想讓 AI 影片裡一直是同一個人?
我的做法:先出參考圖,每一段都重新把參考圖給它,再發 10 秒的提示詞。
這一支我覺得品質還是很高的。但人物一致性有隨機性,要抽卡。
提示詞(先傳三張參考圖,再發):
The 3 attached images show the same woman, Xiao Man. Use them as appearance references only.
Create a 10-second vertical 9:16 1080p video. The camera is her boyfriend's phone; she talks to the lens.
PERSON: "Xiao Man", a 26-year-old adult Chinese woman, shoulder-length black hair with soft natural waves, oval face, light natural makeup, realistic skin texture, a small mole under her left eye, slim build. Keep her face and hair identical to the attached images.
VOICE: a soft, bright young female voice, natural conversational Mandarin Chinese, playful and warm, not a broadcast voice.
OUTFIT: cream knit cardigan over a white top, light jeans.
SCENE: She sits across a café table by the window. She pushes a slice of cake toward the camera, then rests her chin on her hand and watches.
DIALOGUE (spoken on camera in Mandarin, accurate lip sync):
00:01 "你嘗一口,這個真的很好吃。"
00:06 "怎麼樣?我沒騙你吧。"
Real handheld phone footage, soft daylight, natural shake.
Audio: her voice and quiet café ambience. No music, no subtitles, no on-screen text, no watermark.
參考圖的提示詞和兩版 30 秒的對比,在長文裡 👇
https://t.co/YCCUWUx91J
1995 年夏天的家庭錄影帶質感,Muse 生成的。
畫面的時代感非常強,配音的部分還是有欠缺。
提示詞:
Create a 10-second horizontal 16:9 video that looks like real home-video footage from 1995 in a southern Chinese city: soft VHS texture, faded colors, slight tracking noise, 4:3 picture inside the frame.
Summer evening on an old street: adults sit on bamboo chairs outside their doors fanning themselves, a man rides past on a bicycle with a block of ice on the back, a small shop has a freezer by the door, an old TV glows inside.
Handheld camcorder feel.
NARRATION (warm, nostalgic middle-aged male voice, Mandarin Chinese, voice-over): "那時候的夏天,是從一根冰棍開始的。"
Audio: the narration, cicadas, bicycle bell, distant TV sound. No on-screen text, no date stamp, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUx91J
冰箱裡一盒牛奶開口抱怨:「我已經在這裡躺了三個星期了。」
Muse 免費額度,10 秒。
提示詞:
Create a 10-second vertical 9:16 video, photorealistic with a cartoon twist.
Inside an open fridge, a carton of milk with a small animated face and tiny arms sits at the back of the shelf, looking grumpy. It sighs, looks at the camera and complains. A tomato next to it rolls its eyes.
VOICE (tired, sarcastic middle-aged male voice, Mandarin Chinese, the milk carton is speaking with lip sync): "我已經在這裡躺了三個星期了。"
Audio: the voice and fridge hum. No music, no readable text on any package, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUxGRh
第一人稱視角,一覺醒來在唐朝的長安。
Muse 免費額度做的 10 秒,旁白是修過的版本。
提示詞:
Create a 10-second vertical 9:16 first-person POV video, photorealistic.
You wake up in a wooden bed in Tang dynasty Chang'an. Your own hands push open a carved wooden window: below is a busy morning street with merchants, camels, lanterns and steaming food stalls. A servant in the doorway bows to you.
The camera is your eyes: natural head movement, your hands and silk sleeves visible in the frame.
VOICE (your own inner voice, confused young male, Mandarin Chinese): "這裡……是長安?"
Audio: the inner voice, street noise, distant bells. No music, no on-screen text, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUx91J
水墨畫裡的蘇東坡,Muse 用免費額度做的。
畫面沒有問題,第一版配音少配字、還讀錯音,這是讓它自己修完的版本。
提示詞:
Create a 10-second horizontal 16:9 video in Chinese ink-wash painting style that comes alive: flowing brush strokes, rice-paper texture, black ink with light touches of color.
A Song dynasty scholar in a long robe stands alone at the bow of a small boat on a wide river at night, under a full moon, cliffs rising on both sides; his sleeves move in the wind.
NARRATION (deep, calm male voice, Mandarin Chinese, voice-over): "公元一零八零年,蘇軾被貶到了黃州。"
Audio: the narration, water, wind, a soft guqin melody. No on-screen text, no seals, no subtitles, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUxGRh
對著鏡頭講生活小知識的真人主播,也是 Muse 生成的。
神態和表情我覺得品質非常高,最後還有一個關手機的動作。
提示詞:
Create a 10-second vertical 9:16 photorealistic video.
A friendly 40-year-old Chinese woman in a light blue shirt stands in a bright home kitchen, holding a bowl of leftover rice, and talks directly to the camera like a short-video host.
Tripod shot, medium close-up, natural daylight.
DIALOGUE (clear, friendly female voice, standard Mandarin Chinese, spoken on camera with accurate lip sync): "隔夜飯到底能不能吃?今天一次講清楚。"
Audio: her voice and quiet room sound only. No music, no subtitles, no on-screen text, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUx91J
火柴人講解類的影片,Muse 也能做。
畫面一次就出來了,旁白是後來讓它修過的版本。
提示詞:
Create a 10-second horizontal 16:9 explainer animation in simple stick-figure style: black line-drawn stick figures on a plain white background, minimal flat colors, clean and funny.
A stick figure runs on a hamster wheel while coins fall out of a hole in his pocket; he stops, looks at his empty wallet and scratches his head with a question mark above him.
NARRATION (calm, slightly humorous male voice, Mandarin Chinese, voice-over): "為什麼你越忙,越存不下錢?"
Audio: the narration and light cartoon sound effects. No written words or numbers on screen, no subtitles, no watermark.
完整實測和全部提示詞在長文裡 👇
https://t.co/YCCUWUx91J
Muse 做的恐龍紀錄片,畫面和音效我覺得品質非常高。
但第一版旁白把「六千六百萬年前」唸成了 600 萬年前。
我讓它自己修,這是修完的版本。
提示詞:
Create a 10-second horizontal 16:9 video in the style of a high-end nature documentary, photorealistic, cinematic.
Late Cretaceous: a herd of triceratops drinks at a river at dusk. In the sky, a bright streak appears and grows; the dinosaurs lift their heads toward it.
Slow, steady camera.
NARRATION (deep, serious male documentary voice, Mandarin Chinese, voice-over): "六千六百萬年前,一顆小行星改變了一切。"
Audio: the narration, river, low dinosaur calls, a low orchestral drone. No on-screen text, no watermark.
它是怎麼修的、修配音的提示詞,都在長文裡 👇
https://t.co/YCCUWUxGRh