七種 AI 貓貓創作指令:跳舞貓、名畫貓、LINE 貼圖,講個大概就做得出來
本來那天要認真工作的。結果我叫 AI 做了一整天的貓 🤣
沒辦法,壓力太大啦,需要貓貓療癒。
而且不是只有我這樣。去 IG 隨便搜一下,百萬、千萬觀看的貓貓影片到處都是。連 AI 做的卡通貓家庭短片,YouTube 上最高的一支都有 1.7 億次觀看。
真的是終極流量密碼無誤啦 🤣(好啦,誰的貓會紅,我沒辦法保證。)
我用 AI 把兩隻根本不存在的貓,玩了七種不同的東西:跳舞、進名畫、闖進我的口播、變成貼圖、演卡通、變成一雙拖鞋,還當了老師。
先講一件老實話:這七種,我其實都只跟 AI 講了個大概,細節是它幫我展開的。
所以下面每一種,我先寫「大概這樣講就可以」,再寫「想調細節,可以多補哪幾句」。簡單做,也做得出效果。
AI 展開後的完整指令,我收在每一種的最下面,想直接貼進生圖或影片工具的人再點開。

七種一次看
| 種類 | 做出來的東西 | 工具 | 我花了多少 |
|---|---|---|---|
| 跳舞貓 | 15 秒影片,四段接起來 | Kling(影片改影片) | 36 點 |
| 名畫貓 | 四張圖 | gpt-image-2.5 | 一張約 1~2 元 |
| 合成進口播 | 14 秒影片,五段接起來 | Kling(影片改影片) | 110 點,前後做了三輪 |
| LINE 貼圖 | 一張八格 | gpt-image-2.5 | 約 1~2 元 |
| 親子卡通 | 10 秒影片 | gpt-image-2.5 加 Kling | 25 點,另有一支沒用上的 25 點 |
| 商品圖 | 三張圖 | gpt-image-2.5 | 一張約 1~2 元 |
| 動畫貓老師 | 8.8 秒影片 | gpt-image-2.5 加 Claude 寫程式 | 3 張圖,約 3~6 元,沒用影片點數 |
想今天就動手,從圖片類開始。名畫貓、商品圖、貼圖都是一張圖幾塊錢,不用碰影片工具。
你需要準備什麼
- 一個能生圖、而且能上傳參考圖的 AI。我用的是 OpenAI 的 gpt-image-2.5,一張約台幣 1~2 元。這是用 API 照張數算的價錢,你用聊天工具內建的生圖,算法不一樣。
- Higgsfield 的帳號,做影片用。裡面有 Kling 這個影片模型,照點數扣。點數換算成多少錢要看你的方案,我這裡不亂換算。
- 一張貓的照片。沒有的話,照下一節生一隻。
- 第七種另外需要一個能自己寫程式、跑程式的 AI。我用的是 Claude Code。
想試 Higgsfield 的話
這篇的影片(跳舞貓、合成貓、親子卡通)我都是在 Higgsfield 裡做的。它把好幾家的影像模型放在同一個地方,我用的 Kling 就在裡面,不用一家一家去註冊。
想試可以從這裡進去:higgsfield.ai(我的推薦連結)
- 優惠:從這個連結登入之後的三小時內,訂閱方案最高五折。(這是 2026 年 10 月 13 日官方頁面寫的)
- 他們的優惠常常改,以你點進去當下看到的為準。
- 這是推薦連結。你從這裡訂閱,我會拿到分潤,你付的錢不會變多。
我這次總共花了多少
前六種加上貼文封面,連試作和沒採用的算進去,一共生了大約 30 張圖,其中十幾張沒用上。影片點數是 196 點:跳舞 36、合成貓 110(重做了三輪)、親子卡通 25,再加上一支做錯方向的 25。
我是怎麼操作的
我是在 Claude Code 裡請 Claude 幫我跑。指令大多是它寫的,生圖、生影片、把片段接起來,也是它去叫工具做。我負責講我要什麼,看了之後說哪裡不對。
你不一定要照我的方式。下面的英文指令是直接給生圖模型和影片模型看的,貼進你習慣的工具也可以試,出來的樣子會跟我的不太一樣。
每個工具的按鈕位置不同,卡住就把畫面截圖丟給你的 AI 問。
先做出你的貓主角
後面七種都靠同一件事:先有一張貓的定裝照。
之後每次生圖、生影片,都把這張當參考圖上傳,貓才會是同一隻。
我的主角有兩隻。一隻是全黑、金色眼睛的短毛黑貓,一隻是金漸層英國短毛貓。
你有自己的貓,可以直接拍一張正面、全身、背景乾淨的照片,跳過這一步。
我用什麼工具、花多少
gpt-image-2.5,一張約 1~2 元。我生了六隻候選,才挑出這兩隻。
大概這樣講就可以
幫我生一張貓的定裝照:【全黑、金色眼睛的短毛貓】,坐著看鏡頭,全身入鏡,背景乾淨。之後我要拿這張當參考圖。
把【】裡的字換成你自己的就好。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 臉要再圓一點。
- 大概六個月大。
- 不要項圈,也不要任何道具和字。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
黑貓:
A young solid-black domestic shorthair cat, about six months old, sleek glossy jet-black coat with no white fur anywhere, a very round face with chubby cheeks, big round golden-yellow eyes, black nose, black whiskers. Photorealistic casting photo of one single cat, sitting upright and facing the camera, full body in frame with a little space around it, looking straight into the lens with a curious, slightly cheeky expression. Plain warm off-white seamless studio backdrop and floor, soft natural window light from the left plus a gentle rim light so the black fur keeps visible texture and shape, gentle shadow on the floor. Shot on a 50mm lens at f/4, true-to-life fur detail, natural proportions of a real domestic cat. No collar, no props, no text, no watermark, no other animals.
金漸層:
A young British Shorthair golden chinchilla (golden shaded) cat, about six months old, plush dense golden fur with a cream chest, extremely round chubby cheeks, big round emerald-green eyes with dark eyeliner rims, small brick-red nose. Photorealistic casting photo of one single cat, sitting upright and facing the camera, full body in frame with a little space around it, looking straight into the lens with a curious, slightly cheeky expression. Plain warm off-white seamless studio backdrop and floor, soft natural window light from the left, gentle shadow on the floor. Shot on a 50mm lens at f/4, true-to-life fur detail, natural proportions of a real domestic cat. No collar, no props, no text, no watermark, no other animals.
要換成你自己的地方:
- 第一句的品種、毛色、年紀、眼睛顏色、鼻子顏色,換成你想要的貓。
sitting upright and facing the camera, full body in frame這段留著。正面、全身、乾淨背景,是定裝照的重點。- 最後一句
No collar, no props, no text, no watermark, no other animals也留著,免得多出項圈和字。
我踩到的坑
- 黑貓很容易變成一團黑影。所以黑貓那段多了一句
plus a gentle rim light so the black fur keeps visible texture and shape,意思是加一道輪廓光,讓黑毛看得出形狀。金漸層那段沒有這句。 - 第一批候選是橘貓、金漸層、賓士貓三隻,後來又加生三隻黑貓才定案。多生幾隻再挑,一隻才一兩塊錢。

一、跳舞貓:把正在紅的舞換成貓來跳
這是什麼
找一支正在紅的舞蹈影片,把裡面的人換成貓。動作、站位、背景、聲音都照原片,只有跳舞的變成貓。
我選的是閃身步。媒體報導,相關影片的全網播放量超過 6 億次。
照報導的說法,原片是北京舞蹈學院趙鐵春教授多年前錄的安徽花鼓燈教學影片。
我讓黑貓跳閃身步,金漸層跳同一支片裡的另一招「浪子踢球」。

我用什麼工具、花多少
- 貓的站立參考圖:gpt-image-2.5。兩隻貓各一張全身、一張臉部特寫,共四張,一張約 1~2 元。
- 換角色:Higgsfield 裡的 Kling 3.0 Omni Edit,就是影片改影片的功能,選 pro。一段只能 3~10 秒,每秒 2 點。
- 原片取 17 秒,切成四段做,一共 36 點。成品後來把開口前的空檔剪掉,剩 15 秒。最長的那段 7 秒多,14 點,等了大約 7 分鐘。
大概這樣講就可以
把這支影片裡跳舞的人換成我的貓(附上貓的照片),動作和背景都照原片,不要改。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 貓不要穿衣服,就是牠原本的毛。
- 影片太長的話,幫我切成幾段分開做,做完再接回去。
- 原片的聲音要留著。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
第一步,先讓貓站起來。把定裝照當參考圖上傳,貼這段:
Use the cat in the reference photo as the exact same character: same solid jet-black short glossy coat with no white fur anywhere, same very round face with chubby cheeks, big round golden-yellow eyes, black nose, black paws. Show him standing upright on his two hind legs like a person, facing the camera, front legs hanging relaxed at his sides like arms, hind feet about shoulder-width apart, full body from the tips of his ears to his hind paws inside the frame with space around. He is a completely natural cat with his own real fur all over: no clothes, no shoes, no accessories. Photorealistic fur, real cat head and paws, with an upright posture and limb proportions that would let him dance like a person. His tail is visible behind him. Plain light-blue studio backdrop wall and a glossy blue-grey dance floor with a faint reflection, soft even studio lighting that keeps the fur texture clearly visible. Photorealistic. No text, no watermark, no other characters.
原片有臉部特寫的鏡頭,就再多生一張臉部特寫當參考。第三種合成貓用的也是這張:
Use the cat in the reference photo as the exact same character: same solid jet-black short glossy coat with no white fur anywhere, same very round face with chubby cheeks, big round golden-yellow eyes, black nose, black paws. Chest-up close-up portrait, centered, standing upright and facing the camera with a calm, serious teacher-like expression, mouth slightly open as if he is about to say something. A completely natural cat with his own real fur: no clothes, no collar, no accessories. Plain light-blue studio backdrop, soft even studio lighting that keeps the fur texture clearly visible. Photorealistic real cat head and fur. No text, no watermark.
第二步,把原片照鏡頭切成 3~10 秒的小段。每一段連同參考圖一起上傳,全身的鏡頭配全身那張,特寫的鏡頭配特寫那張,貼這段:
Replace the man in the video with the black cat from the reference images: a real, photorealistic solid jet-black short-haired cat with a round face and big round golden-yellow eyes, standing upright on his hind legs. The cat performs exactly the same dance as the man: same footwork, same steps, same arm swings, same body lean, same timing and same position in the frame. The cat has his own natural glossy black fur all over, with no clothes, no shoes and no accessories, and a visible tail. Keep everything else exactly unchanged: the blue studio wall, the floor and its reflection, the lighting, the camera framing, and the on-screen caption graphic at the bottom left.
第三步,把做好的幾段照原片的長度接回去,再把原片的聲音放回來。
要換成你自己的地方:
- 每一段裡的毛色、眼睛顏色,換成你的貓。換另一隻貓上場,就把毛色、眼睛和參考圖一起換。
- 第一步的
Plain light-blue studio backdrop wall and a glossy blue-grey dance floor:原片是藍色的牆和藍灰色的地板,我讓參考圖的背景跟它一樣。換成你那支影片的場景。 - 第二步的
the man:換成原片裡跳舞的人。 - 第二步
Keep everything else exactly unchanged後面那一串:改成你那支影片裡要原封不動留著的東西。我的原片左下角有一塊招式名稱的字卡,所以特別把它寫進去。 - 第二步的
no clothes, no shoes and no accessories,第一步的參考圖指令也有一句差不多的。想讓貓穿衣服就拿掉。我做過穿衣服的版本,最後選不穿。
我踩到的坑
- 一段最短要 3 秒。原片有兩個特寫鏡頭只有兩秒多,得先把最後一格拉長、湊滿 3 秒再送,做完再裁回來。
- 出來的影片沒有聲音,原片的聲音要自己接回去。
- 長度會差一點。有一段原片 5.2 秒,出來是 5.04 秒,尾巴要用最後一格補。
- 黑貓站直的時候,毛色會偏深棕。
- 特寫那兩段,貓比原片的人大很多,整個佔滿畫面,嘴型也對不上唸招式名的聲音。我覺得動作有到就好,沒有重做。
- 送出前先看報價。同一段 7 秒,我查到另一種換物件的功能要 56 點,Kling 這個是 14 點。
版權提醒:閃身步是傳統民間舞的基本功,舞步本身沒有版權的問題。但我拿來改的那支教學影片,畫面和聲音是別人的。要不要這樣玩,你自己評估。
二、名畫貓:把名畫裡的人換成貓
這是什麼
挑一張大家都認得的名畫,構圖、顏色、筆觸照原畫,把畫裡的人換成貓,再加一個貓才會做的動作當笑點。
我做了四張:《拾穗》、《神奈川沖浪裏》、《星夜》、《吻》。貼文放前兩張,後兩張放在這裡當加碼。

我用什麼工具、花多少
gpt-image-2.5,一張約 1~2 元。上傳貓的定裝照當參考圖,兩隻都出場就傳兩張。
大概這樣講就可以
把【米勒的《拾穗》】裡的人換成我的貓(附上貓的照片),構圖和畫風都照原畫。
把【】裡的字換成你自己的就好。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 讓牠做一個貓才會做的動作,像是尾巴翹起來、伸掌去撥東西。
- 不要出現現代的東西,也不要字。
- 直式,比例 4:5。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
《拾穗》:
A humorous parody of Jean-François Millet's 1857 painting "The Gleaners", keeping the original composition, the warm golden harvested wheat field with stubble, the hazy pale sky, the big haystacks, the loaded cart and the tiny harvesters far in the background, painted in Millet's soft realist oil style with muted earthy colours, visible brushwork and aged canvas texture. The three peasant women are replaced by three cats in the same positions and the same bending poses, wearing the same simple peasant blouses, aprons, long skirts and headscarves. On the left, the solid black cat from the reference photo (round face, golden-yellow eyes) wears the blue headscarf and bends low, one black paw reaching for a single stalk of wheat on the ground, the other paw holding a small bundle of wheat behind his back. In the middle, the golden shaded British Shorthair from the reference photo (plush golden fur, very round cheeks, emerald-green eyes) wears the red headscarf and bends low in the same way. On the right, a tiny solid black kitten with big round eyes wears the yellow headscarf, only half bent, distracted from the work and happily chewing on a stalk of wheat he is holding in both paws. The two adult cats' tails stick straight up out of the back of their skirts. The cats have realistic heads, paws and fur but are blended into the painting's lighting and brush texture. Vertical 4:5 crop of the scene. No text, no signature, no frame, no modern objects.
《神奈川沖浪裏》:
A humorous parody of Hokusai's woodblock print "The Great Wave off Kanagawa", keeping the original ukiyo-e composition, flat Prussian-blue and cream colors, the claw-like foam of the great wave, the small fishing boats and Mount Fuji in the distance, with authentic woodblock print texture and paper grain. Hiding inside the curl of the giant wave is a huge solid black cat with round golden-yellow eyes (the cat from the reference photo), peeking over the crest with dilated pupils and one big black paw reaching out of the foam to swat the tiny boats like toys. The cat is rendered with realistic fur but blended into the print's flat lighting and line work. Vertical 4:5 crop of the scene. No modern objects, no extra text.
加碼,《星夜》:
A humorous parody of Vincent van Gogh's 1889 painting "The Starry Night", keeping the original composition, the swirling deep-blue night sky with glowing yellow stars ringed by halos, the sleeping village with the church steeple and the rolling blue hills, all painted in Van Gogh's thick impasto oil brushstrokes with visible directional strokes and canvas texture. The tall dark cypress tree in the left foreground is replaced by the solid black cat from the reference photo (round face, big golden-yellow eyes): he sits tall like the cypress, his black fur painted in the same flame-like swirling strokes, looking up with dilated pupils and stretching one front paw high into the sky to bat at the biggest swirling star as if it were a toy. The crescent moon in the top right is replaced by the golden shaded British Shorthair from the reference photo (plush golden fur, very round cheeks): he is curled up fast asleep in a glowing golden crescent shape, with the moon's soft yellow halo around him. Both cats keep recognisable faces and fur but are fully blended into the painting's brushwork, colours and lighting. Vertical 4:5 crop of the scene. No text, no signature, no frame, no modern objects.
加碼,《吻》:
A humorous parody of Gustav Klimt's 1908 painting "The Kiss", keeping the original composition, the shimmering gold-leaf background, the flower meadow at the bottom and the two ornate robes painted in Klimt's Art Nouveau oil-and-gold style with visible brushwork and aged canvas texture. The embracing couple is replaced by the two cats from the reference photos: the golden shaded British Shorthair (plush golden fur, round cheeks, emerald-green eyes) wears the golden robe with black-and-white rectangles and leans in to press his cheek against the solid black cat (golden-yellow eyes), who wears the robe with colorful circular flower patterns and looks slightly annoyed, pushing one paw against the golden cat's face the way cats do when they do not want a kiss. The cats' heads and paws are rendered with realistic fur but blend into the painting's lighting and brush texture. Vertical 4:5 museum-painting look. No text, no signature, no frame.

要換成你自己的地方:
- 開頭的畫名、畫家,還有
keeping the original composition後面那串原畫的特徵。《拾穗》寫的是麥田、草堆、馬車,《神奈川沖浪裏》寫的是浪花、小船、富士山。換一張畫,這串就換成那張畫最好認的東西。 - 誰站哪個位置。《拾穗》是照原畫三個人的位置,一隻一隻指定是哪隻貓、戴什麼顏色的頭巾。
- 那個笑點。四張各有一個:尾巴從裙子後面翹起來、躲在浪裡伸掌撥小船、伸掌去撥星星、被親的時候用掌推開對方的臉。換成你家貓最常做的事。
- 貓的毛色和眼睛。
Vertical 4:5:這是 IG 貼文的比例,要別的比例就改這裡。- 結尾的
No text, no signature, no frame, no modern objects留著。
我踩到的坑
- 只挑作者過世很久的畫。比較近代的畫家,作品可能還在著作權保護期內,我直接避開。
- 我先做《吻》和《神奈川沖浪裏》兩張試試看,覺得可以才加做《拾穗》和《星夜》。先試兩張再說,不用一次做滿。
- 四張都做了,貼文只放兩張。四張擠在一起,每一張都太小。
- 名畫配貓這個玩法很多人做過。你可以先搜一下別人用過哪些畫和梗,挑還沒人玩過的。
三、把貓合成進自己的口播影片
這是什麼
拿一支已經拍好的口播,請 AI 在畫面裡加一隻貓。人的臉、表情、嘴型都不動,貓走上桌之後一直煩我:拍我的麥克風、爬到我肩膀上,最後走過來把鏡頭整個擋住。
我用的是上一支老闆娘叫我用 AI 做宣傳片的口播。

我用什麼工具、花多少
- Higgsfield 裡的 Kling 3.0 Omni Edit,選 pro,跟跳舞貓是同一個功能。一段 3~10 秒,每秒 2 點。
- 輸入是一段口播影片,加一張黑貓的臉部參考圖。
- 我先拿一段 5 秒的測試,再做正式的五段,一共 48 點。後來覺得貓只是走過來坐著、沒有梗,又重做兩輪:一輪讓牠一直煩我(30 點),一輪讓牠每一句都張嘴喵(32 點)。三輪加起來 110 點,成品 14 秒。最後定案用的是第二輪那版。
大概這樣講就可以
在我這段口播影片裡加一隻黑貓(附上貓的照片),牠在桌上一直煩我。畫面裡的我完全不要動到。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 貓從左邊走上桌。
- 一直拍我手上的麥克風。
- 爬到我的肩膀上。
- 最後走過來,把鏡頭整個擋住。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
Add one real, photorealistic cat to this video: the solid jet-black short-haired cat with a round face and big golden-yellow eyes from the reference image. 【這裡寫貓的動作,例如:At the very start the cat is already stepping onto the wooden table at the bottom-left of the frame; it walks along the table toward the man with its tail raised high, reaches him and sniffs the hand that holds the small microphone.】 The cat has natural weight, real fur and a soft shadow on the table, and is lit by the same warm room light. Keep everything else exactly unchanged: the man's face, glasses, hair, expression, lip movements, shirt and hands, the microphone, the wall, the lamp, the plant, the shelf, the camera framing and the lighting. Do not change or redraw the man in any way. The cat never covers the man's face.
每一段只換中間括號裡那句貓在做什麼。我五段寫的動作是:
- 從左下走上桌,走向我,聞我拿麥克風的手。
- 一開始就在桌上,來回蹭我的手和前臂。
- 站起來,前掌搭在我胸口,頭蹭胸口和領口,不超過下巴。
- 坐在桌上靠著我,看鏡頭,慢慢眨眼。
- 坐著抬頭看我,再轉向鏡頭喵一聲。
這是第一輪的寫法。定案那版,我把後面四句改成:一直拍我的麥克風、對著我喵、爬到我肩膀上、走向鏡頭把畫面擋住。
要換成你自己的地方:
- 貓的描述,換成你的貓。
- 括號裡的動作。寫成英文再貼,可以請 AI 幫你翻。貼的時候把【】和裡面那句中文說明拿掉,只留你的英文動作。
the wooden table at the bottom-left of the frame:換成你的畫面裡貓可以站的地方。Keep everything else exactly unchanged後面那一串。我列的是我自己畫面裡的東西:臉、眼鏡、頭髮、表情、嘴型、襯衫、手、麥克風、牆、燈、盆栽、架子。照你的畫面重列一次。the man換成畫面裡的人。- 最後兩句
Do not change or redraw the man in any way. The cat never covers the man's face.留著。
我踩到的坑
- 先花一段的點數測試。我最怕臉被 AI 重畫,所以先拿最長的那段 5 秒去試,確認臉、表情、嘴型都沒變,才做其他段。
- 出來的影片可能沒有聲音,有也不可靠。要自己把原聲配回去。
- 貓不太聽細的動作。我寫蹭我的手,出來是貓站在我面前、尾巴翹得很高,沒有蹭。走過來、坐著、看鏡頭這幾種比較穩。
- 第二段開始,要寫明第一格貓就已經在桌上。不然每一段貓都會重新走進來一次。
- 每個剪接點,貓的位置會跳。一段最長 10 秒,只能一段一段分開做,貓在上一段結尾和下一段開頭不會剛好在同一個位置。
- 出來的影片每秒格數會變,長度也會跟原檔差 0.1 到 0.3 秒。要放回原本的剪輯,得先轉成跟原檔一樣。
- 我還試過幫貓配叫聲。先用 AI 生的貓叫,聽起來很假;換成真貓的錄音,又對不上牠的嘴。最後決定不配,沒有聲音反而比較好笑。
四、LINE 貼圖:一次八格
這是什麼
把兩隻貓重畫成可愛的手繪風,排成一張八格的貼圖。每一格一個動作,加一句中文。
我的八句:收到、好喔、不要、謝謝、累了、蛤?、愛你、晚安。
我用什麼工具、花多少
gpt-image-2.5,一張約 1~2 元。上傳兩隻貓的定裝照當參考圖。
大概這樣講就可以
用我的兩隻貓做一張八格的 LINE 貼圖(附上貓的照片),可愛的手繪風,每一格配一句聊天常用的話。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 八句話我要:收到、好喔、不要、謝謝、累了、蛤?、愛你、晚安。
- 黑貓臭臉,金色那隻愛撒嬌。
- 每一張要有白色的貼紙邊。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
A sheet of 8 LINE-style chat stickers arranged in a clean grid of 2 columns by 4 rows on a plain white background, each sticker clearly separated by generous white space. Cute hand-drawn kawaii illustration style with soft flat colours, clean thick dark-brown outlines and a thick white die-cut border with a faint grey drop shadow around each sticker; chibi proportions with big round heads. The two characters are the cats from the reference photos redrawn in this cartoon style: a solid black cat with a round face and big golden-yellow eyes, and a golden shaded British Shorthair with plush golden fur, a cream chest, very round cheeks and emerald-green eyes. Each sticker has bold rounded hand-lettered Traditional Chinese text, exactly as written here:
1. The black cat saluting with one paw, serious face. Text: 收到
2. The golden cat winking and raising one paw happily. Text: 好喔
3. The black cat turning his head away with half-closed eyes, one paw pushing outward. Text: 不要
4. The golden cat bowing politely with little sparkles. Text: 謝謝
5. Both cats melted flat on the floor like puddles, exhausted. Text: 累了
6. The black cat peeking out of a cardboard box with huge round eyes. Text: 蛤?
7. The golden cat rubbing his cheek on the black cat with little hearts, the black cat looking annoyed. Text: 愛你
8. Both cats curled up asleep together under a small blanket with a crescent moon. Text: 晚安
The Chinese text is large, legible and correctly written. No other text, no numbers, no watermark, no logos.
要換成你自己的地方:
- 八格的動作,還有每一格
Text:後面的字。挑你自己每天真的會傳的那幾句。 - 畫風那段,
Cute hand-drawn kawaii illustration style開頭的那一句。 - 兩個角色的描述。
- 格數和排法,
2 columns by 4 rows那裡。
我踩到的坑
- 中文字要一格一格放大檢查。我這次八格的字都對。指令裡是把每一句原封不動列出來,前面寫
exactly as written here,最後再補一句The Chinese text is large, legible and correctly written。 - 這張只是預覽圖,離真的上架還有一段。要一格一格裁開、去背,再照 LINE 官方的規格出檔。我還沒做到那一步。
- 上架的規格,還有 AI 生成的圖能不能上架,我都還沒查。要送審之前,請以 LINE Creators Market 的官方說明為準。
五、親子卡通:貓貓生活小教室
這是什麼
3D 卡通風的貓媽媽和小貓,演一個十秒的小意外。再加上「貓貓生活小教室」的標籤和中英對照字幕,看起來像一集兒童節目。
我的劇情:小黑貓想幫媽媽做蛋糕,勾倒了麵粉袋,整隻變成白貓。
這類卡通貓母子的短片,YouTube 上最高的一支有 1.72 億次觀看,這是 2026 年 10 月 13 日查的數字。最常見的劇情是小貓闖禍、媽媽崩潰或收拾,幾乎沒有對白。
提醒一下:這些片是不是 AI 做的,多數是從畫面判斷,只有少數頻道自己有標。

我用什麼工具、花多少
- 開頭、結尾兩張圖:gpt-image-2.5,一張約 1~2 元。
- 讓它動起來:Higgsfield 裡的 Kling v3.0,選 pro、含聲音、10 秒、直式 9:16,把兩張圖分別設成起始畫面和結尾畫面。25 點,等了大約 6 分鐘。
- 字幕:請 Claude 寫程式疊上去,不用點數。你用任何剪輯 App 加字也可以。
大概這樣講就可以
幫我做一支十秒的兒童卡通:小黑貓打翻麵粉,整隻變成白的,貓媽媽嚇一跳。3D 動畫電影那種風格,直式。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 先生開頭和結尾兩張圖給我看,我說可以再做成影片。
- 不要對白,只要音樂和喵叫。
- 幫我加中英文字幕,最後講一個小道理。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
開頭那張圖。上傳兩隻貓的定裝照當參考:
A frame from a 3D animated children's cartoon in a polished Pixar-like style, vertical 9:16. Cute chibi proportions: big round heads, small bodies, short limbs, huge glossy expressive eyes, soft stylised fur, warm cosy lighting, bright cheerful colours. Setting: a warm family kitchen with cream cabinets, a wooden counter, a window with morning sun and a few hanging pots. Two characters, redrawn from the reference photos as cartoon characters: the mother is a golden shaded British Shorthair cat with plush golden fur, a cream chest, very round cheeks and emerald-green eyes, wearing a simple pale-blue apron; she stands at the counter in the background on the right, cheerfully stirring a mixing bowl with a wooden spoon, not looking at the kitten. Her child is a tiny solid black kitten with a round head and big golden-yellow eyes, wearing a small white diaper; he stands on tiptoe on a little wooden stool in the foreground on the left, tongue sticking out in concentration, stretching one tiny paw up to hook the bottom corner of a big paper bag of flour that sits right at the edge of the counter above his head, the bag just starting to tip. The mother stays a cat with a cat's body, not a human figure. Calm empty space at the top of the frame. No text, no letters on the bag, no watermark, no humans.
結尾那張圖。把剛剛生好的開頭那張當參考圖上傳:
The same scene a moment later, as the next key frame of the same 3D animated children's cartoon: keep exactly the same kitchen, the same camera angle and framing, the same lighting, the same wooden counter and stool, the same two characters and the same polished Pixar-like chibi style as the reference frame. What has changed: the big paper flour bag has tipped over the edge of the counter and now hangs on its side, half empty, with flour still trickling down, and a soft white cloud of flour floats in the air. The tiny kitten in the diaper now sits on the wooden stool completely covered in white flour from the tips of his ears to his tail, so he looks like a fluffy little white kitten; only his two big round golden-yellow eyes and his tiny pink tongue show clearly, with a few small patches of black fur peeking through, and he blinks innocently up at his mother. A little heap of flour sits on his head, on the stool and on the floor around it. The golden mother cat in the pale-blue apron has turned to face him with huge shocked round eyes and a wide open mouth, both front paws pressed to her cheeks, the wooden spoon left standing in the mixing bowl. Vertical 9:16. No text, no letters, no watermark, no humans.
讓它動起來:
3D animated children's cartoon in a polished Pixar-like style, one continuous shot, fixed camera, no cuts. First three seconds: the tiny black kitten in a diaper wobbles on tiptoe on the wooden stool, tongue sticking out, and tugs the corner of the big paper flour bag while the golden mother cat happily stirs her mixing bowl. Then the bag tips over the edge of the counter and a big soft cloud of white flour whooshes down over the kitten with a poof. For the last five seconds the flour cloud settles: the kitten sits on the stool completely white, blinks his big golden eyes twice and gives one tiny sneeze that puffs out a little flour, while the mother cat turns around, gasps with huge round eyes and presses both front paws to her cheeks. Bouncy, squashy cartoon animation with expressive faces. Audio: playful light music, small kitten mews, a soft poof, a tiny sneeze and the mother's surprised meow. No dialogue, no human voices, no on-screen text.
字幕一共三句:
| 出現的時間 | 中文 | 英文 |
|---|---|---|
| 0.25~4.45 秒 | 小黑想幫媽媽做蛋糕 | Little Black wants to help Mom bake. |
| 4.75~7.25 秒 | 哎呀!麵粉倒了 | Oops! The flour fell down. |
| 7.45~10 秒 | 下次先問媽媽喔 | Next time, ask Mom first. |
左上角一個橘色的「貓貓生活小教室」小標籤,整支都在。
標籤下面是一張米白色的字卡,中文大字在上、英文小字在下。每一句有一個詞標成橘色:幫媽媽、哎呀!、先問媽媽。
字卡放左上,是因為媽媽的臉在右上、小貓在左下,那一塊剛好是空的。
要換成你自己的地方:
- 場景
a warm family kitchen和那個意外。開頭圖、結尾圖、影片三段指令講的是同一件事,要一起換。 - 誰當媽媽、誰當小孩,還有牠們的毛色和眼睛。
- 小貓的尿布、媽媽的圍裙。高觀看的同類片裡,小貓常常穿尿布,我是照著做的。不喜歡可以拿掉。
- 影片指令裡的秒數
First three seconds和For the last five seconds,還有Audio:後面那串聲音。 - 三句字幕。我的寫法是:發生什麼事、出包了、下次怎麼做。
我踩到的坑
- 第一支做成了寫實風,25 點。我要的其實是兒童卡通那種,是我沒先講清楚。那支現在只能當備用 💸
- 結尾那張圖,要拿開頭那張當參考圖來生,場景才會一致。
- 圖裡不要有字。兩張圖的指令最後都寫了
No text,開頭那張連麵粉袋上都寫了no letters on the bag。字幕全部事後再加。 - 字幕的時間要看成品來對。我的指令寫前三秒勾袋子,實際出來是前四秒,第五秒麵粉才倒下來。
- 十秒只夠演一個動作。我查到的高觀看同類片是 19 到 60 秒,十秒就挑一個瞬間。
- 聲音是模型自己配的。出片後你可以自己戴耳機聽一遍,再決定要不要用。
六、文創商品圖:一雙貓拖鞋
這是什麼
把貓主角變成一款商品,生出像選物店貼文的情境照。主圖、穿在腳上、細節特寫,三張一組。
我做的是一雙毛絨室內拖鞋,左腳黑貓、右腳金漸層。
先講清楚:這雙拖鞋不存在,三張都是 AI 生的示意圖。

我用什麼工具、花多少
gpt-image-2.5,三張,一張約 1~2 元。
大概這樣講就可以
幫我設計一雙貓拖鞋,一腳黑貓、一腳金色的貓,生一張放在臥室地板上的商品照。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 再一張穿在腳上的,只拍腳和鞋,不要拍到臉。
- 再一張刺繡的特寫,要跟第一張是同一款。
- 下面留一塊空白,我之後要放字。
AI 幫我展開後的完整指令(想直接貼進工具的人再點開)
第一張,主圖:
Photorealistic lifestyle product photo for an Instagram select-shop post, vertical 4:5. A pair of plush cat-shaped indoor slippers placed side by side, toes pointing toward the camera, on a warm light-oak wooden floor beside a cream knitted rug, next to a low bed with beige linen bedding, soft morning window light from the left. The LEFT slipper is a solid black cat: jet-black short plush coral fleece, two small upright ears at the toe, embroidered round golden-yellow eyes with dark pupils, a tiny embroidered nose and fine white whisker stitches, and a small stuffed black tail at the heel. The RIGHT slipper is a golden British Shorthair cat: plush golden-apricot fleece with a cream patch around the muzzle, two small rounded ears, embroidered round emerald-green eyes, a tiny brick-red embroidered nose, whisker stitches, and a short fluffy golden tail at the heel. Closed-toe slip-on design with beige non-slip soles, soft and chubby, clearly a real manufactured product with visible embroidery stitching and fabric texture. The slippers sit in the upper two thirds of the frame; the lower third is calm empty floor for a caption. No text, no logos, no people, no real cats.
第二張,穿在腳上。把主圖當參考圖上傳:
Use the exact same pair of plush cat slippers from the reference photo: one solid black cat slipper with embroidered golden-yellow eyes, white whisker stitches and a small black tail at the heel, and one golden-apricot British Shorthair cat slipper with embroidered emerald-green eyes, cream muzzle patch and a fluffy golden tail at the heel, both with beige soles. Show them being worn by a young woman, framed so that ONLY her feet, ankles and lower calves and the slippers are visible: no face, no upper body, no hands. She is sitting on the edge of a low bed with beige linen bedding, feet resting on a warm light-oak wooden floor next to a cream knitted rug, toes slightly turned toward each other in a cute relaxed way, bare ankles with loose cream knit lounge pants ending above the ankle. Slightly high three-quarter camera angle from the front so both cat faces on the slippers are clearly visible. Soft morning window light, natural skin, photorealistic lifestyle product photo, vertical 4:5. Calm empty floor in the lower third for a caption. No text, no logos, no real cats.
第三張,刺繡特寫。把前兩張當參考圖上傳:
Use the exact same pair of plush cat slippers from the reference photos: one solid black cat slipper with embroidered golden-yellow eyes, a small grey embroidered nose, white whisker stitches and a round black pom-pom tail at the heel, and one golden-apricot British Shorthair cat slipper with embroidered emerald-green eyes, a cream muzzle patch, a small pink embroidered nose and a golden pom-pom tail, both with beige soles. Show an extreme close-up macro product detail shot of the front of the black cat slipper, filling most of the frame: the satin-stitch embroidery of the two golden-yellow eyes with black pupils and a tiny white highlight stitch, the small embroidered nose, the fine white whisker stitches, and the soft dense black plush fibres around them, every thread clearly visible. The golden-apricot slipper sits just behind it, softly out of focus, with its green embroidered eye still recognisable. The slippers rest on a light-oak wooden floor. Soft morning window light from the left, shallow depth of field, photorealistic macro product photography, vertical 4:5. Keep the top quarter of the frame calm and uncluttered for a caption. No text, no logos, no labels, no real cats, no people.
要換成你自己的地方:
- 商品本身。
plush cat-shaped indoor slippers和後面整段材質、刺繡、尾巴的描述,換成你想做的東西,馬克杯、帆布袋、抱枕都可以。材質和做工寫細一點,主圖裡還有一句clearly a real manufactured product,要它看起來像真的做得出來的商品。 - 兩隻貓的特徵:毛色、眼睛顏色、鼻子顏色。
- 場景。我寫的是木地板、針織地毯、床邊、早晨的窗光。
- 留白的位置。主圖寫了下面三分之一留空,是要放文字的。不上字就拿掉。
- 第二張的
ONLY her feet, ankles and lower calves:只拍腳,不拍臉。
我踩到的坑
- 生出來的細節不會完全照你寫的。主圖我寫的是
a small stuffed black tail,一條小尾巴,出來是一顆毛球。所以到第三張,指令改成照主圖實際長的樣子寫,變成a round black pom-pom tail,鼻子的顏色也一起改,三張才會是同一雙。 - 一次一張。先把主圖定下來,後面兩張都拿它當參考圖。
- 三段指令都留著
no real cats。主題是貓,我會提醒自己加這句,免得畫面裡多一隻真的貓。 - 我會提醒自己:這是示意圖。真的要拿去賣東西,要換成實品照,不然圖和貨會對不上。
七、動畫貓老師:黑板上的字一行一行寫出來

這是什麼
一隻貓老師站在黑板旁邊,黑板上的粉筆字一行一行寫出來,像在上課。
我做出來的是 8.8 秒、可以一直循環的短片:黑貓老師戴眼鏡打領結,拿教鞭一行一行指,金漸層小貓在下面舉手。
我用什麼工具、花多少
- 教室場景圖:gpt-image-2.5,一張約 1~2 元。
- 寫字的動畫:請 Claude 寫程式做,不用影片點數。
我實際用了 3 張圖:一張教室底圖,一張只改老師眨眼張嘴,一張只改學生舉手。影片點數是 0。
大概這樣講就可以
做一支貓老師上課的短動畫:黑貓老師站在黑板旁邊,拿著教鞭,黑板上的字一行一行寫出來。黑板要寫的是:【你的三到五行字】。
把【】裡的字換成你自己的就好。
想調細節,就多補一句
不補也做得出來。想要更接近你心裡的樣子,就像這樣多講一句:
- 黑板上的字用程式寫上去,不要讓 AI 畫字。
- 旁邊加一隻小貓學生在舉手。
- 做成可以一直循環的。
比較完整的做法和指令(想照著一步一步做的人再點開)
做法是兩步:
- 先用生圖做一張教室場景。貓老師站在一邊,黑板要大,而且完全空白。
- 請 Claude 寫程式,把粉筆字一行一行寫到黑板上,輸出成影片。
黑板上的字不讓生圖模型畫,跟我做宣傳片那一篇的做法一樣:字用程式疊上去,中文才不會糊,之後也改得動。
下面這段是照我的做法整理的,我自己還沒拿它從頭跑完一次。每個人的環境不同,卡住就直接問 Claude 怎麼修。
把前三行換成你自己的內容,整段貼給 Claude:
我想做一支 10 秒左右的直式短片(1080×1920):一隻貓老師站在黑板旁邊,黑板上的粉筆字一行一行被寫出來。請一步一步帶我做,每做完一步先停下來給我看,我說可以再往下。
我的貓:___(例如:全黑、金色眼睛的短毛貓。有照片的話我會上傳)
這堂課的主題:___(例如:貓貓教你三個收納小訣竅)
黑板上要寫的字:___(三到五行,每行不超過十個字)
開始之前先問我兩件事:
- 你現在這個環境,能不能自己寫程式、跑程式,並且輸出影片檔?
- 我有沒有可以生圖的工具?
其中一樣沒有的話,先告訴我有哪些替代做法,不要直接往下做。
請照下面的順序做:
1. 場景圖:幫我寫一段生圖用的指令。畫面是教室,貓老師站在黑板的一側。黑板要佔畫面一大塊,上面完全空白,不能有任何字、圖案或粉筆痕跡。先不要畫教鞭,也不要畫老師靠黑板那隻手,這兩樣之後用程式畫,才指得準每一行。有貓的照片就當參考圖。圖生好先給我看。
2. 量位置:告訴我黑板在圖上的範圍,上下左右各到哪裡。之後的字只能寫在這個範圍裡,不能蓋到貓。
3. 寫字動畫:用程式把我給的字一行一行寫上去。要看得出是一筆一筆出現的,不是整行突然冒出來。字用粉筆質感的白色或淡黃色,底圖不動。
4. 出片:輸出 mp4。出片後自己抽幾格檢查:字有沒有超出黑板、有沒有蓋到貓、中文有沒有變成方框或亂碼。
限制:
- 圖裡不要用 AI 畫字。所有的字都用程式寫上去,之後我要改字,改程式就好。
- 字要避開畫面最上面和最下面,那裡會被 App 的介面蓋住。
- 有任何一步要花錢,先跟我說大概多少,我說好再做。
要換成你自己的地方:
- 前三行:你的貓、這堂課的主題、黑板上要寫的字。
- 片長和比例。要橫式或更長,改第一句。
我實際用的教室底圖指令長這樣。裡面把黑板、老師、學生的位置都用百分比講死,之後疊字才對得上:
A cute cartoon classroom scene, vertical 4:5 illustration. Cute hand-drawn kawaii illustration style with soft flat colours, a subtle crayon-like paper texture, clean thick dark-brown outlines and chibi proportions with big round heads — exactly the drawing style of the two cats in the reference image. Use the reference only for the character designs and the drawing style; do not copy its poses, its white sticker borders or its yellow accent marks.
Layout (important, follow it closely):
- A large classic green chalkboard with a light wooden frame hangs on a warm cream-coloured wall. It spans from near the left edge of the image to about 72% of the image width, and from about 7% down to about 55% of the image height. The chalkboard surface is completely EMPTY: a clean, flat, uniform dark green with only a very faint chalk-dust texture — absolutely no writing, no drawings, no lines, no doodles, no smudged letters. A thin wooden chalk tray with two chalk sticks and a small eraser runs along its bottom edge.
- The teacher: the solid black cat from the reference (all-black short fur, round face, big golden-yellow eyes), now wearing thin round wire-rimmed glasses and a green bow tie. He stands upright on the right side of the image, just to the right of the chalkboard, behind a small wooden lectern that hides his lower body. His head is between about 22% and 44% of the image height and he occupies roughly the right 27% of the image width. He is turned slightly toward the chalkboard (three-quarter view) with a gentle, friendly teacher expression, eyes wide open, mouth closed in a small smile. His paw farther from the chalkboard rests on top of the lectern. IMPORTANT: do NOT draw his other arm (the one nearest the chalkboard), and do NOT draw any pointer, stick, ruler or wand — that arm and the pointer will be added afterwards as a separate animation layer. That side of his body is simply plain black fur.
- The student: the golden kitten from the reference (plush golden-apricot fur, cream chest, round cheeks, emerald-green eyes) sits at a small wooden school desk in the lower-left corner, below the chalkboard, seen in front three-quarter view looking up toward the teacher, happily raising one paw to answer. His whole face is between 60% and 80% of the image height.
- The bottom 18% of the image shows only the wooden floor, the lower part of the lectern and the legs of the student desk — nothing important there.
Background: keep the wall simple and uncluttered (at most a small potted plant or the edge of a window), soft warm daylight.
STRICTLY NO text anywhere in the image: no letters, no numbers, no Chinese characters, no posters, no clock, no calendar, no labels, no book titles, no signature, no watermark.
我踩到的坑
- 教鞭不要讓生圖模型畫。它畫的是一個固定姿勢,對不準黑板上的五行字。我後來在底圖指令裡寫明不要畫那隻手和教鞭,改成程式畫,每一行都指得到。
- 想讓老師眨眼、學生換手,我是拿底圖當參考、請它只改一處。結果它其實整張重畫,位置偏了幾個像素,直接切換整個畫面會跳。解法是只把有變的那一小塊貼回底圖,其他地方都用原圖。
- 字型要先確認每個字都有。我一開始選的幾套字型沒有 ①②③,會變成空白,後來換了一套才齊。
- 黑板一定要先是空白的。每做完一步先停下來看,不要讓它一路做到底。
最後
這些都不難,不會寫程式也做得出來。簡單、好玩、又療癒。
你也可以用貓貓來創作喔 🤣
先做個貓貓踏踏馬殺雞來看看?
想試的話,從最便宜的開始:生一張你家貓的定裝照,再挑一張名畫。兩張圖,台幣不到五塊。
傳給你喜歡貓的朋友,一起來玩 🐾
資料來源
- TVBS 新聞網,2026 年 10 月 2 日,〈閃身步是什麼?陳念琴賽後魔性跟跳!由來、爆紅原因與膝蓋受傷風險一次看〉:https://news.tvbs.com.tw/life/4030565
- YouTube,Lovento Films,〈Cute Kitten of Cat is doing Makeup for the First Time〉,觀看數查詢日 2026 年 10 月 13 日:https://www.youtube.com/shorts/DiEJQ3ZgQvI
想用 AI 讓生活與工作更輕省,追蹤 AI 生活實驗室 👉 @life.coach.mtcity;想每天看見 AI 可以做什麼,追蹤 @ailifelab.co,我們一起玩 AI!