指南 / 影像製作

製作爆紅的懷舊風格 AI 影片

那些讓人停下滑動的影片看起來就像鄰居的舊家庭錄影:一個平凡的地方、一台廉價的相機,以及一個不該出現在那裡卻被視為正常的事物。以下是三個 NOLGIA 專案如何以這種風格製作,逐鏡頭解析,包含所有實際發送的提示詞、未達標的草稿以及兩者之間的變化。

  • 閱讀時間 28 分鐘

本文由英文自動翻譯。閱讀原文

鬣狗燒烤主持人黑豹蝙蝠鱷魚爆炸電梯

依序列出你製作的內容。點一下即可跳到對應步驟。

本指南介紹三個為 NOLGIA 製作的真實專案,皆採用相同風格。「House animals」由五段 4 秒的 Seedance 2.5 影片組成,內容是不屬於普通美國郊區的動物,剪輯成 18 秒。「Shoot out」由六段 4 秒的地下犯罪電影時刻影片組成,搭配 Lyria 3.5 配樂,剪輯成 19.5 秒。「NOLGIA prompting」呈現 1990 年代末嘻哈音樂錄影帶的世界,逐鏡頭拍攝:細節、汽車、低角度、無人機,以及由角色表生成的兩位主持人。這是提供本指南的影片背後的專案。以下每個提示詞皆為原始發送內容;唯一的改變是將長破折號改為連字號。影片中的所有人物皆由 NOLGIA 生成。

這種風格取材自舊式家庭錄影:廉價的相機、普通的午後或深夜、鏡頭後方會做出反應的人。懷舊感來自相機與場景;吸引目光的關鍵,則是畫面中那個不該出現的事物,以及眾人對它的平靜反應。

為什麼這些短片能讓人停下滾動

完成的 House animals 影片:一隻黑豹走過在泳池邊閱讀的祖母身後,祖母撫摸著躺椅旁的一隻黑豹,祖母對著生日蛋糕上的蝙蝠開槍,兩隻鱷魚游過後院泳池中的男孩,一隻繫著牽繩的鬣狗將鼻子推近鏡頭,隨後出現 NOLGIA 標誌
House animals, 18 秒:五個片段,最後兩秒覆蓋 NOLGIA 標誌。
完成的 Shoot out 影片:一名男子離開倉庫爆炸現場,一架直升機掠過老人的房屋,一名男子在唱片行,一名戴金鏈的老人,一名男子在廚房開槍,一名男子在夜間商店外踢車門,以及一名男子在電梯內,隨後 NOLGIA 標誌出現在關閉的電梯門上
Shoot out, 19.5 秒:七個鏡頭搭配 Lyria 3.5 配樂。

每個片段都基於相同的邏輯運作:在一個平凡且可信的世界中,放入完全異常的事物,而人們則將其視為正常,或像真實的人一樣反應。笑點周圍的一切刻意保持平淡,讓那唯一異常的事物成為焦點。

場景平凡的世界唯一的異常事物人們的反應鏡頭動作
不尋常的遛狗傍晚的郊區人行道兩隻被牽著的斑鬣狗老先生像狗主人一樣安撫牠們:「嘿。嘿。過來。」當鬣狗靠近鏡頭時後退
隨意的撫摸溫暖夕陽下的後院泳池一隻黑豹躺在她的躺椅旁她撫摸牠並對著鏡頭微笑站在幾公尺外,像家庭錄影帶一樣漂移並變焦
生日派對擠滿氣球和蛋糕的客廳在天花板盤旋的蝙蝠孩子們繼續奔跑;奶奶冷靜地朝蝙蝠開槍每次開槍時驚嚇並失去奶奶的身影
泳池派對後院泳池旁的家庭烤肉泳池裡的兩隻鱷魚沒人會在意他在派對中閒逛,跟著一隻鱷魚經過一個浮具
彩蛋:黑豹走了進來一位祖母在泳池邊看報紙一隻黑豹從她後方的灌木叢中走出她凝視了一會兒,然後又喝了一口飲料先注意到,嚇了一跳,然後鏡頭拉近
家中的寵物。
場景平凡的世界發生了什麼事他如何反應
停車場傍晚時分的工業區空地他身後的倉庫爆炸了他繼續朝鏡頭方向走
直升機一條安靜的街道,一輛老轎車,一間小房子一架軍用直升機低空掠過屋頂他戴上太陽眼鏡
便利商店夜間的商店停車場他以巨大的力量踢了一扇車門他看了看車門,然後走開了
廚房午夜時分的小公寓廚房另一間房間傳來的槍聲回敬一槍,然後又吸了一口菸
唱片行打烊後的 1990 年代唱片行鏡頭外的一槍回敬一槍,然後他消失了
電梯老舊公寓大樓的電梯他在門關閉時開槍門在他面前關上
交火場景,正如每個提示詞所描述的那樣。

從概念開始,而不是從提示詞開始

最強的片段始於一個你一秒就能理解的概念,遠在建立任何提示詞之前。請依此順序,由以下要素構建:

  1. 年代: 拍攝的時代,主要指使用的設備。
  2. 地點: 一個平凡的地方:人行道、後院、廚房、停車場。
  3. 角色: 一個普通人,穿著尋常,年齡可信。
  4. 日常活動: 他們在任何一天都會做的事:遛狗、在泳池邊閱讀、在櫃檯抽菸。
  5. 意外元素: 唯一不該出現的事物。
  6. 自然反應: 他們恐慌、忽略它,或異常冷靜。
  7. 鏡頭行為: 拍攝者身體上的反應。
  8. 真實感: 真實的比例、真實的光線、真實的解剖結構。
  9. 質感: 你所指定相機的影像質感。

提示:先決定反應. 這裡的幽默大多來自冷面笑匠式的風格:祖母啜飲她的飲料,泳池裡沒人看鱷魚,從爆炸現場走出來的男人從不回頭。在撰寫其他內容之前,先決定人們是恐慌、無視還是保持冷靜,因為反應決定了鏡頭如何運作以及片段如何結束。

透過選擇攝影機來決定年代

對於這種風格,攝影機是故事的一部分。不要要求通用的電影感影像,而是具體說明記錄設備及其限制:廉價的消費級光學鏡頭、手持移動、不完美的構圖、反覆對焦、數位雜訊、壓縮、動態模糊以及變化的曝光。

你可以改變年代並保留其他所有元素:1990 年代初期的消費級攝影機、1990 年代後期的家用錄影機或 2000 年代初期的數位 Handycam,各自提供不同的質感。無論你選擇哪一種,都要保持一致:地點、服裝、攝影機行為和影像都應屬於同一個世界。House animals 從未指定年份;兩個 Shoot out 提示詞則有指定,停車場採用 2000 年代初期的低預算街頭錄影風格,唱片行則具有正宗 90 年代消費級相機的感覺。

質感能支撐起一個逼真的場景,但無法取代它。在巨型動物身上加上顆粒感,它依然是隻巨型動物。大多數提示詞都以「視覺 DNA」區塊結尾,用來描述攝影機的影像及其行為,而質感相關的描述行幾乎每次都相同:

斑鬣狗提示詞中的質感描述行(其他《八家將》動物提示詞也使用幾乎相同的描述)
Cheap consumer-camera optics.
Slight motion blur.
Subtle rolling shutter.
Digital noise.
Compression artifacts.
Dirty imperfect lens.
Natural exposure changes.

注意:要求攝影機的效果,而非其疊加圖層. 唱片行提示詞要求「輕微的 VHS 式柔和感,但不要假的 VHS 疊加圖層」,然後逐一排除:不要時間戳記。不要 REC 圖示。不要文字疊加。描述老式攝影機對畫面造成的影響,並明確表示你不想要濾鏡會添加的圖形。若要在此基礎上鎖定某個年代,請用該年代的攝影機來命名,就像 寫實影片指南 要求使用真實的 1990 年代末美國攝影機一樣,並檢查每個鏡頭中的汽車、手機和服裝。

保持世界平凡,讓動物維持真實尺寸

荒謬感之所以有效,是因為周圍的一切都很尋常。郊區的人行道看起來就該像郊區的人行道,生日派對也該像真正的家庭生日派對。描述尋常的服裝、可信的年齡和自然的行為,然後加入那件不尋常的事物,並說明每個人如何圍繞它表現。

對於動物,當比例很重要時,請說明正常的現實世界尺寸和解剖結構,這樣不尋常的動物才不會變成奇幻怪獸。每個保留的 House animals 提示詞都有類似以下的內容:

  • Hyenas: "The hyenas are NORMAL REAL-WORLD SIZE. Realistic adult spotted hyena proportions. NOT oversized. NOT monstrous. NOT fantasy animals."
  • Panther: "approximately the size of a real adult black leopard. NOT oversized, NOT gigantic, NOT a fantasy creature."
  • Crocodiles: "approximately 2.5 to 3 meters long, normal real-world crocodile size. NOT GIANT. NOT MONSTROUS."

Shoot out 對物理效果也做同樣的事。便利商店的提示詞要求衝擊感「沉重且物理上可信」,並排除「誇張的超級英雄動作」;停車場的提示詞則要求「RAW REALISM」且沒有英雄式的慢動作。

讓攝影機像是一個人手持著它

攝影機應該表現得像一個拿著廉價相機的真人,而不是擁有完美控制的虛擬操作員。請寫下腳步聲、輕微晃動、緊張反應、突然重新構圖、意外變焦、自動對焦搜尋、歪斜構圖,以及主體離開畫面的時刻。接著,給操作員對異常事物的反應:正是這種反應讓畫面感覺像是被記錄下來,而不是被渲染出來的。

場景操作員的行為,如文字所述
不尋常的遛狗邊走邊拍攝,當鬣狗靠近鏡頭時本能後退,後退時相機隨之晃動
生日派對每次開槍時都劇烈驚嚇,弄丟了奶奶又找到她,同時有孩子直接從鏡頭前走過
黑豹走了進來在她發現之前先注意到黑豹,變得緊張且手抖,並緩慢變焦拉近
停車場爆炸時驚嚇並側身猛晃,失去對其臉部的對焦,然後在他持續靠近時後退
直升機無法控制地晃動,當它從頭頂掠過時低頭閃避,然後抬起頭發現他仍站在原地
電梯對開槍反應稍慢,向後猛退,停留在緊閉的門上,然後走近一步

接著,排除所有鄰居拿相機也做不到的工具。每個提示詞結尾都列出一串「不要」。House animals 提示詞都排除了穩定器、斯坦尼康和慢動作;Shoot out 提示詞主要針對精緻外觀:不要英雄式慢動作、不要平滑跟拍、不要音樂錄影帶式的精緻感、不要好萊塢動作片攝影風格。

提示:使用作曲家的鏡頭運動. 在提示詞下方,運鏡列有二十種運鏡方式,分為三種強度:推鏡與拉鏡、搖鏡與俯仰、環繞、升降與橫移、跟拍、變焦、希區考克变焦、移焦、手持、視差漂移、無人機穿越與固定鏡頭。選擇其中一種會在你的提示詞末尾增加一句話。所有六個 Shoot out 片段和 House animals 獎勵片段都使用了強度為 Strong 的 Handheld,這些提示詞的最後一句話「Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.」就是該選擇的結果。甩鏡和急推變焦不在該列中,因此請將它們寫入提示詞;運鏡指南提供了實現每種效果的用語。開啟設定為 Handheld 且強度為 Strong 的編輯器。

撰寫提示詞:世界、動作、鏡頭、禁忌事項

這些提示詞不會微觀管理每一幀畫面。它們為 Seedance 2.5 提供清晰的世界觀、角色、動作、必須保持不變的內容以及鏡頭語言,並讓模型生成動態。House animals 的提示詞都是按照相同的方式建構:

  1. 標題包含模型與格式:SEEDANCE 2.5 VIDEO PROMPT,9:16。
  2. 世界觀以簡短行描述:地點、時段,以及幾個能營造氛圍的道具。
  3. 人物:年齡、服裝與正在進行的動作。
  4. 異常現象,並用 IMPORTANT 行鎖定其大小。
  5. 動作,每行一個節拍,對話內容加引號。
  6. 禁止事項:無攻擊、無受傷、無血腥。
  7. 視覺基因:攝影機行為,接著是影像風格。
  8. 排除項:家庭錄影不會有的工具與外觀。
  9. 一句話說明畫面應有的感覺。

Shoot out 提示詞遵循相同順序,但其中三個省略標題與 VISUAL DNA 小節,將攝影機影像放在結尾段落。兩種方式皆可;重點是每個部分都要存在。

不尋常的遛狗(Seedance 2.5)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A completely ordinary American suburban neighborhood during late afternoon.

Quiet residential street.
Driveways.
Parked cars.
Small front yards.
Mailboxes.
People occasionally visible through windows.

An elderly man in an old windbreaker, loose pants and worn sneakers is casually walking down the sidewalk.

He is walking TWO REALISTIC ADULT SPOTTED HYENAS on sturdy leashes.

IMPORTANT: The hyenas are NORMAL REAL-WORLD SIZE. Realistic adult spotted hyena proportions. NOT oversized. NOT monstrous. NOT fantasy animals.

The man behaves as if he is simply walking his dogs around the neighborhood.

The camera operator is walking several meters ahead of him, filming casually with a cheap handheld camera.

The hyenas suddenly notice the camera.

They become alert.

One hyena walks toward the camera and gets extremely close to the lens.

Its face fills most of the frame.

It stares directly into the camera.

The hyena begins aggressively growling and making realistic low snarling sounds.

The second hyena pulls toward the camera too.

The old man struggles slightly with the leashes.

He says calmly, almost like an annoyed dog owner:

"Hey. Hey. Come on."

He pulls the leashes backward and tries to guide both hyenas away from the camera.

The first hyena continues staring directly into the lens while growling.

The camera operator instinctively backs away.

The camera shakes as the operator retreats.

The old man keeps pulling the leashes and calmly trying to regain control.

The hyenas eventually turn away and continue walking down the sidewalk.

The old man casually follows them.

CUT.

No attack.
No biting.
No injury.
No gore.

The tension comes entirely from the animals getting extremely close to the camera and the old man struggling to control them.

VISUAL DNA:
Extremely raw handheld consumer-camera footage.
Real human camera operator walking backward.
Constant handheld micro-shake.
Uneven footsteps.
Natural human breathing movement.
Crooked framing.
Camera operator reacts nervously as the hyena approaches.
Sudden backward movement.
Camera briefly loses the old man.
Autofocus hunts between the hyena's face and background.
The hyena's face occasionally becomes extremely close to the lens.
Natural focus breathing.
Realistic animal fur.
Realistic hyena anatomy and proportions.
Realistic ears, teeth and facial movement.
Natural afternoon lighting.
Realistic suburban shadows.
Cheap consumer-camera optics.
Slight motion blur.
Subtle rolling shutter.
Digital noise.
Compression artifacts.
Dirty imperfect lens.
Natural exposure changes.

NO giant hyenas.
NO oversized animals.
NO fantasy creatures.
NO CGI appearance.
NO cinematic wildlife photography.
NO gimbal.
NO Steadicam.
NO drone.
NO perfect tracking.
NO slow motion.
NO polished Hollywood cinematography.
NO music-video look.

Make it feel like a random neighbor filmed an old man taking his extremely unusual pets for their daily walk.
一位身穿米色雨衣的老人在夕陽下的郊區人行道上牽著兩隻斑點鬣狗;其中一隻走向鏡頭直到口鼻充滿畫面,張開嘴,然後兩隻都退回他身邊
不尋常的遛狗。 4 秒,此提示詞的首次生成。

建立鏡頭:細節、汽車、低角度與無人機

第三個專案,NOLGIA 提示詞撰寫,是一堂關於鏡頭語言的課,其規則是提示鏡頭,而不只是主體。說明發生了什麼、相機在哪裡、如何移動、光線狀況以及影像應如何表現。這正是將通用片段轉化為具有單一視覺識別度素材的關鍵。其中的每個提示詞都遵循一個公式:

鏡頭公式
Subject + Environment + Action + Camera position + Camera movement + Lighting + Image texture + Timing

手部或細節鏡頭讓小物件感覺像是由真正的攝影師捕捉到的:緊湊的構圖、略微不完美的手持相機、用於產生硬高光效果的直接閃光燈,以及必須找到細節的自動對焦。烤瓷牙鏡頭是較難的版本,因為細節是文字:NOLGIA 以鑽石形式呈現,在她的牙齒上,在洗手間鏡子中。

草稿:NOLGIA 這個詞以大鑽石字母呈現,覆蓋她的嘴唇並延伸至牙齒之外
草稿:字母覆蓋她的嘴唇。
草稿:鑽石烤瓷牙顯示為 NOLGA,缺少了 I
草稿:NOLGA,缺少了 I。
草稿:片段開始於她身穿黑色禮服直立站立,距離洗手間鏡子較遠,然後鏡頭推進至她的烤瓷牙
草稿:開始於她直立站立,距離鏡子較遠。
保留版本:一位金髮女子倚靠在夜店洗手間的鏡子上,用一根手指將下唇往下拉;鏡頭不斷推近,直到鏡中出現寫著 NOLGIA 的鑽石牙套填滿畫面
保留版本, 第九次生成:她倚靠在鏡子上,拉下嘴唇,鑽石牙套上的 NOLGIA 字樣清晰可見約一秒鐘。

每次失敗都成為提示詞中的一條規則。當字母擴散到她的嘴唇上且字母 I 消失後,提示詞加入了「NO floating text(無浮動文字)」、「NO lettering on lips(嘴唇上無文字)」以及「one letter per tooth(每顆牙齒一個字母)」,並指定 N = 一顆牙齒,直到 A = 一顆牙齒。當她以站立姿勢且距離鏡子過遠時,提示詞加入了一個 IMPORTANT BODY POSITION(重要身體姿勢)區塊:她必須 LEANING FORWARD TOWARD THE MIRROR(身體前傾靠近鏡子),而不是 NOT standing upright several feet away(直立站在幾英尺外)。這個場景總共使用了 Seedance 2.5 生成了十三次;最終保留的是第九次。

鑽石牙套(使用 Seedance 2.5,並將 Rack focus 設定為 Strong)
DURATION: 6 SECONDS

AUTHENTIC VERY LOW-QUALITY LATE-1990s CONSUMER CAMCORDER FOOTAGE.

A beautiful, hot-looking young blonde woman in her late 20s with a wealthy late-1990s hip-hop nightlife aesthetic is standing in front of a large mirror inside an upscale 1990s nightclub restroom.

She looks rich and glamorous, wearing a revealing low-neck black 1990s-style dress, diamond earrings, layered gold necklaces, a luxury diamond watch, rings, and expensive jewelry. Her blonde hair is slightly messy and her makeup is glamorous but imperfect and slightly smudged, giving her a late-night party look.

IMPORTANT BODY POSITION:

She is LEANING FORWARD TOWARD THE MIRROR.

Her upper body is bent naturally toward the mirror, bringing her face extremely close to the glass. Her mouth is only a few inches from the mirror.

She is NOT standing upright several feet away.

She is physically leaning into the mirror to inspect her grill at extremely close range, matching the intimate composition and proximity of the provided reference.

The camera is positioned very close to her reflection, creating an almost claustrophobic close-up of her face and mouth.

She is looking at HERSELF IN THE MIRROR, not directly at the camera.

IMPORTANT GRILL ACTION:

She slightly opens her mouth and uses one finger to gently pull her lower lip downward, exposing the entire custom grill while examining it in the mirror.

Her mouth and grill should dominate the frame.

DENTAL GRILL:

A luxurious custom diamond grill fitted precisely over her natural upper and lower teeth.

The grill spells:

N O L G I A

The letters must be BOLD, LARGE, CLEAR, AND READABLE.

Each letter occupies its own individual tooth position:

N = one tooth
O = one tooth
L = one tooth
G = one tooth
I = one tooth
A = one tooth

The letters are physically constructed from diamonds and integrated directly into the grill.

Use a STRONG CLASSICAL OLD-STYLE SERIF FONT, similar to a bold Times New Roman / traditional Roman serif.

Straight serif letterforms.
Bold vertical strokes.
Clearly defined serifs.
Traditional Roman proportions.
Luxury custom-jewelry appearance.

NOT cursive.
NOT calligraphic.
NOT handwritten.
NOT gothic.
NOT futuristic.
NOT thin or delicate.

The six letters must clearly read:

NOLGIA

The letters must remain entirely ON THE TEETH.

NO floating text.
NO text overlay.
NO lettering on lips.
NO lettering on gums.
NO lettering on skin.
NO lettering outside the mouth.

CAMERA MOVEMENT:

0-1 SECOND:
Begin with a very close handheld view of her leaning toward the mirror. Her face and mouth are already extremely close to the glass. The camcorder is slightly out of focus.

1-2.5 SECONDS:
The operator moves closer with a crude optical zoom toward her mouth. She continues leaning toward the mirror, examining her grill.

The autofocus hunts badly, briefly focusing on her nose and lips before shifting onto the teeth.

2.5-4 SECONDS:
EXTREME CLOSE-UP of the grill.

The camera is almost macro distance from her reflection. Her mouth fills most of the frame.

She gently pulls the lower lip downward with her finger so the grill is completely visible.

The autofocus finally locks onto the individual diamond letters.

The bold N-O-L-G-I-A lettering should be clearly readable despite the degraded footage.

4-5 SECONDS:
The camera briefly loses focus again. The diamonds bloom into soft highlights. The image becomes blurry for a moment, then autofocus snaps back onto the grill.

5-6 SECONDS:
The camera pulls back slightly with a crude optical zoom, revealing more of her face and upper body while she remains LEANING TOWARD THE MIRROR.

She continues looking at herself in the mirror and admiring the grill.

CAMCORDER AESTHETIC:

THIS MUST LOOK LIKE REAL LATE-1990s CONSUMER CAMCORDER FOOTAGE.

Very low resolution.
Heavy coarse analog grain.
CCD noise.
Soft image.
Color bleeding.
Crushed blacks.
Blown highlights.
Slight yellow/green analog cast.
Chromatic aberration.
Motion smear.
Autofocus hunting.
Focus breathing.
Exposure pumping.
Imperfect white balance.
Crude optical zoom.
Analog compression artifacts.
Unstable handheld framing.

The camera should feel physically operated by a real person standing extremely close to her.

NO modern digital camera.
NO cinematic commercial photography.
NO DSLR look.
NO clean sharp image.
NO gimbal.
NO Steadicam.
NO modern stabilization.
NO perfect focus.
NO polished Hollywood cinematography.

The WOMAN and her JEWELRY should look expensive and luxurious.

THE FOOTAGE SHOULD LOOK CHEAP, OLD, GRAINY, AND RAW.

The visual reference should primarily inform the extreme proximity, leaning posture, mouth-to-mirror composition, and grill inspection action.

DIEGETIC AUDIO ONLY:
quiet nightclub restroom ambience, distant muffled voices, subtle breathing, jewelry movement, clothing movement, and faint camcorder handling noise.

NO BACKGROUND MUSIC. Camera: locked off on a tripod with a fast, decisive rack focus, the focus snapping from a blurred foreground to the subject and back, very shallow depth of field.

汽車或動作鏡頭最適合將動作與攝影機指令分開:先描述汽車原地打轉,再說明攝影機如何觀看。Porsche 提示詞將操作者置於街道層級,低角度且緊貼路面,呈前側四分之三角度,同時橫向移動並後退,以保持汽車在構圖中,而輪胎煙霧則滾入鏡頭。影片播放到一半時,同一段片段會變成第四種鏡頭。

無人機或俯拍鏡頭能展現地點、規模或動作的整體樣貌。請說明高度、角度與移動方式:高空、幾乎垂直向下看、緩慢旋轉,讓輪胎煙霧的圓圈從上方清晰可見。保持相同的光線與質感,讓空拍視角仍屬於同一部影片。

低角度鏡頭來自將攝影機物理性地放置在主體下方,而非要求戲劇化的角度。教堂提示詞從低處開始,緊貼一名站在紅色 Lamborghini 車頂、雙臂張開的男子身後,讓汽車、他的背影與拱形天花板在同一個畫面中層疊呈現。接著鏡頭粗糙地推近至三位震驚的牧師,橫搖掃過會眾,並拉遠至整個中殿。

一輛黑色 Porsche 911 在夜間棕櫚樹夾道的邁阿密路口甩尾,低角度從街道拍攝,輪胎煙霧繚繞,背景棕櫚樹燃燒,接著切換至高空俯視,汽車畫出一圈煙霧
Porsche 鏡頭。 4 秒,第一次生成:先街景平視,再高空俯視。
一座宏偉的羅馬教堂內正在舉行儀式:一名穿皮夾克的男子站在紅色 Lamborghini 的車頂,位於走道上雙臂張開,低角度從後方拍攝,接著推近至祭壇前三位震驚的牧師、會眾,以及中殿的廣角視圖
教堂鏡頭。 4 秒,此場景的第五次生成,其中一次使用 Wan 3.0。
Porsche(Seedance 2.5,選擇了 Drone fly-through 的 Medium 強度)
DURATION: 4 SECONDS

A realistic early-1990s luxury hip-hop music-video shot in a wealthy Miami residential neighborhood at night. A period-correct Porsche 911 performs controlled, aggressive donuts at an open residential intersection, rotating around the same center point with believable vehicle weight, tire grip and momentum. The Porsche has realistic proportions, authentic period bodywork and glossy paint reflecting the surrounding streetlights.

In the background, several tall palm trees and tropical trees are burning intensely, with realistic flames climbing the trunks and branches. Thick natural smoke drifts across the street and catches the light. The houses remain believable upscale Miami residential architecture, with warm practical lights visible through windows and subtle reflections on parked cars.

CAMERA - FIRST HALF:
A real handheld camera operator stands safely at street level facing the Porsche. Begin with a low front three-quarter angle close to the pavement. The Porsche enters frame already sliding and begins rotating into a donut. The camera operator instinctively shifts sideways and slightly backward to keep the car framed, creating subtle natural handheld shake and imperfect framing. Tire smoke rolls toward the camera and briefly softens the image.

As the Porsche completes another rotation, the operator quickly tilts upward for a moment, catching the burning palm trees behind the car.

CAMERA - SECOND HALF:
Seamlessly transition to a high overhead drone perspective looking almost straight down. The Porsche continues its donuts below, leaving a circular pattern of tire smoke on the pavement. The drone slowly rotates around the spinning car while maintaining a wide overhead composition, revealing the residential streets, surrounding houses, burning trees and drifting smoke.

Keep the Porsche’s donut trajectory physically consistent and believable. No impossible drifting, no floating vehicle, no exaggerated suspension movement.

Authentic early-1990s analog music-video image character: realistic film grain, subtle halation around streetlights and flames, natural exposure changes, slightly imperfect optics, realistic reflections, believable night lighting and natural atmospheric smoke.

The people and environment should remain realistic and grounded. No parody, no exaggerated proportions, no videogame appearance, no overly polished CGI.

AUDIO: No background music. Only diegetic sound: Porsche engine revving, tires squealing against asphalt, tire smoke and road noise, distant neighborhood ambience, crackling flames, wind moving through palm trees and the occasional echo of the engine between houses. Camera: a first-person drone flight forward through the scene, banking smoothly around obstacles at a steady pace, the horizon tilting with each turn.
教堂(Seedance 2.5,選擇了 Handheld 的 Strong 強度)
DURATION: 7 SECONDS

Authentic late-1990s handheld camcorder footage inside an enormous historic Roman church in Rome. Massive stone columns, towering arches, marble floors, ornate religious architecture, stained glass and an extremely high vaulted ceiling create a huge sense of scale. A religious service is already taking place.

A bright red 1990s Lamborghini is positioned inside the church near the central aisle.

A young adult man in authentic late-1990s luxury hip-hop clothing is standing completely still on the roof of the Lamborghini, seen entirely from behind.

He wears a dark oversized 1990s leather jacket, loose trousers and expensive sneakers. From the rear angle, his clothing completely covers his neck and no chain or necklace is visible from behind.

He is not walking or moving forward.

He stands firmly on the roof of the car with both feet planted and both arms raised and spread outward, confronting the altar and the people in front of him. His posture is confident and defiant but physically natural.

The congregation has already noticed him. People throughout the church are turned toward him, confused and staring.

At the front of the church, three pastors are standing behind their proper preaching podium/lectern at the altar, facing the congregation. They have stopped preaching and are staring directly toward the man with genuine surprise and confusion.

CAMERA

Everything is filmed from the perspective of someone holding a late-1990s handheld camcorder.

0.0-2.0 SECONDS - BACK OF THE MAN

Begin with a low-angle handheld shot directly behind the man.

The camera is close enough to clearly see his back, shoulders, raised arms and the roof of the red Lamborghini.

Do not show his face.

Do not show his chain.

The framing should emphasize:

man from behind → red Lamborghini → enormous church interior ahead.

The camera operator makes tiny natural movements while standing in place: subtle shake, breathing movement, slight side-to-side drift and occasional framing correction.

The man remains stationary with both arms spread.

The high ceiling and massive architecture should already be visible above him.

2.0-3.5 SECONDS - SUDDEN ZOOM TO PASTORS

Without cutting, the camcorder suddenly performs a crude optical zoom toward the altar.

The operator physically zooms in while slightly adjusting the camera.

The image becomes temporarily soft and grainy during the zoom.

The three pastors behind the podium become the focus.

All three are staring toward the man.

Their expressions are genuinely shocked and confused - subtle raised eyebrows, widened eyes, exchanged glances and uncertain body language.

They remain behind their preaching podiums.

3.5-5.0 SECONDS - AUDIENCE

The camera quickly pans away from the pastors and moves across the congregation.

People are already turning around and looking toward the Lamborghini.

Some people whisper to each other.

Some stare directly toward the man.

The camera passes behind shoulders and briefly loses focus on individual people before naturally correcting.

Keep this movement messy and handheld.

5.0-7.0 SECONDS - PULL OUT / FULL REVEAL

The operator physically steps backward while pulling the camera out to a much wider composition.

The entire scene gradually becomes visible:

the man standing on the red Lamborghini with both arms raised,

the congregation surrounding him,

the three pastors behind their podiums,

the altar,

the massive marble architecture,

the towering columns,

and the enormous vaulted ceiling.

The camera continues to shake naturally as the operator backs away.

End on a wide handheld camcorder frame where the scale of the Roman church completely surrounds the strange confrontation.

CAMCORDER LOOK

This must look like genuine late-1990s handheld camcorder footage, not modern cinematic footage.

Strong natural handheld shake, slight walking/breathing movement, imperfect framing, crude optical zoom, autofocus hunting, exposure pumping, motion blur, low-light CCD noise, analog grain, chromatic noise and soft consumer-camcorder optics.

The image should occasionally look slightly underexposed and then compensate naturally.

No gimbal.

No Steadicam.

No smooth robotic camera.

No modern stabilization.

No perfect cinematic focus pull.

The camera operator should feel like a real person standing inside the congregation, physically reacting to what they are seeing.

The Lamborghini, man, congregation and architecture must maintain realistic scale and proportions.

The scene should feel completely real despite the bizarre situation - like genuine 1990s footage of an unexpected interruption during a religious service.

AUDIO

NO BACKGROUND MUSIC.

Only diegetic sound: enormous church reverb, quiet congregation murmurs, people whispering, footsteps on marble, clothing movement, distant church ambience, the pastors’ confused voices and natural camcorder handling noise.

No narrator.

No cinematic sound effects.

The final shot should feel like a bizarre piece of raw archival camcorder footage from a 1990s hip-hop world. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

注意:選擇與提示詞相符的運鏡方式. 運鏡選項加入的句子會放在你的提示詞之後,而且不會檢查兩者是否一致。grill 提示詞要求手持攝影機和粗糙的變焦,但其 Rack focus 選項卻加入了「locked off on a tripod」。Porsche 提示詞要求緩慢的俯視旋轉,但其 Drone fly-through 選項卻加入了「a first-person drone flight forward」。這次兩個片段都遵循了它們的提示詞。請選擇與提示詞意思相同的運鏡,或者不選。

提示:描述攝影機,而非形容詞. 單獨使用 cinematic 或 realistic 這類詞彙效果有限。請說明攝影機實際在做什麼:「低角度手持攝影機沿著汽車跟拍,操作者勉強維持構圖,粗糙的光學變焦,自動對焦短暫搜尋,直接閃光燈捕捉車身,深陰影和強烈的 CCD 顆粒感。」主體是內容;攝影機、光線和不完美之處才是風格的核心。

讓每個鏡頭處於同一個世界

改變鏡頭不代表改變外觀。在所有鏡頭中保持相同的視覺配方,只改變攝影機的位置和移動方式。NOLGIA 的提示詞鏡頭都使用了這個配方:

元素保持一致
光線直接機頂閃光燈、深重陰影、昏暗暖色實用光源、些許冷調藍紫溢光
質感厚重類比顆粒、低光 CCD 噪點、光學柔化、輕微色散
曝光壓暗的黑色、過曝的高光、不完美的曝光與對焦呼吸
相機感覺手持微顫、不完美的構圖、自動對焦搜尋與粗糙的光學變焦

閃光燈是將它們聯繫在一起的關鍵。海濱場景的提示詞讓操作員沿著一排汽車奔跑並連續閃光:先照整群人,再照法拉利鍍鉻部件,接著照一位對著 1990 年代手機爭論的老婦人,然後畫面回到黑暗、充滿顆粒的曝光。

夜晚的海濱停車場與城市天際線:一群穿著 1990 年代末嘻哈服飾的人站在一排跑車旁,一位戴金鏈的老婦人對著 1990 年代手機爭論,同時相機閃光燈閃爍
海濱場景。 6 秒,此提示詞的首次生成。
海濱場景 (Seedance 2.5,選擇了強烈的手持效果)
DURATION: 6 SECONDS

Authentic late-1990s hip-hop camcorder footage at night in an industrial waterfront parking area overlooking a major American city skyline. A large group of realistically proportioned adults stands casually around a lineup of expensive 1990s luxury and sports cars - period-correct Ferrari, Porsche, Lamborghini, Mercedes and other high-end vehicles. The cars are immaculate and physically believable, with headlights and small practical lights glowing against the dark pavement.

The group wears authentic late-1990s luxury hip-hop and subtle punk-influenced streetwear: oversized leather jackets, baggy denim, dark designer clothing, heavy diamond chains, pendants, watches, rings and sunglasses. The styling feels genuinely expensive and naturally worn, never like costumes.

An elderly white woman stands prominently among the group, dressed in an unexpectedly elegant old-fashioned outfit with a substantial diamond hip-hop chain around her neck. She holds a chunky 1990s-style handheld cellular phone against her ear and is angrily arguing with someone on the phone. She gestures naturally with her free hand, visibly irritated and animated while everyone around her remains relatively calm.

CAMERA - FAST HANDHELD CAMCORDER MONTAGE

The entire sequence feels like someone physically running around the scene with a late-1990s handheld camcorder, rapidly capturing fragments of the night.

0.0-1.2 sec:
Start with a shaky wide shot of the entire group and lineup of luxury cars. The operator quickly steps sideways while filming, causing natural handheld bobbing and imperfect framing. Several bright camera flashes fire during the movement, briefly illuminating faces, jewelry and car bodywork.

1.2-2.2 sec:
Abrupt handheld move toward a Ferrari. Brief close-up of the front emblem and headlights. The camera shakes while the operator moves around the front corner of the car. A harsh direct flash briefly blows out the chrome and paint reflections.

2.2-3.3 sec:
Hard handheld whip toward the elderly woman. She is already mid-conversation, angrily speaking into the large 1990s phone. The camera gets unusually close to her face, slightly off-center. Another direct flash suddenly illuminates her face, diamond chain and phone.

3.3-4.3 sec:
Quick low handheld shot moving past several pairs of expensive sneakers and car tires, then tilting upward toward the group. The operator briefly loses framing before correcting.

4.3-5.2 sec:
Fast medium shot of the woman continuing her furious phone conversation while the luxury cars sit behind her. She gestures sharply with her free hand. The people around her barely react.

5.2-6.0 sec:
The camera quickly backs away into a wide shot, revealing the entire group and cars again. Another burst of direct flash briefly freezes the scene before the image falls back into dark, grainy camcorder exposure.

IMAGE CHARACTER

Make this look like real late-1990s consumer/professional camcorder footage, not modern cinema.

Strong but believable analog/digital camcorder grain, low-light noise, slight chromatic fringing, blown highlights from direct flash, dark crushed shadows, imperfect autofocus, exposure pumping, motion smear, subtle interlacing/CCD character, occasional overexposure from the flash and slightly washed-out colors.

The flash should feel like a real on-camera photographic flash, creating sudden harsh illumination against the otherwise dark environment.

The handheld movement is extremely important: no gimbal, no Steadicam, no perfectly smooth dolly, no robotic movement. The operator should physically stumble, reposition, tilt, pan and correct framing while trying to capture the group.

Human proportions, vehicles, clothing, jewelry and environmental scale must remain completely realistic.

The overall feeling is wealthy, mysterious, badass late-90s hip-hop culture captured by someone with a camcorder after midnight - raw, spontaneous and expensive at the same time.

AUDIO - NO BACKGROUND MUSIC.

Only diegetic sound: distant city ambience, car engines idling, footsteps on pavement, clothing movement, camera handling noise, occasional camera-flash mechanical sounds, distant traffic and the elderly woman’s angry voice coming through the phone conversation. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

同一套配方讓兩位主持人跨越兩個鏡頭。每位主持人最初都是角色設定圖:八種表情加上同一人穿同一套服裝的全身照,背景為純灰色。接著是一張沒有主持人的豪宅背景圖:一隻黑豹坐在黑色皮沙發上,由相機閃光燈直接打光,以及一位戴兜帽的男子坐在房間深處的椅子上,並為主持人預留了空間。兩段主持人影片都在影片編輯器中的 參考圖片 下附加這三張圖片,且兩段提示詞都以命名這些圖片開頭:「使用上傳的男性角色設定圖作為該男子的精確身分參考。」

男性主持人的角色設定圖:八張同一位金髮男子的肖像,身穿黑色皮夾克、白襯衫並配戴金項鍊,每張表情不同,另有一張全身照,背景為純灰色
他的設定圖, GPT Image 2.5 Flare。
女性主持人的角色設定圖:八張同一位金髮女子的肖像,身穿黑色洋裝與皮夾克並配戴金飾,每張表情不同,另有一張全身照,背景為純灰色
她的設定圖, GPT Image 2.5 Flare,使用相同的提示詞但修改了細節。
他的角色設定圖(GPT Image 2.5 Flare)
Create a professional photorealistic character reference sheet of ONE fictional adult white male, approximately 28-32 years old. This is a BASE CHARACTER SHEET for maintaining identical character likeness across future AI-generated images and videos.

CHARACTER IDENTITY: handsome masculine face, strong defined jawline, straight proportional nose, slightly deep-set expressive eyes, natural eyebrows, subtle cheekbone definition, realistic lips, short-to-medium messy blonde 1990s hair with the exact same hairline, length, texture and styling in every frame. Realistic facial asymmetry, visible pores, subtle skin imperfections and completely natural human skin texture. Athletic masculine build, broad shoulders and realistic proportions.

WARDROBE MUST REMAIN IDENTICAL IN EVERY PANEL: oversized black leather jacket, loose white/off-white shirt, baggy dark 1990s trousers, heavy gold chain, gold watch and several gold rings. No logos, no text, no modern fashion.

Create ONE clean character-sheet composition containing 8 clearly separated portrait/reference panels of the SAME EXACT PERSON. The face, hairstyle, age, skin tone, body proportions and wardrobe must remain completely consistent across all panels.

PANEL 1 - straight-on neutral expression.
PANEL 2 - straight-on subtle confident expression.
PANEL 3 - slight serious/intimidating expression.
PANEL 4 - subtle smirk.
PANEL 5 - natural genuine smile.
PANEL 6 - surprised/alert expression with slightly raised eyebrows.
PANEL 7 - angry/intense expression, controlled rather than exaggerated.
PANEL 8 - relaxed thoughtful expression.

Also include one full-body neutral standing view beside or below the portrait grid, showing the complete outfit and body proportions.

Use consistent neutral studio lighting so the face can be clearly studied. No dramatic shadows, no extreme color grading, no beauty retouching. Photorealistic skin and facial detail.

The most important requirement is IDENTITY CONSISTENCY: every panel must look unmistakably like the exact same person, not different variations of the character. Preserve the exact facial structure, eye shape, nose, mouth, jawline, hairline, hairstyle, skin tone and proportions across every expression.

Straight photographic reference-sheet presentation, clean neutral background, realistic photography, high facial detail, natural proportions, no text, no labels, no duplicate people, no extra characters.
豪宅背景圖(GPT Image 2.5 Flare)
Create a photorealistic cinematic background plate for an underground late-1990s American hip-hop music video. Show ONLY the environment and background characters described below - do NOT include the main male character or the female character.

INTERIOR: an enormous dark luxury mansion room at night. The room feels expensive, mysterious and slightly sinister. Black and very dark brown interior, dark polished wood, black leather furniture, heavy dark curtains, subtle antique/luxury details, large shadowy architectural elements and a deep black background. The space should feel like a private mansion room rather than a modern studio.

FOREGROUND: a large black leather couch positioned prominently in the room. A large BLACK PANTHER is calmly sitting/lying on the couch, relaxed but alert, realistic anatomy, glossy black fur catching tiny highlights from the camera flash. The panther should feel like a real animal naturally occupying the room, not posed like a fantasy creature.

BACKGROUND: several meters behind the couch, place ONE adult man restrained in a simple heavy chair. His wrists are visibly secured to the chair, but there are NO injuries, NO blood, NO wounds and NO graphic details. He wears a plain white sleeveless undershirt and dark trousers. His face is completely covered by a loose dark cloth hood/bag, creating a mysterious silhouette. He sits tense and uncomfortable, suggesting an intense off-screen interrogation without showing any violence.

LIGHTING IS CRITICAL: reproduce the feeling of a cheap early-2000s/late-1990s consumer camcorder recording in an extremely dark room with the CAMERA FLASH ON. A harsh direct on-camera flash hits the foreground objects, producing bright blown highlights on the black leather couch and the panther's fur while the rest of the mansion falls rapidly into deep darkness. Add subtle dim warm practical lamps far in the background, barely illuminating parts of the room. Mix warm amber practical light with cold blue-violet reflections and flash spill.

CAMERA LOOK: handheld consumer camcorder photograph, slightly low-light and imperfect. Strong optical softness, heavy analog/CCD noise, dense film grain, subtle chromatic aberration, crushed blacks, blown flash highlights, slight lens bloom, imperfect exposure and natural focus falloff. The foreground should be sharper while the background man and mansion architecture gradually fall into soft darkness. The image should feel like a frame captured accidentally during a secret late-1990s/early-2000s underground recording.

COMPOSITION: wide-medium establishing shot showing the couch and panther in the foreground/midground and the restrained hooded man clearly visible deeper in the room. Leave enough empty space around the couch for the female character to be composited beside the panther later, and enough space in the background for the male character to be added later.

No main male character. No female character. No modern objects. No text. No logos. No weapons. No injuries. No blood. No gore. Photorealistic, raw, mysterious, luxurious and unsettling.
豪宅背景圖:一隻黑豹躺在黑色皮沙發上,被相機閃光燈照亮,而在昏暗房間深處,一名戴兜帽的男子坐在椅子上
背景圖, 為主持人預留了空間。
被相機閃光燈照亮的黑暗豪宅:一名身穿黑色皮夾克、戴金項鍊的金髮男子轉向鏡頭說話,接著鏡頭平移至坐在黑色皮沙發上撫摸黑豹的金髮女子
開場鏡頭。 6 秒,兩次生成中的第二次。
開場鏡頭(Seedance 2.5,附上角色設定圖與背景圖,Handheld 強度選 Strong)
Use the uploaded male character sheet as the EXACT identity reference for the man. Preserve his face, blonde 1990s hair, black leather jacket, white shirt, gold chain, jewelry and proportions exactly.

Use the uploaded female character sheet as the EXACT identity reference for the woman. Preserve her face, blonde 1990s hair, black dress, leather jacket, gold jewelry and proportions exactly.

Use the uploaded mansion/location image as the EXACT environment reference. Preserve the dark mansion, black leather couch, black panther, wooden architecture and background composition.

Create ONE continuous 4-second handheld camcorder shot with three connected beats. No cuts.

BEAT 1 - MAN:
Begin with the man standing beside the restrained hooded man in the chair in the background. His back is partially toward the camera. He looks physically tired and composed after an intense off-screen interrogation. His hands are visibly dirty/stained but show no graphic injury. He pauses, then slowly turns his upper body and head toward the camera.

As he turns, the camera operator reacts naturally and performs a crude optical camcorder zoom toward him. The zoom is imperfect and mechanical, briefly losing focus before autofocus catches his face. He looks directly toward the camera and says:

“These videos have been going viral lately.”

His delivery is serious, calm and slightly intimidating. Keep his movement natural and restrained.

BEAT 2 - PAN TO WOMAN:
Immediately after he finishes the line, the camera quickly but imperfectly pans from the man toward the woman sitting on the black leather couch beside the panther.

IMPORTANT CAMERA DIRECTION:
Do NOT pan across the entire mansion or make a huge sweeping movement. Pan through the SHORTEST DIRECT PATH between the man's position and the woman's position. Keep the movement compact and motivated, as if the camera operator is physically turning toward the person who begins speaking.

During the pan, allow natural motion blur, exposure fluctuation and brief autofocus hunting. The camera settles on the woman and panther.

The woman is calmly petting the black panther while looking toward the camera. She says:

“But nobody teaches you how to make them.”

She delivers the line confidently and naturally. While speaking, she continues slowly stroking the panther. The panther remains calm, occasionally blinking or subtly moving its ears.

BEAT 3 - MAN MOVES CLOSER:
As soon as the woman finishes, the camera shifts back toward the man.

The man now walks naturally toward the camera and toward the couch, closing the distance. The camera operator backs up slightly while maintaining a loose handheld frame. The camera performs another crude optical zoom/focus adjustment as the man approaches.

He stops beside the couch near the woman and looks directly into the camera. The camera settles on his face in a slightly imperfect medium close-up.

He says:

“So we got you.”

His delivery is short, serious and confident.

CAMERA STYLE THROUGHOUT:
Authentic late-1990s / early-2000s consumer camcorder footage. VERY handheld. Real human operator movement: micro-shakes, imperfect framing, tiny corrections, slight stumble in movement, natural motion blur and occasional exposure pumping. No gimbal. No smooth modern tracking. No polished Hollywood camera movement.

Use crude optical zoom rather than digital zoom. Zooms should have slight mechanical jerk, focus breathing, autofocus hunting and temporary softness before locking onto the subject.

LIGHTING:
Extremely dark mansion interior with dim warm practical lamps in the background. Harsh DIRECT ON-CAMERA CAMCORDER FLASH illuminates whoever the camera is focused on, producing blown highlights on faces, gold jewelry, black leather and the panther's glossy fur. Deep black shadows surround the subjects. Mix subtle warm amber practical light with faint cool blue-violet tones.

IMAGE CHARACTER:
Heavy CCD noise, analog grain, low-light color noise, optical softness, crushed blacks, blown flash highlights, slight chromatic aberration, mild lens bloom, imperfect exposure and occasional focus hunting. No HDR, no modern sharpness, no cinematic stabilization.

The entire shot should feel like a raw piece of underground 1990s/early-2000s footage captured by someone physically holding an old camcorder with the flash on.

No cuts. No subtitles. No text overlays. No extra characters. Preserve exact character likeness and exact mansion environment. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

結尾鏡頭被寫成直接的延續:「same dark mansion, same black leather couch, same black panther, same furniture, same lighting」,且鏡頭是「the SAME OLD HANDHELD CAMCORDER from the previous scene」。當兩個鏡頭必須剪接在一起時,請在第二個提示詞中說明,並列出延續的元素。

同一座豪宅:女子坐在黑色皮沙發上撫摸黑豹並對鏡頭說話,而男子從沙發後方探身過來
結尾鏡頭。 11 秒,三次生成中的第三次。
結尾鏡頭(Seedance 2.5,附上角色設定圖與背景圖,Handheld 強度選 Strong)
Use the uploaded male and female character sheets as the EXACT identity references. Preserve their faces, hairstyles, body proportions, wardrobe and jewelry exactly.

Use the uploaded mansion/location reference as the EXACT environment. This must feel like a direct continuation of the previous shot: same dark mansion, same black leather couch, same black panther, same furniture, same lighting and same visual atmosphere.

ONE CONTINUOUS 4-SECOND SHOT. No cuts.

COMPOSITION:
The woman remains seated on the black leather couch beside the black panther, calmly petting it throughout the scene. The man is now positioned BEHIND the couch, slightly crouched and leaning forward toward the woman. He is close enough to feel naturally connected to her but remains behind the couch.

DIALOGUE / PERFORMANCE:
The woman looks toward the camera and says naturally:
“Everyone is making these. It’s time you give it a shot.”

She continues gently petting the panther while speaking.

The man then leans slightly forward from behind the couch and says:
“Getting creative is the hardest part.”

The woman turns her attention slightly toward him, then back toward camera and says:
“knoll-jia makes it easy.”

The man then looks toward the camera and says:
“If you want the free guide to follow along…”

The woman finishes:
“…reply nol,  and I’ll send it to you.”

Keep the delivery conversational, confident and understated. Do not make them sound like actors delivering exaggerated commercial dialogue. Natural pauses and subtle facial reactions.

CAMERA:
The camera is operated like the SAME OLD HANDHELD CAMCORDER from the previous scene. Maintain exact visual continuity.

Begin with a loose handheld medium-wide composition containing the woman, panther and man behind the couch. Slight natural camera shake and imperfect framing.

As the woman speaks, make a very subtle crude optical zoom toward her. Let autofocus briefly breathe/hunt before settling on her face.

When the man speaks, naturally shift the camera's attention upward/back toward him with a small handheld reframing movement. Do NOT perform a large cinematic pan.

When the woman speaks again, gently drift the camera back toward her.

For the final “reply Nol” line, settle into a slightly closer handheld framing of the woman and panther.

CAMERA CHARACTER:
Authentic late-1990s/early-2000s consumer camcorder. Strong handheld micro-shake, imperfect framing, tiny operator corrections, slight motion blur, crude optical zoom, autofocus hunting, focus breathing and occasional exposure fluctuation.

LIGHTING:
Same extremely dark mansion lighting as the previous shot. Harsh direct ON-CAMERA FLASH illuminates the characters, black leather, gold jewelry and the panther's glossy fur. Deep black shadows surround the room. Dim warm practical lamps remain visible in the background with subtle cool blue-violet tones.

IMAGE TEXTURE:
Heavy CCD noise, analog grain, optical softness, crushed blacks, blown flash highlights, subtle chromatic aberration, mild lens bloom and imperfect low-light exposure. No modern stabilization, no HDR, no polished Hollywood look.

The final result must feel like the exact SAME handheld camcorder recording continuing naturally from the previous scene.

No cuts. No subtitles. No text overlays. No new characters. No wardrobe changes. Preserve exact likeness and environment. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

提示:將名稱拼寫成發音的樣子. 結尾提示詞將 NOLGIA 寫成 knoll-jia,讓模型能正確讀出。對於任何模型可能誤讀的品牌或名稱,也請這樣做。

注意:時長選項設定長度,而非提示詞. 兩位主持人的提示詞都要求 ONE CONTINUOUS 4-SECOND SHOT,而教堂提示詞則寫著 DURATION: 7 SECONDS。實際影片長度分別為 6、11 和 4 秒:這是時長選項設定的結果。請根據時長選項來安排你的節奏。

所需素材

  • 他的角色設定圖WebP 圖片 · 1600x900 · 152 KB下載
  • 她的角色設定圖WebP 圖片 · 1600x900 · 163 KB下載
  • 豪宅場景圖WebP 圖片 · 900x1600 · 126 KB下載
兩位主持人皆為虛構角色,並為 NOLGIA 生成。

使用 Seedance 2.5 生成

從 建立 選單開啟 影片,並停留在 文字轉影片。House animals 和 Shoot out 片段不使用角色、起始畫面或參考資料,僅使用提示詞;對於主持人,請先在 參考圖片 下新增表格和底板。在模型卡片上按 更改 並選擇 Seedance 2.5,將長寬比選項設為 9:16,時長選項設為 4s,並將品質保留為 720p 且 Audio On 開啟,以便片段自帶聲音。在 專案 下拉選單中選擇你的專案,讓每個鏡頭都存放在那裡,貼上提示詞並按 生成影片。按鈕上會顯示價格,且在執行任何操作之前就會顯示。

影片編輯器,標有六個編號:模型卡片上的 Seedance 2.5、文字轉影片、Audio On、專案下拉選單、4s、9:16 和 720p 選項,以及顯示價格的生成影片按鈕
設定 House animals 片段。 (1) Seedance 2.5。(2) 文字轉影片:無起始畫面,無參考資料,僅使用提示詞。(3) Audio On,以便片段自帶聲音。(4) 每個鏡頭存放的專案。(5) 4s、9:16、720p。(6) 這些設定的價格,在執行任何操作之前顯示。
House animals 場景執行次數保留版本之前的內容
不尋常的遛狗1無:保留第一次生成結果
黑豹(保留兩段)4四個構想,各生成一次:在露台桌下、在烤肉派對、走進院子(保留為額外片段)以及躺在躺椅旁(保留)
生日派對4餐廳場景,接著正式晚餐場景生成兩次,兩次都是蝙蝠突然出現;然後是蝙蝠從第一幀飛出的畫面
泳池派對3鱷魚爬出,接著巨大的鱷魚現身,兩者都設定為驚慌場景;然後是正常大小的鱷魚,無人問津
十二次生成,保留五個片段。每次生成皆使用 Seedance 2.5,僅文字提示,4 秒,720p,9:16。Shoot out 的生成次數計入其專屬章節。

4 秒是 Seedance 2.5 最短的片段長度,在這裡它構成快速剪輯的一個節拍。時間也非常有限:需要鋪陳、揭示、反應和結尾的提示詞,未必能完全容納。生日派對和泳池派對展示了這種情況。每個片段的價格取決於模型、長度和品質;在時間軸上剪輯和匯出不計費。即時費率:

  • Seedance 2.5Pro and up

    Per 5s clip

    480p 26 點 · 720p 56 點 · 1080p 137 點

    1080p renders at 16:9 and 9:16

  • GPT Image 2.5 FlareEvery plan

    Per image

    Native 8 點 · 2K 21 點 · 4K 58 點

頁面載入時從目錄即時讀取。 所有價格都會在生成前顯示。 查看所有費率.

黑豹:什麼都沒發生,然後它走了進來

撫摸的提示詞幾乎沒有動作:她撫摸黑豹,黑豹靠向她的手,看向鏡頭,然後把頭放回去。「沒有其他戲劇性的事情發生。」笑點在於每個人都把真正的黑豹當作家庭寵物,所以提示詞把篇幅花在讓黑豹顯得平凡:放鬆、眨眼、正常大小。

隨意撫摸 (Seedance 2.5)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A completely ordinary suburban American backyard on a warm late afternoon.

An elderly woman in a simple house dress and slippers is sitting comfortably in a lawn chair beside her backyard swimming pool.

A REGULAR-SIZED ADULT BLACK PANTHER is lying peacefully beside her chair.

IMPORTANT: The panther is a realistic NORMAL-SIZED adult panther, approximately the size of a real adult black leopard. NOT oversized, NOT gigantic, NOT a fantasy creature.

The old woman casually pets the panther's head and slowly scratches behind its ears.

The panther is completely relaxed and behaves like a real domesticated animal, calmly leaning into her hand and blinking.

The woman smiles and continues petting it.

The camera operator is standing several meters away filming this like a casual home video.

The old woman occasionally looks toward the camera while continuing to pet the panther.

The panther briefly looks directly into the camera.

Then it calmly puts its head back against the woman's leg.

Nothing else dramatic happens.

The entire scene should feel bizarre only because everyone is treating a real panther like an ordinary household pet.

VISUAL DNA:
Extremely raw handheld consumer-camera home-video footage.
Real human camera operator.
Constant subtle handheld micro-shake.
Uneven human movement.
Slightly crooked framing.
Imperfect amateur composition.
Occasional accidental digital zoom.
Camera operator sometimes moves closer and then stops.
Autofocus naturally hunts between the old woman and panther.
Natural exposure fluctuations.
Realistic animal fur and anatomy.
Realistic panther proportions.
Natural eye reflections.
Realistic breathing and subtle body movement.
Warm natural afternoon sunlight.
Realistic backyard shadows.
Cheap consumer-camera optics.
Slight motion blur.
Subtle rolling shutter.
Digital noise.
Compression artifacts.
Slightly dirty imperfect lens.

NO giant panther.
NO oversized animal.
NO fantasy creature.
NO CGI-looking animal.
NO attack.
NO aggression.
NO dramatic animal behavior.
NO gimbal.
NO Steadicam.
NO drone.
NO perfect tracking.
NO slow motion.
NO polished wildlife cinematography.
NO cinematic lighting.

The footage should look like someone casually filmed their elderly neighbor petting her completely normal-sized panther in her backyard.
一位穿著白色家居連身裙的祖母坐在泳池邊的躺椅上,在傍晚的陽光下,手放在身旁躺臥的黑豹頭上;黑豹看向鏡頭
隨意撫摸。 4 秒,此提示詞的首次生成結果。

額外的好處是 House animals 片段中有一個反轉鏡頭:一位祖母在泳池邊看報紙,一隻黑豹從樹籬後走出來,操作者在祖母發現前就注意到並拉近鏡頭,而祖母又喝了一口飲料。黑豹在片段開始時已經在鏡頭邊緣,反應很小,這正是它合適的原因。在剪輯中,它作為影片的開頭,修剪為 2.4 秒。

黑豹走進來 (Seedance 2.5,選擇了 Strong 的手持效果)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A completely ordinary American suburban backyard in the late afternoon.

An elderly woman in a simple house dress and slippers sits comfortably in a plastic lawn chair beside a swimming pool. She is drinking iced tea and reading a newspaper.

The camera operator films casually from the patio.

Everything feels peaceful.

A huge BLACK PANTHER slowly walks into the backyard from behind a hedge.

The camera operator notices it before the woman does.

The camera immediately becomes nervous and shaky.

The panther silently walks across the grass only a few meters behind the woman.

The operator slowly zooms in.

The old woman finally looks up.

She sees the panther.

Long silence.

Instead of panicking, she casually takes another sip of her drink.

The panther walks past her and sits beside the pool.

The woman looks toward the camera with an expression like this is completely normal.

She goes back to reading her newspaper.

CUT.

No attack. No gore. No exaggerated animal behavior. The panther behaves like a real wild animal.

VISUAL DNA:
Extremely raw handheld consumer-camera footage.
Real human camera operator.
Constant subtle handheld micro-shake.
Uneven framing.
Slightly crooked horizon.
Camera operator reacts nervously and moves backward.
Occasional accidental digital zoom.
Autofocus briefly hunts between the woman and panther.
Natural exposure fluctuations.
Realistic fur detail.
Natural afternoon sunlight.
Realistic shadows.
Cheap consumer-camera optics.
Slight motion blur.
Subtle rolling shutter.
Digital noise and compression artifacts.
Dirty imperfect lens.
No gimbal.
No Steadicam.
No drone.
No perfect composition.
No cinematic tracking.
No slow motion.
No polished Hollywood wildlife cinematography.
No CGI-looking animal.

The footage should feel like a neighbor accidentally captured something completely insane happening in an ordinary backyard. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.
從庭院上方的甲板拍攝:一隻黑豹從灌木发走出,經過一位在池畔躺椅上讀報的祖母,她手中拿著一杯飲料
黑豹走進來。 4 秒,此提示詞的首次執行。

注意:一個詞可能破壞尺寸鎖定. 額外提示詞寫著 "A huge BLACK PANTHER" 且沒有尺寸說明。如果你重複使用,請將 huge 換成撫摸提示詞中的那行:approximately the size of a real adult black leopard, NOT oversized。

生日派對:將所有內容放入第一幀

這個場景的早期版本講述了一個故事:平靜的派對,接著蝙蝠突然飛入,然後恐慌,最後祖母拿著手槍站起來。在 4 秒內,故事無法容納。在下方的第二版正式晚餐草稿中,片段以平靜的餐桌開始,鏡頭快速搖轉,蝙蝠以天花板上的深色模糊影像出現,而祖母從椅子上開火。

草稿:吊燈下的正式燭光晚餐;蝙蝠以天花板上的深色模糊影像呈現,一位身穿晚禮服的祖母從椅子上向上開手槍
草稿: 正式晚餐。以平靜的餐桌開始,蝙蝠以模糊影像進入。
一個擁擠的客廳生日派對,有氣球和蛋糕:蝙蝠飛過賓客頭頂,孩子們從鏡頭前跑過,而蛋糕桌旁的祖母正朝天開槍
保留: 影片開始時蝙蝠已經在飛了。

保留的提示詞說明派對在鏡頭開始前就已經進行:蝙蝠從第一幀就開始飛且從未停止,祖母已經拿著槍,孩子們已經在跑。它在 IMPORTANT 下再次說明,並在 noes 中重申:NO bats suddenly appearing(沒有蝙蝠突然出現),NO bats disappearing(沒有蝙蝠消失)。重複那些影片不能出錯的部分。

生日派對 (Seedance 2.5)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A chaotic suburban American children's birthday party inside a crowded living room during the afternoon.

Colorful balloons everywhere.
Birthday cake on a table.
Paper decorations.
Kids running around.
Parents talking.
Food and drinks on tables.
Bright ordinary household lighting.

FROM THE VERY FIRST FRAME, MULTIPLE REALISTIC BATS ARE ALREADY FLYING THROUGH THE ROOM.

The bats NEVER stop flying.

They constantly circle around the ceiling, weave between balloons, fly above the guests and occasionally swoop through the middle of the room.

The party continues despite the bats.

Kids are running around laughing and playing.

Some adults duck occasionally as bats fly past them.

An elderly grandmother stands in the middle of the room holding a handgun.

She is completely calm.

She keeps looking upward and firing cinematic shots toward the flying bats throughout the entire scene.

The camera NEVER presents the shooting as precise or instructional. The gun is simply part of the absurd fictional chaos.

Each time she fires, the camera operator flinches violently.

Muzzle flashes briefly illuminate the room.

The bats keep flying.

The kids keep running around.

People duck, crouch, move around furniture and continue trying to get through the party.

A kid runs directly across the foreground while a bat flies above them.

Another bat briefly flies between the camera and Granny.

Granny calmly tracks the flying bats and continues firing toward empty areas of the room.

NO BAT IS HIT.
NO BLOOD.
NO GORE.
NO ANIMAL INJURY.

The scene should feel completely chaotic but strangely casual.

IMPORTANT:
The bats must already be present when the video begins.
Do NOT introduce the bats halfway through.
They must remain continuously visible and active throughout the entire clip.
Granny must already be holding the gun and actively shooting during the scene.
Kids must continuously run around the room.
The party must feel like it was already happening before the camera started recording.

VISUAL DNA:
Extremely raw handheld consumer-camera home-video footage.
Real human camera operator.
Constant imperfect handheld shake.
Uneven movement.
Camera operator physically reacts to every gunshot.
Sudden involuntary camera jerks.
Occasional accidental reframing.
Kids randomly cross directly in front of the lens.
People partially block the frame.
Camera sometimes loses Granny and then finds her again.
Autofocus constantly hunts between foreground people, bats and background objects.
Fast-moving bats create realistic motion blur.
Bats have realistic wings, anatomy and unpredictable flight paths.
Natural household lighting.
Exposure briefly clips during muzzle flashes.
Realistic smoke from gunfire.
Cheap consumer-camera optics.
Soft image.
Digital noise.
Compression artifacts.
Subtle rolling shutter.
Dirty imperfect lens.
Realistic motion blur.

NO gimbal.
NO Steadicam.
NO perfect tracking.
NO cinematic composition.
NO slow motion.
NO dramatic Hollywood lighting.
NO polished action-movie look.
NO CGI-looking bats.
NO bats suddenly appearing.
NO bats disappearing.
NO bat being shot.
NO blood or gore.

The final result should feel like an insane family birthday home video that somebody happened to record in real life.

注意:明確指出不能發生的事,就像明確指出必須發生的一樣. 提示詞從未將射擊描述為精準或具指導性,沒有蝙蝠被擊中,也沒有血液、血腥或受傷。每一點都單獨列為一行。

泳池派對:沒人理會

兩份鱷魚草稿都是為了製造恐慌而寫:一隻鱷魚爬出來,或是一隻巨大的鱷魚出現,派對注意到並慌亂起來。第二份結果顯示一隻鱷魚直衝鏡頭,在泳池邊緣填滿畫面,而派對的人坐在遠處觀看。沒有人慌亂。

草稿:一隻鱷魚直衝鏡頭,在泳池邊佔滿畫面,派對人群坐在遠處觀看
草稿:一隻巨大的鱷魚,目的是讓派對人群驚慌逃散。
陽光明媚的後院泳池派對:兩隻鱷魚緩慢游過泳池,經過一個男孩和一個浮具,兩名女子在草地上交談,有人將腳懸在池邊
保留版本:兩隻正常大小的鱷魚,沒人理會。

保留的提示詞去掉了揭露和恐慌。鱷魚從一開始就在泳池裡,長 2.5 到 3 公尺,而且不是巨大,奇怪之處明確寫出:沒人理會。一個小孩游過其中一隻,一個成年人將腳浸入水中,兩名女子繼續交談。冷面笑匠式的場景適合 4 秒;一連串的反應則不適合。

泳池派對 (Seedance 2.5)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A completely ordinary suburban American family pool party on a bright summer afternoon.

A medium-sized backyard swimming pool surrounded by adults and children. People are sitting in lawn chairs, eating barbecue, talking, laughing and relaxing. Kids are playing in the pool.

THERE ARE TWO NORMAL-SIZED ADULT CROCODILES IN THE POOL.

The crocodiles are realistic, approximately 2.5 to 3 meters long, normal real-world crocodile size. NOT GIANT. NOT MONSTROUS. They simply swim slowly through the pool between the people.

The strange part is that NOBODY CARES.

People casually continue swimming, talking and eating as if crocodiles being in the pool is completely normal.

One kid casually swims past a crocodile.

An adult sitting on the pool edge dips their feet into the water while a crocodile slowly swims nearby.

A woman continues talking to another woman without even looking at the crocodile.

The camera operator casually walks around filming the party.

The camera briefly follows one crocodile as it swims past an inflatable pool float.

The crocodile behaves like a completely normal animal: slow, natural swimming, subtle head movement, realistic eyes, realistic water movement.

No attacks.
No screaming.
No panic.
No one reacts dramatically.
No one runs away.
No crocodile behaves aggressively.

The humor and surrealism come entirely from how NORMAL everyone treats the crocodiles.

VISUAL DNA:
Extremely raw handheld consumer-camera home-video footage.
Real human camera operator.
Constant subtle handheld micro-shake.
Uneven footsteps.
Slightly crooked framing.
Imperfect amateur composition.
Occasional accidental digital zoom.
Camera sometimes drifts away from the subject.
Autofocus naturally hunts between foreground people and crocodiles.
Natural exposure fluctuations in bright sunlight.
Realistic water reflections.
Realistic crocodile skin, anatomy and movement.
Natural summer daylight.
Realistic shadows.
Cheap consumer-camera optics.
Slight motion blur.
Subtle rolling shutter.
Digital noise.
Compression artifacts.
Slightly dirty imperfect lens.

NO gimbal.
NO Steadicam.
NO drone.
NO perfect tracking.
NO cinematic camera movement.
NO slow motion.
NO polished Hollywood cinematography.
NO giant crocodiles.
NO fantasy monster crocodiles.
NO CGI-looking animals.

It should look like a completely ordinary family home video where, for some inexplicable reason, normal-sized crocodiles are just part of the pool party.

拍攝結果:那個沒有反應的男人

第二個專案將動物換成動作場景,但保留了那種面無表情的風格。在其中三個提示詞裡,就在一名男子旁邊發生了巨大且暴力的事件,鏡頭慌張移動,但他卻無動於衷:他身後的倉庫爆炸,他繼續行走;一架直升機低空掠過他的屋頂,他戴上太陽眼鏡;他用力踢開車門,然後只是看著它。提示詞明確指出反應才是重點:「But the man DOES NOT react.」

黃昏時分的工業區停車場:一名身穿深色工作服的男子朝鏡頭走來,他身後的倉庫爆炸成火球,他繼續穿過煙霧行走,直到靠近鏡頭
停車場。 4 秒,第一次生成:這個片段入選了。
同一個停車場提示詞第二次生成:一名身穿工作服的男子在黃昏時分穿過停車場朝鏡頭走來,他身後升起火球,他繼續前進直到充滿畫面
同一個提示詞,再次生成。 爆炸和行走的畫面第二次也成功呈現。
多雲午後的郊區街道:一位身穿白色背心的老人站在一輛木紋轎車旁,一架軍用直升機低空掠過他的房屋,揚起滿街的塵土與紙張,鏡頭隨之晃動
直升機。 4 秒。
夜晚便利商店外的日光燈棚下:一名身穿皮夾克的壯碩男子踢向一輛老式轎車的駕駛座車門,畫面瞬間幾乎全黑,接著他看向車門後轉身離開
便利商店。 4 秒。
停車場(Seedance 2.5,選擇了 Strong 的手持效果)
Vertical 9:16 raw handheld consumer camcorder footage, early-2000s low-budget street video aesthetic.

A completely ordinary industrial parking lot in the late afternoon. A huge black sedan is parked beside an abandoned warehouse. A rugged-looking man in dark work clothes slowly walks toward the camera carrying nothing, completely emotionless.

The camera operator is standing across the parking lot filming casually. The framing is imperfect and slightly crooked. The man gets closer.

Suddenly a massive explosion erupts behind him from the warehouse.

The camera violently flinches and jerks sideways from the shock. Autofocus loses the man's face for a moment. Exposure blows out from the fireball. Dust and debris rush across the parking lot.

But the man DOES NOT react.

He keeps walking directly toward the camera through the smoke as if absolutely nothing happened.

The camera operator instinctively backs away, causing uneven footsteps and chaotic handheld shake. The man gets closer until he fills the frame.

RAW REALISM. No heroic slow motion. No polished action cinematography. No dramatic crane shot. No perfect composition.

Cheap consumer camera optics, ugly digital zoom, autofocus hunting, motion blur, blown highlights, muddy blacks, imperfect exposure, natural environmental sound, microphone clipping from the explosion.

The footage should feel accidentally captured, dangerous and completely real. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.
直升機(Seedance 2.5,選擇了 Strong 的手持效果)
Vertical 9:16 raw handheld footage.

A quiet residential neighborhood on a cloudy afternoon.

An elderly man in a simple white tank top stands beside an old luxury sedan parked outside a small house.

The camera operator films him from across the street.

The old man calmly looks toward the sky.

A military-style helicopter suddenly descends extremely low behind the houses.

The rotor wash violently throws dust, leaves and loose papers through the street.

The camera shakes uncontrollably as the operator tries to keep filming.

The old man doesn't move.

He simply puts on a pair of sunglasses.

The helicopter passes directly overhead.

The camera operator instinctively ducks.

When the camera comes back up, the old man is still standing there completely calm.

He looks directly into the camera.

Very short, absurd, intimidating moment.

No heroic cinematography. No smooth tracking. No perfect composition.

Cheap handheld consumer camera, natural cloudy daylight, blown sky, imperfect autofocus, exposure fluctuations, dirty lens, subtle digital noise, real motion blur, unstable framing, accidental zoom.

It should feel like someone randomly filmed an unbelievable moment on the street. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.
便利商店(Seedance 2.5,選擇了 Strong 的手持效果)
Vertical 9:16 raw handheld street footage.

Nighttime outside a grimy convenience store.

A huge older black sedan is parked underneath a flickering fluorescent sign.

A heavily built man wearing a black leather jacket stands beside the car.

The camera operator films from across the parking lot.

The man slowly walks toward the driver's door.

He suddenly kicks the door with enormous force.

The entire car violently rocks sideways.

The camera operator flinches and takes an involuntary step backward.

The man casually looks at the damaged door.

Then he walks away.

No exaggerated superhero movement.

The impact must feel heavy and physically believable.

Camera behavior is the star: sudden reaction shake, imperfect framing, autofocus briefly losing focus during the impact, exposure pumping from the fluorescent lights, accidental digital zoom, slight rolling shutter.

Dirty late-night consumer footage. Realistic skin, realistic clothing, real metal deformation.

No slow motion. No music-video polish. No Hollywood action cinematography.

It should feel like a random person accidentally captured something violent and absurd. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

提示:讓鏡頭呈現男子拒絕表現的反應. 每個提示詞都將戲劇張力轉移到操作者身上:鏡頭劇烈顫抖並橫向甩動,自動對焦失去男子的臉部,曝光因火球而過曝,操作者在直升機飛過時閃避。「鏡頭行為是主角,」便利商店的提示詞寫道。男子越冷靜,鏡頭的反應就應該越強烈。

外景拍攝:單鏡到底,鏡頭顫抖

其餘三個場景都是普通房間裡的單發槍聲:午夜的廚房、打烊後的唱片行、老式電梯。每個提示詞花在鏡頭對攝影機產生什麼影響上的篇幅,都比花在鏡頭本身上的篇幅還多:槍口閃光在瞬間讓高光過曝,自動對焦鎖定在煙霧而不是男人身上,操作者反應稍微遲了一點,並向後猛退。

夜間螢光燈管下的小廚房,從走廊拍攝:一個在吧檯抽菸的男人拿起手槍,朝旁邊黑暗的房間開了一槍,然後轉回身面向攝影機,嘴裡還叼著菸
廚房。 4 秒。
打烊後的唱片行:一個穿黑衣的男人站在唱片架之間看著攝影機,接著閃光出現,攝影機掉落到一個唱片架後面
唱片行。 4 秒。
一個穿皮夾克的男人站在昏暗燈光下的老式電梯裡,當刮傷的金屬門滑上關閉時,他舉起手槍指向鏡頭,攝影機則停留在關閉的門上
電梯。 4 秒。提示詞將鏡頭瞄準一個看不見的目標;在影片片段中,他則是瞄準鏡頭。
草稿:夜晚的空蕩蕩餐館,紅色卡座,窗上有雨;一名男子整段影片都坐在遠處的卡座裡,鏡頭只搖擺一次
草稿:餐館。提示詞描述的槍戰從未發生。
廚房(Seedance 2.5,Handheld 選取為 Strong)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A tiny apartment kitchen at midnight.

Old refrigerator humming. Dirty dishes. Harsh overhead fluorescent tube.

A man stands at the kitchen counter smoking a cigarette.

The camera operator films him from the hallway.

Suddenly a loud gunshot comes from another room.

The man freezes.

He looks toward the doorway.

The camera operator takes a nervous step backward.

The man grabs a handgun from the counter and fires ONE shot toward the unseen room.

Huge muzzle flash.

The camera violently jerks upward.

Smoke briefly hangs in the kitchen.

The man slowly turns toward the camera.

He takes another drag from his cigarette.

CUT.

No injury or gore.

VISUAL DNA:
Cheap handheld consumer-camera footage.
Human operator.
Constant subtle shake.
Uneven footsteps.
Crooked framing.
Slight camera breathing.
Natural focus hunting.
Autofocus briefly locks onto smoke instead of the subject.
Muzzle flash temporarily blows out highlights.
Realistic motion blur.
Low dynamic range.
Dirty lens.
Digital noise.
Compression artifacts.
Practical fluorescent lighting.
No artificial cinematic lighting.
No gimbal.
No Steadicam.
No perfect framing.
No slow motion.
No polished action cinematography.
No glossy commercial aesthetic. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.
唱片行(Seedance 2.5,Handheld 選取為 Strong)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

An underground 1990s record store after closing.

Stacks of vinyl records. Posters on the walls. Fluorescent ceiling lights. Old stereo equipment.

A mysterious man in oversized black streetwear stands between the shelves.

The camera operator slowly approaches him.

He looks directly at the camera.

Suddenly someone off-screen fires a shot.

The camera snaps violently toward the sound.

Records shake on the shelves.

The man calmly pulls a handgun and fires once toward an unseen direction.

The camera operator ducks behind a shelf.

The camera slowly rises again.

The man has disappeared.

Only the spinning record player remains.

CUT.

VISUAL DNA:
Authentic 90s consumer-camera feeling.
Very handheld.
Unstable framing.
Cheap optical zoom.
Focus hunting.
Slightly soft image.
Natural fluorescent flicker.
Ugly exposure.
Muzzle flash briefly overwhelms the camera sensor.
Realistic motion blur.
Subtle VHS-like softness without fake VHS overlays.
Natural film grain and digital noise.
Dirty lens.
No timestamp.
No REC icon.
No text overlays.
No stabilization.
No cinematic crane.
No polished music-video look.
No slow motion. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.
電梯(Seedance 2.5,Handheld 選取為 Strong)
SEEDANCE 2.5 VIDEO PROMPT - 9:16

A brutal, raw underground crime-film moment captured on a cheap handheld consumer camera.

Inside a cramped old apartment-building elevator at night. Dirty mirrored walls, scratched metal doors, flickering fluorescent ceiling light.

A tough-looking man in a black leather jacket stands silently inside the elevator, staring straight ahead.

The camera operator is standing outside the elevator filming him.

Just as the doors begin closing, the man suddenly pulls a handgun and FIRES one shot toward an unseen target outside the frame.

The muzzle flash violently illuminates the tiny elevator.

The camera operator instinctively jerks backward and almost loses the framing.

The elevator doors slam shut.

The camera stays pointed at the doors for a beat.

Then slowly steps closer.

CUT.

The gunshot is sudden and shocking, not glorified. No instructional firearm handling. No visible injury or gore.

VISUAL DNA:
Very raw handheld consumer-camera footage.
The camera operator is a real human holding a small cheap camera.
Constant subtle handheld micro-shake.
Uneven human movement.
Imperfect framing.
Camera reacts slightly late to the gunshot.
Violent involuntary camera jerk at the muzzle flash.
Autofocus briefly loses focus.
Exposure blows out for a fraction of a second from the muzzle flash.
Natural motion blur.
Subtle rolling shutter.
Cheap optical characteristics.
Slight digital compression and image noise.
Dirty imperfect lens.
Flickering practical fluorescent lighting.
No gimbal.
No Steadicam.
No polished Hollywood composition.
No slow motion.
No dramatic crane movement.
No glossy CGI appearance.

Make it feel like illegal underground footage somebody accidentally recorded. Camera: aggressive handheld, the frame shaking and lurching with the action, quick reframes and an urgent run-and-gun feel.

注意:虛構內容,非操作指南. 電梯提示詞用一行說明:「槍聲突然且令人震驚,而非美化。無教學式槍械操作。無可見傷害或血腥。」廚房提示詞在 CUT 之後加上「無傷害或血腥」。將這些行寫入每個包含武器的提示詞中。

六段 Shoot out 影片來自一次包含十七次 Seedance 2.5 執行的工作階段:保留的六段、第二次停車場執行,以及十次未採用的想法執行。其中之一是餐館,位於上方網格的最後:凌晨兩點的餐館,外面的槍聲本應引發槍戰。影片從未達到那個階段;男子整段時間都坐在他的卡座裡,鏡頭只搖擺一次。在評估其他任何事情之前,請觀看每一段_take_以確認你要求的動作。

評分並剪輯成一支影片

House animals 使用一條在剪輯中鋪設的音軌。Shoot out 則使用在 NOLGIA 製作的配樂:一首快節奏、陰暗的 90 年代初期 boom-bap 器樂曲,帶有稀疏的鋼琴 riff 和幾句 ad-libs,與影片片段處於相同的地下 90 年代氛圍。在剪輯中,配樂幾乎完全取代了片段本身的聲音。

Shoot out 配樂 (Lyria 3.5)
Fast, underground early-90s hip-hop instrumental, around 100-106 BPM. Dark, mysterious, badass boom-bap energy. Hard dusty kick and snare, tight swung breakbeat, deep gritty bassline, raw vinyl texture, subtle tape hiss and analog saturation. Main musical element is a haunting, sparse minor-key piano riff - hypnotic, cold, memorable, with occasional chopped piano stabs and reversed piano textures.

Instrumental-focused arrangement with NO full rap verse and NO lead vocal. Only occasional short male hip-hop ad-libs used as rhythmic texture: “ayo,” “uh,” “yeah,” “yo,” “what,” “huh,” scattered naturally throughout the beat, sometimes distant or whispered, sometimes layered over drum hits. Keep the ad-libs sparse - they should feel like part of the production, not a song vocal.

Raw underground 90s New York feel, mysterious nighttime atmosphere, confident and intimidating, dusty sample-based production, imperfect human groove, gritty cassette/vinyl character. Minimal, hypnotic, fast-moving, hard-hitting. No modern trap, no singing, no full vocals, no polished pop production.

從 建立 選單開啟 音訊,在 你需要什麼? 下選擇 音樂,挑選 Lyria 3.5,貼上提示詞並按下 生成音訊。沒有長度設定:這次生成的結果約為兩分半鐘,剪輯使用了從約 0:55 開始的 19.5 秒。挑選最有力的段落,並在時間軸上修剪至該長度。

  • Lyria 3.5Every plan

    Per generation

    Standard 5 點

頁面載入時從目錄即時讀取。 所有價格都會在生成前顯示。 查看所有費率.

House animals 以 9:16 比例播放 18 秒:黑豹走進來,修剪至 2.4 秒,接著是撫摸、生日派對、泳池派對和鬣狗,每段約 4 秒,最後 2 秒覆蓋 NOLGIA 標誌。它在每個片段上鋪設粗糙的顆粒感,比片段本身更重,並為整個剪輯鋪設一條音軌,取代大部分片段本身的聲音。

Shoot out 長 19.5 秒,剪輯節奏較快,每個鏡頭 1.5 到 4 秒:停車場、直升機、唱片行、戴金鍊的老人、廚房、便利商店和電梯,NOLGIA 標誌出現在關閉的電梯門上。戴金鍊老人的鏡頭不在專案中,因此這裡沒有它的提示詞。

兩個剪輯版本都是在 NOLGIA 之外編輯,然後上傳到各自的專案。若要在 NOLGIA 中進行剪輯:

  1. 開啟專案的 Workspace 並按 新時間軸,命名後將畫面尺寸設為 直式 9:16,然後按 建立。
  2. 在 專案媒體 中依順序新增片段:將每個片段拖曳到時間軸上,或將滑鼠移到片段上並按 + 以在播放頭位置新增。
  3. 將每個片段修剪至所需的節奏:使用 選取 工具拖曳片段邊緣,或在新增前於預覽中設定 標記入點 和 標記出點。
  4. 若要讓片段看起來像同一捲底片,請選取一個片段,在檢查器中開啟 調色 並選擇一個 底片 預設集,例如 Fuji Superia 400。編輯器沒有顆粒控制功能,因此顆粒效果需透過提示詞或在其他編輯器中添加。
  5. 聲音方面,保留每個片段原有的音訊,或是在工具列的 音訊 群組中點擊 新增音訊軌道,將配樂放在獨立軌道上,在 音訊 檢查器中設定其 片段音量,並在結尾加上 淡出。
  6. 按下 匯出影片。完成的 MP4 檔案會存入你的媒體庫。

音訊指南 更詳細地說明了時間軸上的音樂與音效處理。

規則

  • 每個片段只聚焦一個核心概念。
  • 讓周圍的環境保持平凡日常。
  • 使用自然的人類動作與真實的比例。
  • 將攝影機視為由人手持的物理物件,並描述它的動作。
  • 讓質感輔助真實感,而非取代真實感。
  • 將所有元素放在第一幀,並重複那些片段不能出錯的關鍵細節。
  • 當笑點在於無人反應時,使用面無表情的反應,並讓攝影機來做出反應。
  • 一旦世界觀與動作清晰,就讓 Seedance 2.5 處理動態。

目標不是完美的電影鏡頭,而是讓觀眾相信有人不小心錄下了他們本不該看到的東西。

常見問題

這些影片是由哪些模型製作的?
每個片段皆使用 Seedance 2.5,解析度為 720p,比例為 9:16,並開啟音訊。House animals 和 Shoot out 片段僅使用文字提示。NOLGIA 提示詞主持片段附加了兩張角色設定圖與一張使用 GPT Image 2.5 Flare 製作的背景板。Shoot out 的配樂由 Lyria 3.5 生成。
這些人和動物是真的嗎?
不是。片段中的所有人物與事物皆由 Seedance 2.5 根據本指南中的提示詞生成。未使用任何真實人物或動物的照片或影像素材。
為什麼提示詞中的某些部分使用大寫字母?
這些大寫字母標示出片段中不能出錯的關鍵行:動物的體型、片段開始時已發生的動作、男子不做出反應,以及絕對不能發生的情況。我們尚未測試大寫字母本身是否會改變結果;從草稿到最終保留的片段之間,關鍵在於明確地陳述這些要求。
片段長度可以超過 4 秒嗎?
可以,Seedance 2.5 可生成 4 到 30 秒的片段,而 NOLGIA 提示詞片段的長度為 4 到 11 秒。無論提示詞內容如何,時間軸上的時長選項決定了片段的長度。這裡的大多數片段較短,因為每個片段都是快速剪輯中的一個節奏點,而較長的片段需要更多內容來填充。
我可以將場景設定在不同的年代嗎?
可以。請指明年代與對應的攝影機,例如 1990 年代初期的家用攝影機或 2000 年代初期的數位手持攝影機,並保持場景、服裝和汽車都來自同一時期。
影片片段是否包含聲音?
開啟「音訊」後,Seedance 2.5 會連同畫面一起回傳聲音,這些提示詞也會要求生成聲音:低吼聲、槍聲、爆音導致麥克風削波。兩段最終剪輯大多以單一音軌取代原始聲音:「House animals」在剪輯中鋪設了音軌,「Shoot out」則搭配 Lyria 3.5 配樂。

製作你的第一支影片

開啟影片編輯器

影像製作

如何製作逼真的 AI 影片

製作逼真 AI 影片的分步工作流程:使用 NOLGIA Agent 進行規劃,鎖定角色與場景,為 Seedance 2.5 逐拍撰寫每個鏡頭,然後進行配樂與剪輯。

· 閱讀時間 21 分鐘

提示詞

鏡頭運動:每種運動的作用與對應的提示詞

用淺顯易懂的文字介紹十二種鏡頭運動:說明鏡頭如何移動、引發什麼感受、在提示詞中該寫什麼,以及容易與之混淆的運動。每種運動都附有在 NOLGIA 製作的片段。

· 閱讀時間 26 分鐘