Alibaba 多鏡頭影片模型，聲音同步生成
一句提示詞，讓 WAN 3.0 生成最長 30 秒的電影感多鏡頭場景，連聲音都一次到位。
用 Alibaba 最新影片模型 WAN 3.0，拍出舞蹈短片、產品廣告、黏土動畫、開口說話的角色與超現實視覺特效。單次生成 2 到 30 秒的有聲影片，支援 480p、720p、1080p、2K 或 4K，也能依照你提供的圖片、影片與音訊參考創作。本頁每支 WAN 3.0 影片都由我們製作，並附上實際使用的完整提示詞，讓你直接接著創作。在 Pixel Dojo，一個方案就能使用 WAN 3.0、各大頂尖影片模型、六大工作室，以及 AI 導演 Sensei。訂閱期間點數可累積保留，所有方案都含商用授權。
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
我們在 Pixel Dojo 用 WAN 3.0 製作了這六支影片：舞蹈短片、20 s 多鏡頭美食場景、香水廣告、黏土動畫短片、開口說話的角色，以及超現實視覺特效。每支都有自己的音軌。點選任一卡片上的「使用此提示詞」，就能接著創作。
1080p 舞蹈與時尚影像：深紅絲綢、環狀飛濺的水花，加上一道頂部硬光，以緩慢推進的鏡頭貫穿整段 10 s。
Symmetrical wide shot with a slow push-in, one continuous take. A dancer with a sleek black bun and a flowing crimson silk gown spins on a flooded concrete floor under a single hard white top light, her skirt fanning out and flinging a ring of water droplets that sparkle in the beam, a dark empty hall behind her. High contrast, deep blacks, 50mm lens, subtle film grain. Audio: splashing water, rustling silk, echoing footsteps, one low sustained cello note.
一句提示詞，完成 20 s、三個鏡頭的場景：雨中的攤位、火焰中炒鍋的慢動作特寫，最後由同一名廚師微笑上菜。
A 20 second cinematic sequence in three shots about a night market cook. Shot 1 [0-6s]: wide shot, a street food stall glows under red paper lanterns in light rain as a cook in a white headscarf tosses noodles in a wok and a burst of flame leaps up. Shot 2 [7-13s]: hard cut to a close-up of the wok, noodles and vegetables flipping through the flames in slow motion, sparks and steam rising. Shot 3 [14-20s]: hard cut to a medium shot, the cook slides a steaming plate across the counter toward camera and smiles. Keep the cook, the stall and the rain consistent across shots. Audio: sizzling wok, roaring flame, rain on the canvas awning, lively market chatter.
微距香水廣告：水滴在瓶身旁濺起皇冠狀水花，琥珀色玻璃依然清晰銳利，一道光芒掠過金色瓶蓋。
Macro product shot with a slow dolly in, one continuous take, the whole bottle in frame. An unlabeled faceted glass perfume bottle with a brushed gold cap, filled with amber liquid, stands on wet black slate as a single water drop falls beside it and bursts into a crown splash, while a thin band of light sweeps across the glass and makes the liquid glow. Black studio background, glossy reflections, shallow depth of field, 100mm macro lens. Audio: one crisp water drip, a soft rising synth tone.
討喜的黏土動畫：獨眼黏土機器人在長滿苔蘚的浮島上，為一朵橘色小花澆水。黏土保留指紋紋理，清澈水流在苔蘚上匯成小水窪。
Stop-motion claymation, locked-off medium shot. A small round clay robot with a teal body and one big glowing eye tips a tiny watering can over a single orange flower on a floating island of moss, and the flower opens its petals as the water lands. Fingerprints visible in the clay, soft miniature set lighting, pastel sky backdrop, charming stop-motion stutter. Audio: tiny mechanical whirs, a gentle water trickle, a cheerful plucked ukulele melody.
說話角色特寫：他從黃銅燈前抬頭，說出台詞。溫暖燈光映在臉上，身後玻璃上流著雨水。
Medium close-up with a slow push-in, one continuous take. A weathered lighthouse keeper in his sixties with a white beard and a navy wool cap looks up from polishing a brass lamp and says quietly: "Forty years, and the light never went out." Warm tungsten lamplight on his face against blue dusk windows, rain streaking the glass, 50mm lens, fine film grain. Audio: his low gravelly voice, wind against the glass, a distant foghorn.
超現實視覺特效：暮色中，暴風雲化成的鯨魚閃著電光，沿途灑下雨水，游過紅岩峽谷。岩柱上的獨行登山客，襯托出牠的巨大尺度。
Epic wide aerial shot, slow drone push forward, one continuous take. A colossal whale whose entire body is made of billowing grey storm clouds, with no skin, only swirling vapor in the shape of a whale, glides over a red sandstone canyon at dusk, lightning flickering inside the cloud body and rain trailing from its belly onto the rocks, a tiny lone hiker on a ridge in the foreground for scale. Gold and violet sky, volumetric light rays, large format clarity. Audio: deep rolling thunder, a low distant whale song, desert wind.
WAN 3.0 能用一句提示詞生成多鏡頭場景，同步產生聲音，解析度涵蓋 480p 到 4K。這三支影片帶你看完整表現：快速製作的 480p 循環短片、720p 快速模式，以及完整 4K 畫面。
同一組舞蹈提示詞，以 480p 呈現：旋轉動作、絲綢質感與環狀水花都清楚可見，適合快速製作社群循環短片。
Symmetrical wide shot with a slow push-in, one continuous take. A dancer with a sleek black bun and a flowing crimson silk gown spins on a flooded concrete floor under a single hard white top light, her skirt fanning out and flinging a ring of water droplets that sparkle in the beam, a dark empty hall behind her. High contrast, deep blacks, 50mm lens, subtle film grain. Audio: splashing water, rustling silk, echoing footsteps, one low sustained cello note.
720p 快速模式：主廚倒入高湯、擺上叉燒，接著在碗上方剝開溏心蛋。
Overhead close-up, then a slow tilt down to eye level, one continuous take. A chef in a black apron ladles steaming golden broth over noodles in a dark ceramic bowl, then lays down chashu slices and a soft egg split open with a glossy orange yolk, steam curling through a warm spotlight in a dim wooden noodle bar. 85mm lens, rich warm grade, shallow depth of field. Audio: broth pouring, the clink of a ladle, low bar chatter.
4K 暴風鯨：峽谷細節清晰銳利，雲層內閃電交錯，下方垂落片片雨幕。
Epic wide aerial shot, slow drone push forward, one continuous take. A colossal whale whose entire body is made of billowing grey storm clouds, with no skin, only swirling vapor in the shape of a whale, glides over a red sandstone canyon at dusk, lightning flickering inside the cloud body and rain trailing from its belly onto the rocks, a tiny lone hiker on a ridge in the foreground for scale. Gold and violet sky, volumetric light rays, large format clarity. Audio: deep rolling thunder, a low distant whale song, desert wind.
我們把同一組拉麵提示詞交給 WAN 3.0 與 WAN 2.7，都設定為 5 s、720p。WAN 3.0 完整演出我們寫下的步驟：倒入高湯、用手擺上叉燒，再放下切半的蛋。WAN 2.7 則呈現暖調的成品主視覺，搭配緩緩倒入的高湯。連續寫下多個動作，WAN 3.0 就會逐一完成。
相同提示詞： Overhead close-up, then a slow tilt down to eye level, one continuous take. A chef in a black apron ladles steaming golden broth over noodles in a dark ceramic bowl, then lays down chashu slices and a soft egg split open with a glossy orange yolk, steam curling through a warm spotlight in a dim wooden noodle bar. 85mm lens, rich warm grade, shallow depth of field. Audio: broth pouring, the clink of a ladle, low bar chatter.
WAN 3.0：倒湯、放叉燒、再放切半的蛋，全都親手依序完成。
Overhead close-up, then a slow tilt down to eye level, one continuous take. A chef in a black apron ladles steaming golden broth over noodles in a dark ceramic bowl, then lays down chashu slices and a soft egg split open with a glossy orange yolk, steam curling through a warm spotlight in a dim wooden noodle bar. 85mm lens, rich warm grade, shallow depth of field. Audio: broth pouring, the clink of a ladle, low bar chatter.
WAN 2.7：完成的拉麵，搭配緩緩倒入的高湯。
Overhead close-up, then a slow tilt down to eye level, one continuous take. A chef in a black apron ladles steaming golden broth over noodles in a dark ceramic bowl, then lays down chashu slices and a soft egg split open with a glossy orange yolk, steam curling through a warm spotlight in a dim wooden noodle bar. 85mm lens, rich warm grade, shallow depth of field. Audio: broth pouring, the clink of a ladle, low bar chatter.
| WAN 3.0 | WAN 2.7 | |
|---|---|---|
| 提示詞中的動作 | 倒湯、放叉燒、放蛋，依序完成 | 在完成的拉麵上倒入高湯 |
| 720p、5 s 所需點數 | 10 | 10 |
| 影片最長長度 | 30 s | 15 s |
| 最高解析度 | 4K | 1080p |
| 聲音 | 與畫面同步生成 | 與畫面同步生成 |
Pixel Dojo 所有方案都能使用 WAN 3.0，每月 $10 起，訂閱期間點數可累積保留。影片依秒數消耗點數，生成前，工具會依照你的設定顯示確切費用。
每秒費率為：480p 需 1.5 點數、720p 需 2 點數、1080p 或 2K 需 5 點數、4K 需 6 點數；720p 快速模式則為每秒 3 點數。所有費率都包含聲音，也包含參考圖片與參考音訊；參考影片則依秒數，按輸出影片的費率計費。
一支 720p、5 s 的影片需 10 點數。
先寫景別與運鏡，再描述一個主體、一個動作、光線、鏡頭焦段，以及想要的聲音。想要一鏡到底，就加上「一個連續鏡頭」。
多鏡頭段落要為每個鏡頭命名、標上時間區間，並寫清楚哪些元素必須跨鏡頭保持一致，就像我們的夜市提示詞。點選該卡片上的「使用此提示詞」，就能接著創作。
選擇 2 到 30 秒、480p 到 4K；想加快出片，就開啟快速模式。加入參考圖片、影片或音訊，固定視覺風格、運鏡或聲音，接著生成。
WAN 3.0 是 Alibaba 的 AI 影片模型。輸入文字提示詞，或提供自己的圖片、影片與音訊，就能生成 2 到 30 秒的有聲影片，解析度從 480p 到 4K。你可以在瀏覽器中透過 Pixel Dojo 使用，也能透過 Pixel Dojo API 串接。
用一句提示詞，生成有聲音、多鏡頭的電影感場景。我們的 20 s 夜市場景，會照指定順序演出三個鏡頭，全程維持同一名廚師。本頁範例也涵蓋舞蹈、香水廣告、黏土動畫、開口說話的角色與超現實視覺特效。
單次可生成 2 到 30 秒，也能讓模型自行決定長度。寫下帶有時間區間的分鏡清單，較長的影片就能容納多個鏡頭。
480p、720p、1080p、2K 與 4K。社群短片可選 480p 或 720p，主視覺鏡頭用 1080p，想呈現大螢幕細節就選 2K 或 4K，就像上方的暴風鯨。
可以。聲音與畫面同步生成，每支影片都有自己的音軌；在提示詞中寫入台詞，也會生成對白。費用已包含聲音。
可以。所有方案都一樣，你創作的內容就屬於你。
免費帳號每天可獲得 3 次免費特效體驗，試用 Pixel Dojo。每月 $10 起的所有方案都能使用 WAN 3.0，訂閱期間點數可累積保留。一支 720p、5 s 的影片需 10 點數，已包含聲音。