如何使用 Kling AI 分 7 步製作短片

要使用 Kling AI 製作短片,請給它一個完成的起始幀,然後將每個鏡頭寫成一個鏡頭移動、一個動作和一句台詞,並附上要避免的事項清單。

一個人用三腳架上的電影攝影機拍攝的墨水畫

大多數第一次使用 Kling 的影片都會漂移:不同鏡頭之間臉部會改變,鏡頭會游移,而且每次聲音都不同。Karan Archana 製作了《一滴水》(Drop of Water),這是一部 3:21 的旱災劇情片,為 2026 年夏季 Flick 駐村計畫而作,在一個 Flick 畫布上使用了 89 個 Kling 3.0 4K 片段。我們閱讀了每一個提示詞及其設定。本指南說明如何像 Karan 一樣使用 Kling AI,分為七個步驟,每個步驟都附有影片中的真實鏡頭和生成該鏡頭的確切提示詞。

Flick 中的 Kling 影片模型需要 Standard 方案及以上版本。

如何逐步使用 Kling AI 製作短片?

從角色設定表到固定的最終鏡頭,七個步驟,全部來自《一滴水》的工作流程。
  1. 第 1 步

    建立角色設定表

    在製作任何影片之前,為每個角色製作一張旋轉視圖設定表,這樣每個 Kling 起始幀都能使用相同的臉部和服裝。

    • Ken 在純白背景上的六個面板:正面、兩側側面、背面和兩張臉部特寫。
    • 使用 GPT Image 2 從他先前的三張肖像製作而成。
    • Karan 為 Evelyn 製作了相同的設定表,並為每個角色製作了表情設定表。
    Character turnaround sheet for Ken from Drop of Water: front, both profiles, back and two face close-ups on white
    Karan Archana 執導的《Drop of Water》 · GPT Image 2
    Professional character turnaround reference sheet for Ken: a late 30s South Asian man, medium brown skin, lean athletic build around 5'10", short slightly wavy black hair, sharp tired intelligent eyes, strong jaw, wearing dark slim-fit cargo trousers, a fitted grey technical shirt, lightweight jacket, heavy duty boots. Create ONE wide image showing SIX panels: full body front view, full body left profile, full body right profile, full body back view, AND one close-up of face from the right side, AND one close-up of face from the left side. Same character in every panel, 100% identical face and clothing matching the reference exactly, neutral A-pose for full body panels, plain pure white seamless background, even flat lighting, orthographic framing. No scene, no props, no other people.
    含義:肯恩角色設定圖展示南亞裔男子六個角度,正面側面背面全身及左右臉部特寫,白背景平光。

    開啟 Drop of Water 工作流程

  2. 第 2 步

    製作開始畫面

    Kling 的圖像轉影片功能從你的圖片開始,因此 Karan 先確定每一幀,從先前鏡頭的靜止畫面重新取景,並附上 Ken 的兩張設定表。

    • 圖像 1 是場景,圖像 2 和 3 是設定表。
    • 一個變化:Ken 的特寫鏡頭,Evelyn 已經離開。
    • 使用 Nano Banana Pro 製作;全部 89 個 Kling 片段都從這樣的圖像開始。
    Close-up of Ken in glasses drinking green liquid from a clear sachet in a dim dusty room, a start frame from Drop of Water
    Karan Archana 執導的《Drop of Water》 · Nano Banana Pro
    Image 1 is the scene: a two-shot of the man Ken (left, glasses, brown vest over grey tee, curly hair, beard) drinking green liquid from a transparent sachet in a dim dusty room, with the woman Evelyn on the right. Images 2 and 3 are reference sheets of Ken — use them to keep his face 100 percent accurate: exact features, glasses, beard, hairstyle, skin texture. Re-frame this into a tight close up on Ken ALONE, filling the frame, in the exact same moment of raising the transparent sachet to his lips and drinking, eyes lowered. Evelyn is completely gone from the frame and must not be visible at all. Keep everything from image 1 identical: Ken's identity, his clothing, the sachet with green liquid, the dim dusty room, the warm muted color palette, window light, lighting mood and photorealistic texture and grain. Shallow depth of field, background soft. Grounded cinematic realism.
    含義:昏暗房間特寫肯恩獨自舉起透明袋裝綠色液體飲用,眼神低垂,淺景深寫實電影質感。
  3. 第 3 步

    指定一個鏡頭運動

    描述一個鏡頭運動,包括起點、速度和終點,這樣 Kling 就不會自行創造推進或切換鏡頭。

    • 單一橫搖,從左到右,沿著圍欄移動。
    • 速度有明確說明:約為正常橫搖的三分之一。
    • 招牌進入畫面,停留片刻後離開。
    • 長度設定在提示詞旁邊:Kling 3.0 4K 可選擇 3、5、10 或 15 秒。
    Slow pan along a barbed wire fence to a rusted Entry to Northland is prohibited sign under a red desert sky, from Drop of Water
    第一幀:Slow pan along a barbed wire fence to a rusted Entry to Northland is prohibited sign under a red desert sky, from Drop of Water
    第一幀
    00:10
    Karan Archana 執導的《Drop of Water》 · Kling 3.0 4K · 圖像轉影片
    The camera begins at the far left end of the barbed wire fence, framing only the rusted wire strands stretching across the blood-orange desert horizon. A very slow, almost glacial pan moves right along the fence line — the wire fills the frame, repeating rhythmically, the cracked earth below and the smoky red sky above. The pan speed is extremely slow — the camera should move only about one third of the distance it would in a normal pan over the full 5 seconds. The fence seems endless. Then, gradually, the weathered wooden post and the rusted "ENTRY TO NORTHLAND IS PROHIBITED" sign drift into frame. The camera barely lingers on the sign for a beat — just long enough to register its presence — then continues its imperceptibly slow pan to the right. The sign drifts out of frame. Only fence wire remains. The world is silent, red, and still. No rush. No acceleration. The entire shot breathes slowly.
    含義:鏡頭緩慢沿著鐵絲網向右平移,橘紅沙漠天空下,禁止進入北地的生鏽標誌短暫入鏡後消失,畫面靜謐荒涼。
  4. 第 4 步

    列出要避免的內容

    將提示詞分為場景、允許動作的簡短清單,以及 AVOID 清單,這樣 Kling 就能保持安靜的表演維持安靜。

    • 動畫清單有五行,每行一個動作。
    • AVOID 清單禁止表演選擇以及鏡頭運動。
    • 影片中 89 個 Kling 提示詞裡有 59 個以這樣的清單結尾。
    Ken holds a crumpled notice and a water pouch and turns his head toward Evelyn in a dusty room, from Drop of Water
    第一幀:Ken holds a crumpled notice and a water pouch and turns his head toward Evelyn in a dusty room, from Drop of Water
    第一幀
    00:05
    Karan Archana 執導的《Drop of Water》 · Kling 3.0 4K · 圖像轉影片
    Prompt
    
    Continue inside the same weathered house immediately after Ken lowers the government-issued water ration pouch. Maintain identical lighting, wardrobe, production design, and atmosphere. Ken stands near the back sliding door, his body still facing the window. After a long moment of silence, he slowly turns his head just enough to acknowledge Evelyn without fully facing her. His expression is calm, compassionate, and tired. He speaks quietly and gently, without anger or urgency. Evelyn stands a few feet behind him, listening in silence. She remains almost perfectly still. The camera is locked in a medium shot with an almost imperceptible slow push forward. Natural breathing, subtle eye movement, restrained body language. Photorealistic live-action film, grounded realism, naturalistic cinematography, practical production design, documentary authenticity.
    
    Animation
    Ken slowly turns his head.
    Speaks quietly.
    Evelyn listens without moving.
    Dust drifts in the light.
    No other movement.
    
    AVOID :
    
    Dramatic acting, exaggerated emotion, crying, shouting, large gestures, camera shake, handheld movement, camera pan, zoom, cinematic swoops, cartoon style, text, logos, AI artifacts.
    含義:昏暗屋內肯恩慢慢轉頭輕聲說話,伊芙琳靜靜站立聆聽,灰塵在光線中飄浮,寫實自然主義風格。
  5. 第 5 步

    將台詞加上引號

    將確切的台詞加上引號並描述說話方式,因為 Kling 3.0 4K 會隨著畫面一起生成語音。

    • 一句台詞,五個字,前面加上呼吸和搖頭的動作。
    • 描述了說話方式:平靜、低沉、實事求是。
    • 台詞後的停頓,寫入提示詞中。
    • 聲音來自「生成音訊」設定,預設為開啟。
    Over the shoulder shot of Evelyn in a lamplit room telling Ken they are not coming back, from Drop of Water
    第一幀:Over the shoulder shot of Evelyn in a lamplit room telling Ken they are not coming back, from Drop of Water
    第一幀
    00:05
    Karan Archana 執導的《Drop of Water》 · Kling 3.0 4K · 圖像轉影片
    Use the attached image as the exact first frame. Preserve the exact composition, characters, clothing, lighting, colour palette, room, lens, camera position and cinematic realism. Continue directly from the previous shot.
    
    Preserve Evelyn's existing voice exactly as established in the previous dialogue scenes. Use the same voice, accent, tone, cadence, pitch, emotional restraint and vocal characteristics already heard in the film. Do not generate a new voice or alter her vocal performance in any way. This line should sound like it is spoken by the same Evelyn from the preceding scene.
    
    Evelyn maintains eye contact with Ken in silence. Her expression is weary, composed and quietly devastated. She takes one small breath, gives the slightest shake of her head and softly says:
    
    "They're not coming back."
    
    Her delivery is calm, low and matter-of-fact. She is not trying to convince Ken—she has already accepted what he cannot. After speaking, she remains still, holding his gaze for two seconds without speaking.
    
    Ken remains softly out of focus in the foreground and barely moves. The camera stays completely locked off. Natural lip synchronization, subtle breathing, one slow blink, grounded cinematic realism.
    
    Avoid
    
    Do not change Evelyn's voice. Do not synthesize a different accent, pitch, cadence or emotional style. Do not make her sound younger, older, robotic or theatrical. Do not alter her face, hairstyle, clothing, posture or eyeline. Do not change Ken, the room, lighting, framing or colour grade. No camera movement, zoom, push-in, rack focus or reframing. No crying, shouting, trembling lips, exaggerated facial expressions or dramatic gestures. Do not add dialogue before or after the exact line. Keep Ken softly out of focus and Evelyn in sharp focus throughout.
    含義:伊芙琳疲憊平靜地注視肯恩,輕聲說他們不會回來了,接著沉默對望,肯恩在前景失焦,寫實電影風格。
  6. 第 6 步

    切換到回應鏡頭

    將對話拍成每句一個 Kling 片段,每個都從新的過肩鏡頭開始,並要求已經聽到的語音。

    • 上一步的反向角度,從先前拍攝的靜止畫面製作而成。
    • 提示詞重複了語音指示和固定鏡頭。
    • Karan 對用詞的註記仍保留在提示詞中,按原文引用。
    Over the shoulder shot of Ken in glasses quietly asking Evelyn what if they come home, from Drop of Water
    第一幀:Over the shoulder shot of Ken in glasses quietly asking Evelyn what if they come home, from Drop of Water
    第一幀
    00:05
    Karan Archana 執導的《Drop of Water》 · Kling 3.0 4K · 圖像轉影片
    Use the attached image as the exact first frame. Preserve the exact composition, camera angle, lens, lighting, colour palette, production design, costumes and photorealistic cinematic style. Continue directly from the previous dialogue shot.
    
    Preserve Ken's existing voice exactly as established in the previous dialogue scenes. Use the same voice, accent, cadence, pitch, emotional restraint and vocal characteristics already heard in the film. Do not generate a new voice or alter his vocal performance in any way. This line must sound as though it is part of the same continuous conversation.
    
    Ken maintains eye contact with Evelyn, who remains softly out of focus in the foreground. He hesitates for a moment, searching her face before quietly asking:
    
    "What if they come home... and no one is here?"
    
    His voice is gentle, tired and quietly hopeful. He is not arguing with Evelyn. He is holding on to the last possibility that their daughters might still return. The line is spoken softly, with a small pause after "come home..." as he struggles to finish the thought.
    
    After speaking, Ken continues looking at Evelyn in silence. He takes one subtle breath, blinks once naturally and remains completely still. Hold the silence for one to two seconds after the line finishes.
    
    The camera remains perfectly locked off. Natural lip synchronization. Subtle breathing. Grounded cinematic realism.
    
    Performance Notes
    Quiet, restrained delivery.
    No anger.
    No pleading.
    No melodrama.
    His hope is fragile rather than desperate.
    Let the silence after the line carry as much emotion as the dialogue itself.
    Avoid
    
    Do not change Ken's voice, accent, pitch or cadence. Do not alter Ken's face, hair, glasses, clothing, posture or eyeline. Do not change Evelyn's silhouette, the room, fireplace, lighting, framing or warm amber colour grading. No camera movement, zoom, push-in, pan or rack focus. No tears, shouting, dramatic gestures, exaggerated facial expressions or theatrical acting. Do not add dialogue before or after the exact line. Preserve the intimate, quiet realism of the surrounding scene.
    
    I also think the wording "come home" is stronger than "come back." Earlier, Evelyn says, "They're not coming back." Then Ken asks, "What if they come home... and no one is here?" The repetition of the same idea from opposite emotional perspectives gives the exchange a quiet symmetry without feeling repetitive.
    含義:肯恩與失焦的艾芙琳對視輕聲問女兒回家時若無人在怎麼辦,電影風格寫實靜謐,溫暖琥珀色調。
  7. 第 7 步

    修正單一鏡次,保留其餘部分

    當一個好鏡次中只有一個物件出錯時,使用 Kling O3 Pro 編輯該片段而非重新生成,並列出所有必須保留的內容。

    • Ken 手中的小袋在鏡頭之間發生了變化,因此編輯只替換小袋,不改變其他任何東西。
    • 提示詞列出了要保留的內容,從構圖到對話時機。
    • 編輯會保留素材的原始聲音,並按來源片段的秒數計費。
    Ken turns toward Evelyn in a dusty kitchen holding the corrected water sachet, a Kling O3 Pro video edit from Drop of Water00:10
    Karan Archana 執導的《Drop of Water》 · Kling O3 Pro
    Edit only the empty water sachet in Ken's left hand. Do not change anything else.
    
    This is a continuity correction.
    
    Replace the current sachet with the same almost-empty transparent plastic water sachet containing only a small amount of bright green liquid exactly as it appears in the previous shot (reference image attached).
    
    The sachet must:
    
    remain in Ken's left hand throughout the entire shot
    have the same size, shape, folds, transparency, remaining liquid level and orientation as the previous shot
    move naturally with his hand as he turns toward Evelyn
    never disappear, refill, change color or change shape
    
    Do not alter anything else.
    
    Preserve exactly:
    
    camera framing
    composition
    lens
    lighting
    warm amber color palette
    Ken and Evelyn's faces
    body positions
    clothing
    dialogue timing
    background
    room
    facial expressions
    pacing
    
    The only change should be replacing the incorrect sachet with the matching continuity sachet from the previous shot.
    
    This is a precision continuity edit, not a regeneration. Maintain cinematic realism in the style of Denis Villeneuve and Children of Men.
    
    One additional suggestion: attach two reference images if Flick allows it.
    
    The last frame of the previous shot, where the sachet is correct.
    The first frame of the current shot, where everything is correct except the sachet.
    
    That gives Flick a very explicit "copy from A, paste into B" task, which generally produces more reliable continuity edits than relying on a text description alone.
    含義:修正肯恩左手中幾乎空的透明水袋使其符合前一鏡頭中剩餘少量綠色液體的連戲細節,其餘畫面不變。

    開啟 Drop of Water 工作流程

真實短片使用了哪些 Kling 設定?

帶聲音的短片段:Drop of Water 中 89 個 Kling 3.0 4K 片段裡有 57 個為 5 秒長度,且全部 89 個都會生成音訊。
公開 Drop of Water 畫布上的 89 個 Kling 3.0 4K 片段,讀取於 2026-10-26。一個提示詞可能出現在多個列中。
片段要求的內容片段
5 秒長度57
10 秒長度28
15 秒長度4
已生成音訊89
提示詞中包含避免清單59
鎖定或靜止的鏡頭51
從前一個鏡頭繼續50
引號中的對白台詞29
包含結束幀和開始幀1
更多細節:我們如何計算

我們讀取了儲存在影片公開畫布上的每個 Kling 3.0 4K 片段的設定和提示詞。上傳的內容和已刪除的拍攝不計入統計。

  • 避免清單:提示詞中的「avoid」一詞。
  • 鎖定鏡頭:「locked」、「no camera movement」或「the camera remains completely still」。
  • 延續:「previous shot」、「last frame」、「final frame」或「continue」。
  • 台詞:提示詞中的一對雙引號。
  • 提示詞的中位數長度為 176 個字。

關於使用 Kling AI 的問題

什麼是 Kling AI?

Kling 是快手旗下的 AI 影片模型系列,能將起始影像和文字提示詞轉換為影片片段,其中許多還附帶聲音。在 Flick 中,Kling 與影像模型在同一畫布上運作,因此角色設定圖、起始畫面和片段可以並排顯示。

我可以在 Flick 中使用哪些 Kling 模型?

圖像轉影片:Kling 3.0 4K、Kling 3.0 Pro、Kling 3.0、Kling O3 Pro、Kling O3 和 Kling 2.6 Pro。最新的 Kling 4.0 於 2026 年 10 月 26 日加入 Flick,在《Drop of Water》製作完成之後,因此本指南中的片段使用的是 Kling 3.0 4K。Flick 還提供 Kling O3 Pro Multishot(從參考影像生成多鏡頭片段)、Kling AI Avatar v2(用於說話頭像)、Kling O3 Pro 影片編輯以及 Kling Image V3(用於影像生成)。

Kling AI 可以免費使用嗎?

在 Flick 中使用 Kling 影片需要 Standard 方案或更高方案,每月 20 美元起。Kling Image V3 在 Free 方案中提供,註冊 Flick 時會贈送 100 點數。

在 Flick 中生成 Kling 影片需要多少點數?

Kling 3.0 4K 片段在預設長度下需要 330 點數,最長長度下需要 990 點數。Kling O3 Pro 編輯按您編輯的片段秒數計費。

Kling 影片可以多長?

Flick 中的 Kling 3.0 4K 片段可執行 3、5、10 或 15 秒。若要製作更長的場景,請從前一個片段的最後一幀開始下一個片段,就像 Drop of Water 串聯其鏡頭的方式。

Kling AI 可以製作有聲對白嗎?

可以。Kling 3.0 4K 會隨著畫面生成音訊,預設開啟「生成音訊」設定。用引號寫出確切台詞、誰說的、怎麼說,並描述台詞前後的靜默。

如何在每個 Kling 鏡頭中保持相同角色?

先製作角色設定圖,然後用影像模型從中建立每個起始幀,最後再用 Kling 製作動畫。起始幀承載了臉部特徵,因此影片模型無需猜測。

Kling 可以使用起始幀和結束幀嗎?

Kling 3.0 4K 可以選擇性地使用結束幀:可以。給它鏡頭的第一張和最後一張圖片,Kling 就會製作兩者之間的動作。

更多關於 Kling 和 Flick 中 AI 電影製作的資訊

本指南中的影片

本頁面上的每個鏡頭都來自公開的 Drop of Water 畫布,版權屬於其製片人 Karan Archana。提示詞完全按照畫布上的原文引用。

在 Flick 中製作你的第一部 Kling 短片

角色設定圖、起始幀和所有 Kling 影片模型都在同一個畫布上,就像 Drop of Water 的製作方式。

在 Flick 中製作 Kling 影片

免費試用,無需信用卡。

讓 AI 解讀本頁:ChatGPTPerplexityClaude