Alibaba 多鏡頭影片模型,連對白一起生成
一句提示詞,導出完整場景。多個鏡頭、角色對白、完整音軌,全收進一支 1080p 影片。
WAN 2.6 是 Alibaba 推出的影片模型,支援文字生成影片、圖片生成影片與原生音效。把對白放進引號,角色就會開口說。為鏡頭編號,模型就會自動切換畫面。像本頁的蒸汽火車,從高架橋空拍、列車長看錶,到車輪飛馳,15 秒一氣呵成。本頁所有範例都在 Pixel Dojo 製作,每支影片都附上實際使用的提示詞。在 Pixel Dojo,一個方案就能用所有頂尖模型、六大工作室,以及 AI 導演 Sensei。訂閱期間點數可累積,每個方案都包含商用權。
10 credits ($0.63) 所有方案皆包含,每月 US$10 起。
已經是會員? 登入
作者 Brian Lovett · 更新於
我們用 WAN 2.6 Standard 製作了七支影片,其中四支為 1080p。每支都有自己的音軌,開啟影片聲音就能聽見,再點「使用此提示詞」,直接以它為起點創作。
1080p 三鏡頭場景,15 秒。一段提示詞串起空拍、對白與車輪特寫,列車長伴著引擎的轟隆聲說出「Right on time」。
Cinematic multi shot sequence. Shot 1: Aerial wide shot, a black steam locomotive with a glowing headlamp crosses a curved stone viaduct over a snowy pine valley at dusk, white steam trailing behind. Shot 2: Medium close-up inside a warm wood paneled dining car, a conductor in his sixties with a gray mustache and a navy cap checks a brass pocket watch and says, "Right on time." Shot 3: Low close-up of the iron wheels and pistons pounding through falling snow, sparks flying. 35mm lens, amber and teal grade. Sound: chugging engine, a long whistle, clinking glasses, wind.
1080p 對白,5 秒。她挑起眉,冷冷地說出「Coffee's fresh. The pie, not so much」,背景是雨聲與冰箱的低鳴。
Medium close-up, slow push in. A tired diner waitress in her fifties with silver streaked red curls pinned up and a mint green uniform leans on the counter late at night, coffee pot in hand, a pink neon sign glowing through the rain streaked window behind her. She raises an eyebrow and says with a wry smile, "Coffee's fresh. The pie, not so much." Warm tungsten practicals against cold blue street light, 50mm lens, soft film grain. Sound: rain on glass, a humming refrigerator, her low raspy voice.
1080p 商品鏡頭,5 秒。鏡頭緩緩環繞,皇冠狀水花綻開,金色光線掃過多切面玻璃。
Macro product shot, slow orbit. An invented faceted amber glass perfume bottle with a brass cap stands on wet black slate as a single drop falls from above and bursts into a crown of water around its base, a slow sweep of golden light gliding across the glass. Pure black background, 100mm macro lens, crisp reflections, no text. Sound: a deep drip, a soft shimmer, a warm synth pad.
1080p 舞蹈,9:16,5 秒。金色彩帶在晨霧中劃出弧線,直式構圖專為 Reels 和 Shorts 打造。
Vertical full body shot, slow handheld orbit. A dancer in her thirties with a sleek black bun and a flowing white linen outfit spins on a misty lakeside dock at sunrise, swirling two long saffron silk ribbons into wide spirals that ripple through the air, mist rising off the water behind her. Soft golden backlight, 35mm lens. Sound: the swish of silk, bare feet on wood, a slow solo erhu melody.
720p 動作場景,5 秒。泥漿、水花、一路甩過髮夾彎,引擎咆哮同步入軌。
Low tracking shot at bumper height, the camera racing alongside. A mud splattered white rally car with no logos slides sideways through a hairpin bend on a rain soaked forest track, the rear tires flinging a thick rooster tail of mud and water toward the lens, headlights blazing through drifting mist between tall pines. 24mm lens, motion blur, desaturated green grade. Sound: engine scream, gravel and mud spray, rain.
720p 空拍,5 秒。低空高速逼近海蝕柱,大西洋浪潮猛烈拍打岩壁。
Aerial drone shot flying low and fast over a black sand beach toward a towering basalt sea stack as huge Atlantic waves explode white against the rocks, seabirds wheeling through the spray, low silver sunrise light breaking through storm clouds. 24mm wide lens, epic scale, crisp detail. Sound: thundering surf, howling wind, distant gull cries.
720p 極限運動,5 秒。機車從沙丘頂端飛躍而出,落入揚起的金色沙塵。
Low tracking shot from the sand, the camera racing alongside. A rider in a white and orange helmet and a sand colored jersey launches a red dirt bike off the crest of a tall desert dune, hangs in the air against a deep blue sky, and lands in a spray of golden sand that blasts toward the lens, low afternoon sun raking across the ripples. 24mm lens, motion blur, warm high contrast grade. Sound: a screaming two stroke engine, the thump of the landing, sand hissing.
多數 AI 影片模型一次只給你一個鏡頭,故事得靠你事後剪接。WAN 2.6 直接處理整個場景。在提示詞裡寫下「Shot 1」、「Shot 2」、「Shot 3」,它就會在同一支影片中切換鏡頭,讓角色、光線與音軌在切換前後保持一致。
我們在蒸汽火車提示詞中指定了高架橋空拍、餐車裡的列車長,以及車輪特寫。WAN 2.6 用 1080p、15 秒完成全部三個鏡頭,列車長還在中間的鏡頭配合說話節奏,說出「Right on time」。
聲音也一起生成。我們將對白影片跑過語音轉文字檢查,每句辨識結果都與原文完全一致。引擎、雨聲、海浪與風聲,也都配合畫面生成。
單鏡頭也出色,從香水瓶周圍炸開的皇冠狀水花,到越野機車飛躍沙丘,都能精彩呈現。
我們從火車場景擷取一張列車長的靜態畫面,用同一段提示詞交給兩個版本生成影片:他啪地闔上懷錶、抬起頭,說出「Next stop, the summit.」。兩個版本都逐字說出這句對白,背景還有鐵軌的喀噠聲。720p、5 秒影片,Standard 花了 10 點數,Flash 則是 7.5 點數。
WAN 2.6 Standard,圖片生成影片,720p,5 秒,10 點數。
Slow push in. The conductor snaps the brass pocket watch shut, looks up toward the lens with a warm grin and says, "Next stop, the summit." Snow streams past the windows behind him. Sound: the rhythmic clatter of the rails, the click of the watch, his deep friendly voice.
WAN 2.6 Flash,圖片生成影片,720p,5 秒,7.5 點數。
Slow push in. The conductor snaps the brass pocket watch shut, looks up toward the lens with a warm grin and says, "Next stop, the summit." Snow streams past the windows behind him. Sound: the rhythmic clatter of the rails, the click of the watch, his deep friendly voice.
為鏡頭編號,把對白放進引號,就能生成剪接好的場景與搭配的聲音,就像我們的三鏡頭火車旅程。
服務生、列車長、主持人與旁白,以到位的表情與節奏說出台詞。
1080p 水花、環繞運鏡與掃光,直接用於廣告和商品頁。
拉力賽甩尾、沙丘飛躍,追蹤鏡頭緊跟高速動作。
無人機低空掠過海岸、山谷與城市,風聲與海浪聲同步入軌。
選擇 9:16 比例,製作 Reels、TikTok 和 Shorts,就像我們的彩帶舞者影片。
Pixel Dojo 每個方案都包含 WAN 2.6,每月 $10 起,訂閱期間點數可累積。按秒計費,聲音已包含在內。WAN 2.6 Standard 在 720p 下每秒 2 點數,1080p 下每秒 3 點數。WAN 2.6 Flash 在 720p 下每秒 1.5 點數,1080p 下每秒 2.25 點數。
影片長度為 5 至 15 秒。5 秒的 Standard 影片,720p 需 10 點數,1080p 需 15 點數。我們那支 15 秒、1080p 的火車場景花了 45 點數。
WAN 2.6 是 Alibaba 的 AI 影片模型,支援文字生成影片與圖片生成影片。它會隨畫面一起生成聲音,說出你放在引號裡的對白,並在同一支影片中切換多個鏡頭,最高支援 1080p、最長 15 秒。
可以。本頁每支影片都有生成音軌。把台詞放進引號,再用一行 Sound 列出音效。我們的服務生、列車長和火車場景,都逐字說出了指定台詞。
在同一段提示詞中依序寫出鏡頭:「Shot 1: Aerial wide shot...」、「Shot 2: Medium close-up...」、「Shot 3: Low close-up...」。分別指定空拍遠景、中近景與低角度特寫。選擇較長的影片秒數,讓每個鏡頭都有發揮空間。我們的 15 秒火車場景就用了三個鏡頭。
Standard 是畫質更高的版本,支援文字生成影片、圖片生成影片與 1080p,720p 下每秒 2 點數。Flash 則是速度更快、費用更低的圖片生成影片選項,720p 下每秒 1.5 點數。兩個版本都讓我們的列車長逐字說出了台詞。
可以。不論哪個方案,你都擁有自己創作的內容。
免費帳號每天可免費使用 3 次特效,體驗 Pixel Dojo。每月 $10 起的所有方案都包含 WAN 2.6,訂閱期間點數可累積。一支 5 秒、720p 的 Standard 影片需 10 點數。
可以。以 POST 將提示詞連同 mode、duration 和 resolution 傳送至 /api/v1/models/wan-2.6-standard/run,或使用 wan-2.6-flash 進行圖片生成影片。