AI 角色語音生成器

使用文字轉語音為每個角色製作語音台詞,然後讓每個鏡頭都以相同的聲音說話,與您的影片在同一個畫布上。

一幅墨水畫,畫中導演坐在導演椅上,向兩位正在交談的演員遞出場記板,旁邊有一盞架在支架上的影視燈

一個角色在每個鏡頭中看起來相同,但聲音卻不同,這比糟糕的剪輯更快破壞一部影片,而大多數語音工具只能產生永遠無法進入畫面的音訊檔案。AI 角色語音生成器透過為每個角色賦予一個可在每個鏡頭中重複使用的聲音,或將聲音精確寫入提示詞中,讓視訊模型以相同方式建構它來解決這個問題。我們閱讀了 4 個 Eleven v3 語音台詞和 15 個來自 8 部 Flick 製作影片的說話鏡頭背後的提示詞,其中大部分來自 2026 年夏季電影製作駐村計畫。本頁展示每個台詞和鏡頭及其精確提示詞、它們共享的六個習慣,以及為您自己的角色賦予聲音的四個步驟。

描述角色和聲音

描述角色、台詞以及聲音應有的效果,或從下方選擇一個真實範例,它會在新 Flick 專案中作為第一個請求開啟。

什麼是 AI 角色語音生成器?

AI 角色語音生成器可將書面台詞轉換為語音,讓角色擁有專屬聲音,使同一個人在每個場景中都保持相同的聲音。
根據您的最終需求,選擇 Flick 的角色配音工具
您需要在 Flick 中使用功能說明
語音台詞的音訊片段文字轉語音,使用 Eleven v3 或 Eleven v4為一位或多位說話者朗讀台詞,並可加入如 [relief] 等音訊標籤來表達情緒
使用現有聲音生成新的語音文字轉語音,使用 Seed Audio 1.0從您附加的最多三個參考片段中複製聲音
角色以該聲音說話的鏡頭全方位參考,使用 Seedance 2.0 或 MiniMax H3以語音片段作為參考,同時生成畫面和語音
說出錄製台詞的靜態肖像角色對話,使用 Kling AI Avatar v2根據您提供的音訊,讓圖片中的臉部動起來

語音台詞如何變成說話鏡頭?

使用文字轉語音生成語音行,將其附加到 Omni 參考鏡頭,並在提示詞中命名,這樣模型就會用該語音說出這些確切的詞句。
00:03

My Radar Apology, Mila 的聲音

Eleven v3

Mila & Sasa 使用文字轉語音製作了這句台詞,然後將其附加到至少三個鏡頭中,兩次作為 Mila 的聲音,一次作為她的口語回覆。

Katrin Ann Orbeta 執導的《Mila & Sasa》

00:04

Found My Star-Shaped Spice

Eleven v3

為 Mila & Sasa 生成的 19 句文字轉語音台詞之一,這部動畫短片中兩位主角都使用最初以文字轉語音製作的聲音說話。

Katrin Ann Orbeta 執導的《Mila & Sasa》

00:03
Young woman in an ivory lace dress at a carriage window on a rainy street, speaking the line00:05
先用文字轉語音製作台詞,再製作說出該台詞的鏡頭Jaime Yao 執導的《Voice consistency》 · Eleven v3 · Seedance 2.0 Mini · 參考轉影片

鏡頭的提示詞

Use @Using the provided character reference: the young woman with… as the start frame. Locked camera. The camera stays fixed, with only natural small bumps and sway from the moving wagon. The girl leans slightly forward, lightly rests her hand on the window edge, keeps looking out the window, and speaks the content of [relief] We are finally here.. No background music.
Jaime Yao 執導的《Voice consistency》
含義:年輕女子在搖晃馬車上固定鏡頭,她身體微前傾手扶窗邊望向窗外說我們終於到了。

如何讓角色的聲音在每個鏡頭中保持一致?

將角色的聲音片段作為音訊參考附加,並在每句台詞旁用 @ 名稱引用它,就像這 8 部影片中 15 個說話鏡頭中的 9 個所做的那樣。
Extreme close-up of a shocked young woman with dark hair in a hospital room00:07

主角的一個聲音片段

台詞命名了其聲音片段 @Maya's Warning Label,兩句之間的 (beat) 讓 Vivi 在第二個問題前有一個停頓。

In the hospital room @Frame at 6.1s, Vivi @Vivi Lovell Character Sheet stands in front of the door. Extreme close up on Vivi's face, with the door out of focus behind her. The color drains from her face, not literally, just figuratively.
VIVI in voice @Maya's Warning Label : "You... You killed someone?"
(beat)
"To give me one more day?"
M.E. Owens 執導的《One More Day》
含義:醫院房間內薇薇臉部特寫,她臉色驟變問你殺了人只為給我多一天生命嗎。
Winged figure in a dark cloak faces a young woman on a rural street at dusk00:12

兩個角色,兩個聲音片段

Vivi 和 Death 各有自己的聲音片段,在每句台詞上再次命名,這樣 Seedance 就知道四句台詞各由哪個聲音說出。

Still standing outside the rural street @Rural Street at Dusk. Medium two shot. Death @Death Character Sheet, 6'5" tall, handsome, wings at his back, is on the left of the camera, Vivi @Vivi Lovell Character Sheet, 5'3", stands on the right of the camera. They are standing in front of each other and facing each other.
VIVI is looking up at Death as she speaks: "You don't know?" in the voice of @Maya's Warning Label.
DEATH replies: "I've never went against it." in the voice of @Flova Crop Between 00 11 000 00 15 000 202607311259 8394c4 (1).
The hopelessness disappears from VIVI's expression. A grin slowly spreads across her face. Death notices immediately and
narrows his eyes.
VIVI: "Never?" in voice @Maya's Warning Label.
DEATH: "Never." in voice @Flova Crop Between 00 11 000 00 15 000 202607311259 8394c4 (1).
M.E. Owens 執導的《One More Day》
含義:鄉村街道上死神和薇薇面對面站立,她抬頭問你從未違抗過嗎,臉上露出笑容。
Frightened young woman in a medieval headdress running through a dark forest at night00:04

來自一個片段的聲音和口音

當參考提供了聲音時,35 個字的提示詞就足夠了:它從一段片段中擷取 Lady Caerwyn 的聲音和口音,然後讓她說兩個字。

Reference Lady Caerwyn's appearance from @Lady Caerwyn, After Flight @Lady Caerwyn Full-Body Three-View. She's in @ACT0 shot1.3 running away from something. Reference her voice and accent from @Rosemary and Pansies. she says : Please, please.
Jaime Yao 執導的《In ’twixt Vows》
含義:凱爾溫夫人從某處逃跑,不斷懇求說請求請求。
Two men in medieval clothes walking a foggy woodland track, seen handheld00:10

兩個參考聲音,逐行指導

每個男人的聲音和口音都來自他自己的片段,然後每一行都獲得書面指導:之前的呼吸、中間的停頓、結尾的降調。

Aurel's appearance match @Untitled Image @Untitled Image 100%.
Reference Aurel's voice and accent from @Marsh Warning (1).

Aldred's appearance match @Untitled Image @Untitled Image 100%.
Reference Aldred's voice and accent from @Boy In The Fen (1).

Camera: one continuous handheld shot. Start frame matches @Aurel and Aldred on the Track 100%. The camera begins by following Aurel and Aldred slowly toward frame right from a short distance away. The follow is imperfect and human, with clear handheld presence throughout: visible footstep bob, irregular body sway, small balance corrections, and slight reframing as the operator walks through wet uneven ground. The camera movement is steady in intention but physically unstable from walking. As Aurel and Aldred continue forward, the camera gradually falls behind them, slows down, and eventually comes to a stop. After stopping, the camera keeps the same viewing direction while the men continue walking away toward frame right and gradually leave the frame. The large foreground tree remains central in the composition, and the horse remains in the left half of the frame area as it follows behind them. No cuts, zooms, push-ins, or pull-outs.

Keep Aurel, Aldred, the horse, their clothing, equipment, scale, environment, lighting, and starting positions 100% matched to @Aurel and Aldred on the Track.

Aurel and Aldred continue walking toward frame right at a steady, unhurried pace. Keep their relative spacing consistent.

The horse follows behind them naturally rather than staying locked to them. It lags slightly, lowers its head once to casually mouth / nibble at grass near the path, then continues walking and follows after them again. Its movement remains calm and ordinary.

Aldred speaks first:

“Theo-dred is gone.”

Aldred takes a small breath before speaking. His voice is low and slightly worn, at a natural conversational volume. He says “Theo-dred” carefully, then lets a very short pause fall before “is gone.” The final two words are quieter and slightly slower, ending with a downward, subdued cadence.

After Aldred finishes, Aurel waits for a brief half-beat before replying:

“The young hunter.”

Aurel’s voice is calm, low, and even. He speaks the line slowly enough to sound like recognition rather than an immediate response. Slight weight on “young.” The final word “hunter” falls gently in pitch and volume.

After Aurel finishes, Aldred answers:

“Aye, Freya’s betrothed.”

“Aye” comes after a small exhale, low and firm. He continues into “Freya’s betrothed” without rushing, with a little more weight on “Freya’s.”

A short natural pause.

“The vows are this week.”

His voice loses some firmness on the second sentence. He speaks slightly slower, with a small breath audible before “this week.” The final words are quieter, with the pitch falling naturally at the end.

Quiet, matter-of-fact delivery with natural pacing.

As Aldred’s final line continues, the camera has already slowed and then stopped. Aurel and Aldred keep walking toward frame right and gradually leave the frame, while the shot remains behind with the tree, the fog, and the horse still present in the environment.

Audio: wet footsteps through mud and fallen leaves, noticeable cloth and leather movement from walking, subtle horse breathing, tack movement, and soft hoof sounds. The horse may briefly rustle vegetation as it lowers its head. Light wind in the damp woodland. Dialogue remains clear in the foreground. No background music.
Jaime Yao 執導的《In ’twixt Vows》
含義:手持鏡頭跟隨兩名中世紀男子和馬穿過霧中林地,男子討論逝去的獵人和即將到來的婚禮,鏡頭停下他們走出畫面,寫實陰鬱風格。
Woman in a dark wine dress at a dinner table, looking across with quiet disapproval00:15

參考聲音中的一個字台詞

每位父母的外觀和聲音都來自不同的參考,而媽媽的單字「Don.」透過「like in @Mom Voice」與她的片段連結。

Use @Mom Character Sheet Final as the visual reference for Mom.
Use @Dad Character Sheet Final as the visual reference for Dad.
use @Mom Voice as reference for what mom's voice sounds like. use @Dad's Voice as reference for what Dad's voice sounds like. use @Dad's phone as the visual reference for the phone. use @S1 CU MOM Wonder 3 for shot 1
use @S1F for shot 2
Use @S1E for shot 3
Three shots with hard cuts.
Action: Shot 1 — close-up on Mom. She subtly shakes her head, a controlled expression of quiet disapproval, then turns her gaze to frame right with a very hard expectant look. Hard cut to Shot 2 — over-the-shoulder on Mom, matching her turned position from Shot 1. Dad is in the foreground, slightly out of focus, busy working his fork through his salad, lifting a bite and eating, gaze down on his plate, completely elsewhere. Mom clears her throat deliberately. She says in her American accented voice like in @Mom Voice : "Don." Hard cut to Shot 3 — over-the-shoulder on Dad as he takes another bite of salad and chews. He looks up slowly and makes eye contact with Mom across the table. Blank expression as he chews — nothing registering. He's not sure what she's talking about. A beat. Cut back to shot 2: mom gives dad a hard expectant look, commands him with her eyes. CUT BACK TO shot 3: Something shifts behind dad's eyes. He gets it. He says, "oh, right." He quickly swallows, sets his fork and knife down on the plate, takes his phone out of his pocket and says to it: "Call Becka Babysitter." He sets the phone face-up on the table. We can see a dialing user interface pop up on the phone. The dial tone begins ringing through the phone speaker. He looks up toward frame right with an expectant expression, waiting.
Camera: static entire time. Style: David Fincher — clinical, precise, methodical, voyeuristic. 35mm film emulation, halation, bloom, shallow depth of field with hexagonal bokeh. Audio: no music, no score. the ambient clink of cutlery on plates, quiet dining room tone. Dad's fork scraping the salad bowl, the sound of chewing, Mom's deliberate throat clear, her measured delivery of "Don." Dad's chewing close on the mic, the beat of silence as recognition dawns, his swallow, the soft clatter of utensils being set down, his casual voice command to the phone, the phone being placed on the table, the dial tone ringing through the small speaker, the ambient dining room silence around it.
Forbidden: no white blobs on the character's faces.
Gertywood 執導的《Xenogenesis》
含義:媽媽在餐桌上用眼神命令爸爸打電話給保姆,採用芬奇式精準冷調風格與淺景深拍攝。
Hand-drawn 2D girl and a small blue companion share a star-topped tart on a spaceship roof00:15

相同的聲音片段用於新台詞

My Radar Apology 片段再次作為 Mila 的聲音用於三行新台詞,因此她聽起來像自己在說片段中從未說過的話。

Start and maintain background @Mila in Spaceship Interior 100%.
Mila's appearance @Mila con Zaino Giapponese, Senza Ombre 100% and expressions @Mila - Espressioni del Viso.
Mila's voice @My Radar Apology.
Sasa's appearance @Sasa con Zaino Giallo e Pistola Laser 100% and expressions @Sasa - Espressioni Facciali 100%. Remove guns from Sasa. Remove any glow form the characters.
Sasa's voice @Mother's Old Spaceship. 2D hand-drawn animation.
Sasa is smaller than Mila. Tart looks like this @Mila con Crostata. Do not take this image background. The tart is 90% smaller. No glow on their heads.
Preserve reference style and character identity. CAMERA: 2 shots, 1 hard cut. SHOT 1 — Medium close-up, 3/4.Mila and Sasa sit side by side on the top of the spaceship. Mila in front then Sasa beside her. Mila holds a tart sliced into 6 pieces. In the center the tart the star spice in the center.
Sasa, happy for Mila with deep warm paternal male voice: “What were the odds we'd find your mother's old spaceship?” Mila turns to him, calm, visibly happy, smiles softly: “I’ve found my star-shaped spice.” Close up - Mila takes a slice with the star spice on top of the slice and gives a bite to it and closes her eyes. The tart has 5 slices left. She closes her eyes tasting the tart. Close up of Mila with closed eyes and and tasting the tart, she opens her eyes and looks up the sky. Her expression becomes quietly hopeful: “But I have to find her among the stars.” AUDIO: Natural dialogue, no overlap. Sasa's voice is deep, warm and human, never robotic. Mila has natural voice, never robotic. No music. No sound. Hand-drawn 2D animation cozy retro-futuristic sci-fi, warm pastel palette, soft sepia linework, vintage children's book, gentle golden light, whimsical, nostalgic, cozy spaceship interior, cinematic parallax, subtle atmospheric animation.
Katrin Ann Orbeta 執導的《Mila & Sasa》
含義:米拉和薩薩坐在太空船上分享星形香料餡餅,她決心在星空中尋找母親,手繪溫馨復古科幻風格。

如何在提示詞中描述角色的聲音?

沒有聲音片段時,用幾個字在台詞旁邊描述聲音的年齡、質地、口音和情緒,因為模型會單獨從該描述建構聲音。
Grey-haired man in a dark uniform in a static close-up against a grey wall00:05

威嚴沙啞聲音的試鏡

試鏡,一個靜態特寫搭配環境音且沒有音樂,將聲音隔離出來,這樣你可以在角色有場景之前判斷「威嚴沙啞」的效果。

use @VP as the man. Action: close up on the man in a grey room. He says with a commanding gruff voice, "Hi my name is V.P. and this is my voice."
Audio: no music. room tone.
Style: screen test. Camera static.
Gertywood 執導的《Xenogenesis》
含義:一名男子在灰色房間特寫中以粗啞嗓音自我介紹,靜態鏡頭試鏡風格。
Stern woman in a dark wine dress facing the camera against a white studio backdrop00:04

嚴厲企業聲音的試鏡

同一部影片用一行台詞和「a stern American corporate voice」測試了媽媽的聲音,後續的提示詞引用了名為 @Mom Voice 的參考。

use @Mom Character Sheet Final as the woman. forbidden: white blobs over the character's face.
medium shot: The woman is in a studio with a white backdrop, looking directly at camera. She says, with a stern American corporate voice, "hi my name is mom and this is my voice."
Style: Hollywood screen test.
Camera: Locked off static.
Audio: No music. No sound effects. Woman's voice is captured naturally as if in a room with the appropriate reverb.
Gertywood 執導的《Xenogenesis》
含義:一位女性在白色背景攝影棚中對鏡頭說話介紹自己,好萊塢試鏡風格,靜態鏡頭。
Older man in uniform speaking from an old CRT screen in a dark room00:15

倒數計時的聲音鎖定

聲音鎖定在一段文字中設定年齡、音調、口音和麥克風,而「Speak exactly these words and no others」讓倒數計時不會有即興發揮。

Create an exactly 15-second, 24 fps, 2.39:1 cinematic video with synchronized dialogue audio..

Voice lock: identical older male low natural baritone, neutral American state-broadcast English with faint old Mid-Atlantic precision, clipped consonants, slight dry rasp. Controlled urgency builds with each number, but he never turns into a sports announcer and never shouts. Each number is fully intelligible and separated by silence. Close mono microphone, light analog saturation, faint CRT hum, no music, ticking, crowd, gate, or action sound.

HE SAYS COUNTING IT DOWN IN AN EPIC WAY “Five…” “four…” “three…” “Two…” “One.” The first three numbers flow as one grave countdown phrase; the longer gaps before “Two” and “One” make the pressure unbearable. Let urgency enter the breath but keep the volume controlled.

Speak exactly these words and no others:

“Five… four… three… Two… One.”

Do not add, omit, repeat, paraphrase, or caption any word. No subtitles, numerals, clock graphic, title, logo, watermark, other voice, cutaway, sound effect, or music.
Sam Wasserman 執導的《The Miracle》
含義:男性旁白用低沉嗓音倒數五到一,聲音逐漸緊迫,電影寬銀幕格式,無音樂無視覺元素。
Two robbers in cowboy hats aim revolvers in a rooftop restaurant at night
第一幀:Two robbers in cowboy hats aim revolvers in a rooftop restaurant at night
第一幀
00:12

兩種不同脾氣的德州腔

兩名搶匪都帶著德州口音說話,男性冷靜而冷酷,女性則興高采烈,因此兩個聲音不僅因說話者不同而有所差異,態度上也截然不同。

Use the provided close shot as the direct first frame and the provided wide shot as the direct final frame. Create a realistic live-action cinematic shot that performs a smooth continuous dolly back from the first frame to the last frame. Start relatively close on the two gangsters, framed roughly from above the chest upward, then gradually pull back until the shot fully matches the provided wide final frame and reveals the whole restaurant, the hostages, and the right-bottom male victim. Keep the same luxury rooftop restaurant at night, floor-to-ceiling window, city lights outside, warm practical lighting, and premium ARRI Alexa 35 film realism. The male gangster carries a revolver and delivers his line in a calm, cold, threatening Texas drawl: “Everybody cool, this is a robbery.” The female gangster carries a pistol and is more manic, gleeful, and excited by the fear around her; after a man screams, she says in a thick Texas drawl, with deranged enjoyment, “Or everybody dies!” During the pullback, diners around them panic and scream. Near the end of the dolly back, the man in the bottom-right corner of the final wide frame cries out, “Oh my God!” and the female gangster immediately turns and shoots him. After the shot, more people scream. Keep the male gangster emotionally controlled and dangerous; keep the female gangster wild, delighted, and intoxicated by the terror she is causing. No music, no score, only dialogue, screams, gunshot, and natural room tone. Preserve photorealism, true anatomy, realistic gun handling, believable performance, and a continuous smooth camera move from the close first frame to the wide final frame. 无音乐
Zhelun “Jerry” Lin 執導的《The Perfect Proposal》
含義:兩名德州歹徒在高級餐廳搶劫,鏡頭從近景拉遠至全景,女歹徒開槍射擊人質,寫實電影風格。
Hand-drawn woman in a gold headwrap in a yard in front of a wooden house00:10

一聲驚恐的葡萄牙語呼叫

設定「音訊語言:葡萄牙語」並用葡萄牙語引述台詞,就能得到該語言的驚慌呼叫,恐懼感被寫入了演繹之中。

Style: Hand-drawn 2D animation, richly painted gouache and watercolor texture, cel-shaded character over a highly detailed painted background, confident linework with painterly edge variation — not vector-clean, not stiff — natural daylight lighting, warm and soft, animated on twos at 12 frames per second, every frame independently hand-redrawn with visible line jitter and boil, no 3D render, no CGI smoothness, no motion blur, no literal camera lens vocabulary.

MÃE DE DÚRÓ as character reference — (@ImageMae) exact 1-to-1 face, white embroidered blouse, olive-green wrapped skirt with woven sash, gold headwrap, gold necklace pendant, gold earrings and bracelets, warm brown skin tone, matching the reference sheet precisely. CASA DE DÚRÓ as background environment reference — (@ImageCasa) exact 1-to-1 layout, yard, and surrounding trees.

Off-center framing mandatory. Every frame must show continuous, clearly visible motion throughout.

Action Sequence: Medium shot, off-center, the mother standing in the yard behind the house, her posture tense and alert. Her head turns quickly, scanning the tree line at the edge of the clearing, worry etched clearly across her face. She cups one hand near her mouth and calls out, urgent and afraid, her voice carrying real panic: "DÚRÓ, TÁ INDO PARA ONDE?!" Her body leans slightly forward as she shouts, the fabric of her skirt and headwrap shifting with the sudden movement. Warm natural daylight falls across the yard, leaves rustling gently in the background breeze, grounding the scene firmly in the real world.

CONSTRAINTS: MÃE matches (@ImageMae) exactly 1-to-1; environment matches (@ImageCasa) exactly; off-center framing, no centered symmetry, no direct head-on pose; her expression and body language read as urgent, frightened, searching — not calm; natural daylight palette only, no mystical or aged-gold tones; avoid jitter, avoid bent limbs, avoid identity drift, avoid chaotic composition; no text, no watermark. Audio language: Portuguese. Dialogue: mother shouts "DÚRÓ, TÁ INDO PARA ONDE?!" with visible worry and urgency in tone and delivery. SFX: rustling leaves, distant birds falling silent, tense ambient hush.
duration: 10s (single continuous shot — force explicit duration, do not leave automatic)
Walter Mauro 執導的《Duro》
含義:手繪2D動畫,母親在屋後院子裡緊張地轉頭掃視樹林,驚慌呼喊尋找孩子,水粉畫質感和自然光線。
Young woman in an ivory lace dress at a carriage window on a rainy street
第一幀:Young woman in an ivory lace dress at a carriage window on a rainy street
第一幀
00:05

馬車台詞,採用輸入而非錄音

語音一致性也讓馬車鏡頭僅從輸入的台詞「啊,我們終於到了」生成,與使用文字轉語音台詞的版本並列。

Locked camera. The camera stays fixed, with only natural small bumps and sway from the moving wagon. The girl leans slightly forward, lightly rests her hand on the window edge, keeps looking out the window, and cheers "ah, we're finally here." No background music.
Jaime Yao 執導的《Voice consistency》
含義:固定鏡頭拍攝女孩在行駛馬車上倚窗前傾,欣喜說終於到了,畫面隨車身自然晃動。

什麼讓 AI 角色的聲音聽起來自然?

這 15 個鏡頭背後的提示詞將每個聲音綁定到一個來源,描述每句台詞的說話方式,並確保配樂中不出現其他內容。
  1. 1. 為每個聲音提供專屬參考

    在台詞本身標註聲音來自哪個片段,並說明是否也包含口音,這樣模型就不會將一個角色的聲音套用到另一個角色上。

    Reference her voice and accent from @Rosemary and Pansies.
    Jaime Yao 執導的《In ’twixt Vows》
    含義:參考特定角色的聲音和口音。
  2. 2. 同時用文字描述聲音

    即使附加了片段,也要加上簡短描述,因為用文字描述的年齡、溫暖感和節奏,能在台詞偏離參考時提供支援。

    Deep, warm, paternal adult male voice, calm and natural.
    Katrin Ann Orbeta 執導的《Mila & Sasa》
    含義:深沉溫暖的父親般成年男性聲音,平靜自然。
  3. 3. 逐句指導演繹方式

    寫下每句台詞的說話方式,包括演員會從導演那裡得到的呼吸、音量和音調指示,因為演繹方式才是讓聲音聽起來像真人的關鍵。

    His voice is low and slightly worn, at a natural conversational volume.
    Jaime Yao 執導的《In ’twixt Vows》
    含義:他的聲音低沉略帶滄桑,音量為自然對話程度。
  4. 4. 鎖定用詞

    當確切台詞很重要時,告訴模型只說引述的詞句,不要添加其他內容,因為視訊模型可能會增加、重複或改述用詞。

    Speak exactly these words and no others
    Sam Wasserman 執導的《The Miracle》
    含義:完全照字面說出這些話,不多不少。
  5. 5. 讓說話者逐一發言

    在一個有兩個角色的場景中,排除重疊的台詞並指定誰在動嘴,這樣正確的臉部才能搭配正確的聲音。

    DIALOGUE: Natural pauses. No overlapping dialogue. Only the speaker moves their lips.
    Katrin Ann Orbeta 執導的《Mila & Sasa》
    含義:對話有自然停頓,不重疊,只有說話者嘴唇動。
  6. 6. 清除原聲帶

    排除音樂和雜散音效,就像 15 個說話鏡頭中的 9 個所做的那樣,這樣聲音才能保持足夠乾淨以便評估,並與您自己的配樂剪輯搭配。

    Audio: No music. No sound effects. Woman's voice is captured naturally as if in a room with the appropriate reverb.
    Gertywood 執導的《Xenogenesis》
    含義:無音樂無音效,女性聲音以室內自然混響錄製。

如何在 Flick 中為角色賦予聲音?

設計角色,使用文字轉語音生成一句台詞,在 Omni Reference 鏡頭中引用該台詞,並將其附加到該角色的每個後續鏡頭中。
  1. 第 1 步

    設計角色

    生成包含正面、側面和背面視圖的角色設定圖,這樣每個後續承載聲音的鏡頭也能匹配所屬的臉部。

    Character design sheet of Sasa, a small blue hooded 2D character, from the front, side and back
    Katrin Ann Orbeta 執導的《Mila & Sasa》 · Nano Banana 2
  2. 第 2 步

    生成台詞

    使用 Eleven v3 的文字轉語音寫一句台詞,加入音訊標籤例如 [relief] 來表達情緒,並保留符合角色的版本。

    00:05
    Katrin Ann Orbeta 執導的《Mila & Sasa》 · Eleven v3
  3. 第 3 步

    在鏡頭中引用

    開啟 Omni Reference,附加設定圖和語音台詞,並在提示詞中命名該台詞,例如「Sasa 的聲音 @Company Rule Number One」,這樣新的文字就會以該聲音呈現。

    Hand-drawn 2D girl tasting soup in a spaceship kitchen while a small blue companion watches00:15
    Katrin Ann Orbeta 執導的《Mila & Sasa》 · MiniMax H3 · 參考轉影片
  4. 第 4 步

    每個鏡頭都重複使用

    將相同的片段附加到該角色的每個後續鏡頭中,就像 Mila & Sasa 對 Company Rule Number One 所做的那樣,一次作為 Sasa 的聲音,一次作為他的口白台詞。

    Small blue 2D character with a laser gun stands over a frightened girl on a spaceship floor00:15
    Katrin Ann Orbeta 執導的《Mila & Sasa》 · MiniMax H3 · 參考轉影片

人們對 AI 角色聲音有哪些疑問?

什麼是 AI 角色語音生成器?

AI 角色語音生成器可將書面台詞轉換為屬於某個角色的語音。在 Flick 中,文字轉語音可製作台詞,而影片模型可將該台詞作為角色在每個鏡頭中的聲音。

有免費的 AI 角色語音生成器嗎?

有。使用 Eleven v3 或 Eleven v4 的文字轉語音在 Flick 的 Free 方案中每次生成需 8 點數,註冊時可獲得 100 點數,加入 Flick Discord 還可再獲得 200 點數。使用 Seedance 的說話鏡頭需要 Standard 方案(每月 $20),而 MiniMax H3 Max 在 Free 方案中就能將語音片段作為參考。

如何在不同場景中保持角色語音的一致性?

為角色製作一個語音台詞,將其作為音訊參考附加到他們說話的每個鏡頭,並在每句台詞旁以 @ 名稱引用它。同時用幾個詞描述聲音,讓模型同時擁有聲音和文字來匹配。

AI 可以從音訊片段複製聲音嗎?

可以。在 Flick 中,使用 Seed Audio 1.0 的文字轉語音可接受最多三個語音片段,並以該聲音生成新的語音,而 Omni Reference 中的影片模型可將語音片段作為參考。複製聲音時請務必取得說話者的許可。

如何讓 AI 角色的嘴唇與語音同步?

同時生成語音和畫面:Seedance 可從語音片段或描述的聲音製作與影片同步的對嘴語音。對於靜態肖像和錄製的台詞,Standard 方案中使用 Kling AI Avatar v2 的角色對話可將臉部動畫與音訊同步。

AI 角色語音可以表現情感嗎?

可以。Eleven v3 可讀取方括號中的音訊標籤,例如 [relief],來設定台詞的說話方式,而影片提示詞可以描述每句台詞的呼吸、停頓和音調。

AI 角色可以說其他語言嗎?

可以。Eleven v4 支援 90 多種語言,Seedance 提示詞可以引用其他語言的台詞並指定音訊語言,就像某部駐村影片中葡萄牙語台詞的做法一樣。

更多關於在 Flick 中製作 AI 角色的資訊

本頁面的影片

為每個角色賦予聲音

錄製一次台詞,讓每個鏡頭都能使用。您的台詞表、聲音和錄音都在同一個畫布上。

在 Flick 中為您的角色賦予聲音

免費試用,無需信用卡。

讓 AI 解讀本頁:ChatGPTPerplexityClaude