一組好用的萬聖節 AI 提示詞會做到四件事：說清楚要把你變成什麼、要求 AI 保留你的臉、用具體細節描述服裝和場景，並列出不要出現的東西。下面 25 組提示詞都照這個模式來寫。第 1 到 16 組需要一張你的照片（或是伴侶、朋友、寵物的照片）；第 17 到 25 組完全不用照片。文中的範例圖來自第 1 到 3 組，它們就是本站萬聖節特效背後實際使用的提示詞；其餘提示詞沿用同樣的結構，但每張照片做出來的效果都會不太一樣。
提示詞為什麼是英文？ 文中的範例圖就是用這些英文原句生成的，保留英文才能和範例對得上，也可以直接整段貼上使用。想調整細節時，用中文描述你要改的地方也可以。
建議先照原樣複製提示詞，再一次只改一個地方：服裝顏色、場景，或是光線。
萬聖節 AI 照片提示詞怎麼保留你的臉
大多數萬聖節 AI 照片都是用「圖生圖」做出來的：你上傳一張自拍，由提示詞描述新的造型。模型會重新畫出服裝、背景和光線，而照片負責告訴它你是誰。做出來還像不像你，取決於你的照片，但最關鍵的是提示詞裡的一句話。
以下是本站萬聖節女巫特效背後的完整提示詞，一字未改：
Turn the person in the photo into a magical witch for Halloween. Keep the person's face, facial features, skin tone, hairstyle and hair colour clearly recognisable as the same person in the photo. Wide-brimmed pointed hat, dark flowing robes with subtle embroidery, a softly glowing spell of purple light in one hand. Background of an autumn forest at night with a full moon, floating pumpkin lanterns and drifting leaves. Enchanting, warm and spooky, cinematic lighting. One subject only; no text unless asked for, no logos, no watermark, no extra people, no extra fingers or distorted hands.
它由五個部分組成，順序如下：
- 要變成什麼： "Turn the person in the photo into a magical witch."（把照片裡的人變成魔法女巫。）
- 保留句： 臉、五官、膚色、髮型和髮色都維持不變，而且每一項都要點名。只寫一句 "Keep my face"（保留我的臉），模型還是可能自行改動你的頭髮或膚色。
- 服裝： 寫出三、四個看得見的細節（帽子、長袍、發光的咒語），而不是只丟一個 "witch"（女巫）。
- 場景與光線： 你在哪裡、身邊有什麼、光線怎麼打。
- 排除項： 只有一個主體、不要文字、不要多餘的人、不要變形的手。
下面是這組提示詞用一張普通自拍做出來的結果，模型是 GPT Image 2.5 Flare，比例為 3:4 直式：
輸入的自拍，以及上方女巫提示詞的結果。她的臉、頭髮和膚色都延續了下來；帽子、長袍、咒語和森林則來自提示詞。
後面每一組照片提示詞都用同樣的順序。為了節省篇幅，它們用的是較短的保留句："Keep their face, features, skin tone and hair exactly as in the photo."（臉、五官、膚色和頭髮都和照片一模一樣。）如果做出來的臉越來越不像你，就換回女巫提示詞裡那句較長的版本。
自拍變裝用的萬聖節 AI 提示詞
上傳一張清楚的個人照，臉朝向鏡頭，再貼上提示詞。3:4 或 2:3 的直式比例很適合拿來當大頭貼或發文。
1. 魔法女巫
直接使用上面的女巫提示詞即可。
2. 哥德風吸血鬼
這是本站吸血鬼特效背後的提示詞，一字未改。其中 "classy and atmospheric, not gory"（優雅、有氛圍，不要血腥）這一句，讓整體造型保持優雅，而不是血淋淋的。
Turn the person in the photo into an elegant gothic vampire for Halloween. Keep the person's face, facial features, skin tone, hairstyle and hair colour clearly recognisable as the same person in the photo. Pale porcelain skin, subtle crimson eyes, small fangs visible in a slight smile, dark velvet high-collared cloak over Victorian clothing. Moody candle-lit castle interior, deep reds and blacks, soft fog, cinematic low-key lighting. Classy and atmospheric, not gory. One subject only; no text unless asked for, no logos, no watermark, no extra people, no extra fingers or distorted hands.
輸入的自拍，以及吸血鬼提示詞的結果。鬍子、頭髮和臉型都保留下來；膚色變得稍微蒼白，眼睛也帶點紅色，正如提示詞所要求的。
3. Ghostface 拍立得（《驚聲尖叫》熱潮）
Ghostface 熱潮的玩法，是把《驚聲尖叫》（Scream）系列電影裡的面具殺手放到你身後，做成一張 90 年代末恐怖片的劇照。這組提示詞是本站 Ghostface 特效三個場景中的其中一個，稍微縮短過；下面這張圖則是用特效的完整提示詞做出來的。
Turn this photo into a still from a late-1990s horror movie. Keep the person's face, facial features, skin tone, hairstyle and hair colour clearly recognisable as the same person in the photo; do not retouch or reshape the face. An instant Polaroid photo with a white border, shot with harsh on-camera flash in a dark room against a plain white curtain. The person looks into the camera, startled; right behind their shoulder Ghostface leans in close, one gloved hand resting playfully on their shoulder. Ghostface wears a black hooded robe with flowing sleeves, black gloves and the white elongated screaming ghost mask with tall hollow black eyes and a long open mouth. Casual 1990s clothes, grainy 35mm film look, Y2K colours, moody low light. Eerie but playful. Only the person and the masked figure; no text, no logos, no watermark, no blood, no weapons, no extra fingers or distorted hands.
Ghostface 特效完整提示詞裡的拍立得場景，用一張普通自拍生成。"No blood, no weapons"（不要血、不要武器）讓畫面維持詭異，而不會變得暴力。
想做經典版的話，把拍立得那一句換成："The person lies on a satin bed in a 1990s teenage bedroom at night, holding a corded landline phone, a bowl of popcorn on the bed. Behind them, in the dark hallway doorway, Ghostface stands half in shadow, watching."（夜裡，這個人躺在 1990 年代青少年房間的緞面床上，手拿有線市內電話，床上放著一碗爆米花；身後昏暗的走廊門口，Ghostface 半隱在陰影中，靜靜看著。）
4. 嚇人但不噁心的殭屍
Turn the person in this photo into a movie-style zombie for Halloween. Keep their face, features and hair exactly as in the photo, so friends can still tell who it is; only the skin colour changes. Pale grey-green skin, dark circles under the eyes, a few small scratches, torn and dusty everyday clothes. Foggy abandoned street at night, flickering streetlight, cold blue moonlight. Spooky, cinematic, not gory: no open wounds, no blood. One person only, no text, no watermark.
5. 骷髏臉彩繪
Keep the person's face, features, skin tone and hair exactly as in the photo, and add detailed skeleton face paint for Halloween: white base on the cheeks and forehead, black around the eyes and nose, painted teeth over the lips. Black hoodie with a skeleton print. Dark studio background with an orange rim light. Sharp, high-contrast portrait. One person only, no text, no watermark.
6. 變身到一半的狼人
Turn the person in this photo into a werewolf in the middle of transforming. Keep their face, eye colour, skin tone and hairstyle recognisable as the same person. Thick fur growing along the jaw and ears, slightly pointed ears, amber glowing eyes, torn flannel shirt. Misty pine forest under a huge full moon. Dramatic moonlit lighting, cinematic. One person only, no text, no watermark.
7. 維多利亞時代的幽靈
Turn the person in this photo into a Victorian ghost. Keep their face, features and hairstyle recognisable as the same person. Translucent pale skin with a soft blue glow, a high-necked white lace dress or suit faded with age, faint wisps of mist around the shoulders. Dim hallway of an old mansion lit by a single candle, dusty portraits on the walls. Eerie, quiet and elegant. One person only, no text, no watermark.
8. 玉米田裡的稻草人
Turn the person in this photo into a friendly Halloween scarecrow. Keep their face, features, skin tone and hair exactly as in the photo. Straw poking out from the collar and cuffs, a patched plaid shirt, denim overalls, a worn straw hat, a few painted freckles. Golden cornfield at sunset with crows on a fence and carved pumpkins on hay bales. Warm, rustic autumn light. One person only, no text, no watermark.
9. 暗黑精靈
Turn the person in this photo into a dark fairy for Halloween. Keep their face, features, skin tone and hair colour exactly as in the photo. Large translucent black wings with violet veins, a crown of thorns and dried roses, a dark green velvet dress or tunic. Enchanted forest at night with glowing mushrooms and fireflies. Magical, moody and beautiful. One person only, no text, no watermark.
10. 由你主演的復古恐怖片海報
Turn this photo into a 1980s horror movie poster with the person as the star. Keep their face, features, skin tone and hair exactly as in the photo. The person looks over their shoulder in fear; behind them a dark old house on a hill with one lit window, under a red sky. Painted poster style, film grain, dramatic lighting. Leave the top third dark and empty for a title. No text, no logos, no watermark.
如果想在海報上放片名，就用引號把它加進去：Title text at the top: "THE NIGHT SHIFT"。片名越短，生成出來的字越穩定。
11. 療癒系女巫風
如果你想要的是萬聖節溫柔、可愛的一面，而不是嚇人的那一面：
Keep the person's face, features, skin tone and hair exactly as in the photo. They sit by a rainy window in a cosy cottage, wearing an oversized knitted cardigan and a small black witch hat, holding a mug of hot cider. Candles, stacked old books, dried herbs hanging from a beam, a black cat curled up on the windowsill, small pumpkins. Soft warm light, film photo look, autumn colours. One person only, no text, no watermark.
12. 搞笑版：累壞的南瓜頭通勤族
Keep the person's body, clothes and pose exactly as in the photo, but replace their head with a carved jack-o'-lantern with a tired, grumpy face and a soft glow inside. Put them on a crowded morning subway train, holding a coffee cup and looking at their phone, while other passengers ignore them. Realistic photo, natural light, deadpan and funny. No text, no watermark.
這一組刻意不保留你的臉，所以最適合用在服裝和姿勢一看就認得出是你的照片上。
13. 搞笑版：在門口發糖果給一群小怪獸
Keep the person's face, features, skin tone and hair exactly as in the photo. They stand in their front doorway on Halloween night holding a huge bowl of candy, looking nervous, while a crowd of small cute monsters in trick-or-treat bags looks up at them: a tiny vampire, a little mummy, a fluffy werewolf pup. Porch lights, carved pumpkins on the steps. Bright, playful, Pixar-like lighting. No text, no watermark.
情侶、朋友和寵物的萬聖節 AI 提示詞
兩個人以上時，上傳一張所有人都在裡面的合照，再依照每個人在照片中的位置指定服裝。模型不會知道「阿傑」是誰，但它看得出誰站在左邊。
14. 吸血鬼情侶
Turn both people in this photo into a gothic vampire couple. Keep each person's face, features and hair exactly as in the photo; only the skin turns a little paler. The person on the left wears a black velvet high-collared cloak; the person on the right wears a deep red Victorian gown or waistcoat. Pale skin, subtle crimson eyes, small fangs in slight smiles. They stand close together on a stone balcony of a castle at night, candles behind them, full moon. Classy and atmospheric, not gory. Only these two people, no text, no watermark.
15. 朋友變身經典怪物小隊
Keep every person in this photo recognisable: same faces, features, skin tones and hair. Dress them as a classic monster squad, from left to right: a vampire, a mummy wrapped in bandages with the face uncovered, a witch and a werewolf with fur on the jaw. Keep their poses and positions from the photo. Group shot in front of a haunted house with glowing pumpkins on the steps. Fun, cinematic, slightly spooky. Only the people in the photo, no text, no watermark.
16. 讓你家的狗或貓變裝
Keep the pet in this photo exactly the same: same breed, fur colour, markings and face. Dress it in a tiny Halloween costume: a little black cape with a high red collar, like a small vampire. Sitting on a porch step next to a carved pumpkin and a bowl of candy, warm porch light, autumn leaves. Cute and funny, sharp detail on the fur. No people, no text, no watermark.
斗篷也可以換成南瓜裝、女巫帽或蝙蝠翅膀。真正讓牠看起來還是你家寵物的，是寫明品種、毛色和花紋的那一句。
不用照片的恐怖萬聖節 AI 提示詞
這些是文字生圖的提示詞：不必上傳任何東西，直接貼進生成器就好。
17. 維多利亞式鬼屋
A haunted Victorian mansion on a hill at night, broken shutters, one candle-lit window on the top floor, bare twisted trees, a crooked iron gate, fog rolling across overgrown grass, a full moon behind thin clouds. Dark, cinematic, highly detailed, blue and amber light. No people, no text.
18. 南瓜燈靜物
Close-up still life of three carved jack-o'-lanterns on an old wooden table, each with a different face: happy, scary and surprised. Candlelight glowing from inside, wax drips, dried leaves, a few small gourds, dark background with soft bokeh. Warm, moody food-photography lighting. No text.
19. 女巫的藥草鋪
Interior of a witch's apothecary shop: crowded wooden shelves of glass bottles with glowing potions, jars of herbs and mushrooms, an open spell book on the counter, a bubbling cauldron, a black cat sitting among the bottles. Warm candlelight with green and purple glows. Rich detail, cosy and magical. No people, no readable text on labels.
20. 復古萬聖節明信片
A vintage 1920s Halloween postcard illustration: a smiling black cat on a crescent moon, bats, a jack-o'-lantern and a witch flying on a broom in the distance. Muted orange, black and cream colours, printed paper texture with slight wear on the edges. Charming and old-fashioned. No text.
21. 萬聖節派對邀請卡
A Halloween party invitation card, portrait layout. Dark purple background, a border of small bats and orange pumpkins, a glowing moon at the top. Large title text: "HALLOWEEN PARTY". Below it, smaller text: "Saturday, Oct 31 · 8 PM". Elegant spooky style, clean layout with readable lettering, plenty of space around the text.
卡片上每個想出現的字都要放進引號，而且越短越好。如果需要清楚好讀的文字，本站推薦用 GPT Image 2 來生成圖片裡的文字。寄出之前，記得檢查結果裡的拼字。
22. 可愛幽靈貼紙
A sticker sheet of eight cute kawaii Halloween characters on a plain white background: a smiling ghost, a sleepy bat, a pumpkin with blushing cheeks, a tiny witch hat, a candy corn, a black cat, a mummy, a cauldron. Thick white sticker borders, flat colours, soft shading, evenly spaced. No text.
23. 經典恐怖片海報
A 1970s horror movie poster painting: a lonely farmhouse at the end of a long dirt road, a dark figure in a field of tall corn, a blood-red sunset sky, crows overhead. Painted illustration style, film grain and faded print colours. Leave space at the bottom for a title. No text, no logos.
24. 萬聖節手機桌布
A Halloween phone wallpaper, vertical 9:16. A quiet street of old houses at dusk with carved pumpkins glowing on every porch, autumn leaves on the pavement, a purple and orange sky, a few bats in the distance. Leave the top third simple and dark for the clock. Soft, cosy and slightly spooky, detailed. No people, no text.
除了寫在提示詞裡，也要在生成器裡把比例設成 9:16。
25. 暗黑美感俯拍擺設
Top-down flat lay on a dark slate surface: a small pumpkin, dried roses, a brass key, tarot-style cards with plain backs, a black candle, scattered candy corn, a lace doily, autumn leaves. Moody, soft window light, muted orange and black tones, aesthetic Instagram style. No people, no text.
在哪裡使用這些萬聖節 AI 提示詞
在本站：
- 照片提示詞（第 1 到 16 組）： 打開 AI 圖片編輯器的圖生圖生成器，上傳照片、貼上提示詞，再選一個模型。GPT Image 2.5 Flare 是不錯的預設選擇：上面女巫、吸血鬼和 Ghostface 的範例都出自這個模型，它也是本站推薦用來讓同一個角色維持一致的模型。
- 不用照片的提示詞（第 17 到 25 組）： 不上傳任何東西，直接貼進首頁的生成器。派對邀請卡，或任何圖裡需要文字的作品，請用 GPT Image 2。
- 比例： 在生成器裡選擇。肖像用 3:4 或 2:3，手機桌布或限時動態用 9:16，貼紙用 1:1。可選的比例依模型而定。
你也可以把同一組提示詞再丟給第二個模型，比較兩個結果後再決定要哪一張。
新帳號送 10 積分。每張圖的價格取決於模型和解析度，開始前就會顯示在「生成」按鈕上。免費生成的圖片會以該模型最低的解析度輸出，並帶有浮水印，僅供個人使用。購買任何方案或積分包，就能去除浮水印並解鎖更高的解析度。
不想自己寫提示詞的話
女巫、吸血鬼和 Ghostface 的提示詞已經做成一鍵特效：你只要上傳照片，其他什麼都不用做。
- 萬聖節女巫特效：女巫造型
- 萬聖節吸血鬼特效：哥德風吸血鬼肖像
- Ghostface 照片特效：臥室接電話、拍立得或對鏡自拍三種場景，還可以另外加一句簡短的要求，例如 "a red hoodie"（一件紅色帽 T）
- 萬聖節變身短片：你變身成吸血鬼的短影片
- 想看更多變裝造型，可以到特效總覽找所有萬聖節特效
每個圖片特效 10 積分，所以新帳號的免費積分夠試一次。影片短片的費用更高。
萬聖節 AI 做出來不像你時怎麼辦
臉變了。 改用女巫提示詞裡那句長版保留句，並放在第一句之後。上傳更清晰的照片：只有一個人、臉朝鏡頭、光線充足、沒戴墨鏡。合照和側臉能讓模型參考的五官比較少。
頭髮或膚色變了。 把它們寫進保留句。服裝用詞可能會連帶改變它們（例如 "pale vampire"，也就是「蒼白的吸血鬼」，就是刻意改變膚色），所以要說清楚哪些改變是你要的、哪些不要。
多出其他人，或出現第二個 Ghostface。 像上面的提示詞一樣，在結尾加上 "one person only"（只有一個人）或 "only the person and the masked figure"（只有這個人和面具人）。
太血腥或太卡通。 直接寫出你要的調性："spooky but not gory"（恐怖但不血腥）、"classy and atmospheric"（優雅、有氛圍），或是 "realistic photo"（寫實照片）、"cute and playful"（可愛俏皮）。
文字拼錯了。 減少字數、把文字放進引號，並改用 GPT Image 2 試試看。
整體都很接近，只有一個細節不對。 保留原本的提示詞，只改那個細節。整組重寫的話，通常連你喜歡的部分也會跟著變。
萬聖節 AI 提示詞常見問題
萬聖節 AI 熱潮要怎麼玩？
把一張清楚的自拍上傳到圖生圖工具，貼上一組萬聖節提示詞（例如上面的女巫、吸血鬼或 Ghostface 提示詞），然後按下生成。這波熱潮的圖片都有同樣三個元素：你自己的臉、一套服裝，以及一個電影感的場景。不想寫提示詞的話，上面的一鍵特效只要一張照片，就能做到同樣的效果。
這些萬聖節 AI 提示詞是免費的嗎？
是的，25 組提示詞都可以免費複製和修改。在本站生成圖片需要積分；新帳號送 10 積分，大約可以用 GPT Image 2.5 Flare 以最低解析度生成三張圖，足夠在付費前先試幾組提示詞。
這些提示詞可以用在 ChatGPT 或 Gemini 嗎？
可以。它們都是一般的英文描述，沒有任何本站專用的代碼。不同模型的結果會有差異，可能需要調整一兩個細節，但整體結構（要變成什麼、保留句、服裝、場景、排除項）都可以沿用，而且不論用哪個工具，保留句都一樣重要。
有哪些好用的恐怖風格 AI 提示詞？
指定一個電影年代和鏡頭質感（例如 "late-1990s horror movie still, grainy 35mm film"，也就是 90 年代末恐怖片劇照、帶顆粒感的 35mm 底片），只描述一個帶有威脅感的細節，而不是一大堆（例如門口的人影、一扇亮著燈的窗），再把光線壓暗。第 3、10、17 和 23 組都是很好的起點。
祝你萬聖節快樂，挑一張最像你的發出去吧。