow-motion side-on tracking shot — outdoor athletics stadium on a clear evening: bar wobbling slightly after the athlete clears, landing mat visible below, crowd rising in anticipation as the jump unfolds.

用「除了板刃切雪外一片寂靜」反向定義音軌,比列一串音效更逼得出空曠感;waist-deep …

A snowboarder in powder-specific gear drops off a cliff into waist-deep powder snow, carving massive arcs down a steep untouched slope — drone shot flying alongside and slightly above — perfect bluebird day in a remote mountain range: enormous powder plumes erupting with each turn, mountains stretching to the horizon, complete wilderness silence except for the board cutting through snow.

Google 官方用來示範「主體+動作+場景+打光+風格」五件套的標準答案;兩種光源(冷白日…

Medium shot, a tired corporate worker, rubbing his temples in exhaustion, in front of a bulky 1980s computer in a cluttered office late at night. The scene is lit by the harsh fluorescent overhead lights and the green glow of the monochrome monitor. Retro aesthetic, shot as if on 1980s color film, slightly grainy.

別用形容詞調表演,寫「她在哪、笑完什麼感覺」——同一個動詞換情境,模型自己生得出音量、時長與…

a woman laughs

a woman laughs long and loudly, she's in an office meeting and she's embarrassed afterwards

a woman laughs quietly, she's at home watching a tv show

原生音模型最愛自動加罐頭笑聲,得用 (no studio audience) 反向排除、再正…

a standup comic tells an awkward joke at a music festival, sounds of distant bands, noisy crowd, ambient background of a busy festival field (no studio audience)

自拍感的關鍵不是「selfie」這個詞,而是把「手臂必須清楚入鏡」寫成硬性構圖要求;末尾 (…

A handheld selfie-style shot, from the point-of-view of a gorilla in a lush jungle. A large silverback gorilla holds the camera at arm's length. His long, powerful arm is clearly visible in the frame, and his face is perfectly framed. The gorilla says: 'I'm just testing out this actually works and I'm going to post it on TikTok later, Essentially it felt cute might delete it later' (lips moving like he's saying it)

首尾幀只給了頭尾兩個端點、中間怎麼走全靠這句——把 match cut 的三個關鍵寫死:靠什…

Match-cut transition: the circular object in the first frame morphs seamlessly into the circular object in the second frame as the camera pushes in, motion-blurred transition at the midpoint, energetic pacing. A rising whoosh peaks on the cut.

before/after 最容易演成最後一秒硬換圖——`across the clip` 要…

Smooth transformation from the worn, dull version in the first frame to the restored, vibrant version in the second frame, camera slowly pushing in, colors enriching and details sharpening across the clip. Gentle uplifting ambience.

時間流逝要有證據——除了色溫從冷晨走到暖昏,另外給雲緩移與影子拉長兩個獨立的時間指標,才不會…

Static wide landscape shot. The lighting transitions smoothly from the cool dawn of the first frame to the warm dusk of the second, clouds drifting slowly, long shadows lengthening across the scene. Calm ambient wind and distant birds.

走秀片的質感全在打光——`harsh flash photography simulatio…

Fashion video of a model in a metallic silver oversized coat walking down a fog-filled studio runway, slow-motion stride, harsh flash photography simulation, dark studio background, confident expression, single magenta lip accent, Vogue 90s editorial style

POV 動作片最怕主體與威脅同框互相搶焦——第一人稱視角天然把「被追」變成畫面外的壓迫感,只…

A POV shot of someone snowboarding down a mountain while a dinosaur attacks him, slight blue tint, cinematic feel, fast action.

Grok Imagine 的原生音效要「被觸發」才準——公式是把聲音講出名字(board c…

The skateboarder lands the jump, board clatter synced to impact, dust kicks up

靜態人像也能有戲——呼氣是唯一的動作,冷空氣把這口氣變成看得見的證據;`locked cam…

She exhales slowly, a faint mist forming in the cold air, locked camera, dim ambient sound

眩暈鏡的定義就是「推軌前進同時變焦拉遠」,這句把兩個相反的動作+焦段區間(35mm→70mm…

Medium shot, dolly forward while zooming out. 35mm to 70mm equivalent. Background distorts while subject stays same size.

AI 生音樂

短影音 / 廣告片 / TikTok 用、15-20 秒內有 hook + drop

Style: upbeat {{genre}}-influenced pop, {{vocal}} female vocals with breathy emphasis, catchy 4-bar melody, big drop with synth bass + clap layer at 0:08, 120 bpm, dance-floor energy, modern radio pop production, mastered loud, target length 15-20 seconds, suitable for TikTok / Reels / YouTube Shorts intro hook
Lyrics: [Verse]
{{lyric_line_1}}
{{lyric_line_2}}
[Hook]
{{hook_line}}

80 年代都會感的中文 City Pop 情歌,合成器+放克貝斯,復古又抓耳。

[Style] 80s Japanese-style city pop sung in Mandarin Chinese, 105 BPM, lush synths, funky slap bass, gated reverb drums, dreamy vocal, nostalgic neon mood.\n\n[Structure]\n[Intro] 合成器前奏 4 小節。\n[Verse 1] 都會夜景畫面、[場景意象],中文歌詞 4 句。\n[Pre-Chorus] 情緒鋪墊 2 句。\n[Chorus] 抓耳記憶點,主題=[情歌主題],4 句、可重複。\n[Verse 2] / [Chorus] / [Outro] 漸弱。\n\n中文歌詞,風格=[歌詞風格]。

獨立遊戲 / pixel art 作品 / 懷舊廣告配樂 — 8-bit NES 風格 chiptune。60 秒 loop,可用 SNES 或 GameBoy 音色。

[Genre: chiptune, 8-bit NES-style game music] [BPM: 140] [Mood: [mood]] [Instruments: NES pulse wave 1 (melody), NES pulse wave 2 (harmony), NES triangle wave (bass), NES noise channel (percussion)] [Hardware: authentic 2A03 chip sound, no modern sample libraries] [Structure: 60 seconds — 0-15s: melody A with bass, 15-30s: melody B variation, 30-45s: melody A return with harmony layered, 45-60s: melody C bridge, ending frame matches opening frame for perfect loop] [Effects: no reverb (authentic NES had none), slight tape noise] [Scale: [scale]]

K-pop 特有的「三拍 + 高音爆發」hook 公式。TikTok / Reels 吃這個吃很久了。Udio 的韓語發音比 Suno 穩,也可以切英文。

[Genre: Modern K-pop, synth-heavy with trap drums]
[Duration: 15 seconds, designed as a chorus/hook loop]
[BPM: 128]
[Key: F# minor]
[Vocals: Female lead with multiple harmony layers, autotuned in a musical way]
[Instruments: Punchy 808 bass, trap hi-hats (rolling 1/32), bright lead synth, vocal chop sample on offbeats]

Hook structure (K-pop rule of 3):
0:00-0:03 — Entry phrase: 3 beats vocal + descending synth arp
0:03-0:08 — Main hook line, sung 3 times with slight variation:
  "[hook line]" (lower)
  "[hook line]" (same)
  "[hook line]" (higher, belt)
0:08-0:12 — Post-hook: vocal chop (no words) + drop
0:12-0:15 — Re-entry: "[brand name]" or "[tagline]" sung, setup to loop

Lyrics: Choose ONE language (Korean / English / Mandarin). Hook line should be 5-7 syllables.
Reference: NewJeans 'Hype Boy', IVE 'After Like', aespa 'Supernova'.

[Mix: Loud, tight low end, glossy high end, ready for TikTok]

拍 vlog 的最大瓶頸不是剪片是配樂。這條 prompt 產出純器樂的電影感 underscore,情緒推進曲線對 vlog 節奏剪得很順。

[Genre: Cinematic underscore, modern orchestral]
[Mood: Reflective → hopeful → resolved]
[BPM: 92]
[Duration: 2:30]
[Instruments: Felt piano lead, warm strings, subtle bass pulse, soft rhodes pad, no drums]

[Opening] 0:00-0:30 — Solo felt piano, sparse melody, leave space
[Build] 0:30-1:15 — Strings enter pp, bass pulse starts, slight crescendo
[Lift] 1:15-1:50 — Strings reach mf, piano moves to higher octave, hopeful harmony
[Resolve] 1:50-2:30 — Return to piano solo, end on held chord + room reverb

[Vocals: Instrumental]
[Mix: Cinematic, wide stereo, orchestral hall reverb, no compression on strings]
[Reference: Ólafur Arnalds / Max Richter / Nils Frahm]

No drums, no beat drop, no electronic elements. Organic and quiet.

需要客製?用免費產生器把你的需求組成 prompt —