沙漠里漂浮的博物馆

文生视频:一段文字,生成一条带声音的视频

4 个模型可选,能做 4–30 秒、最高 2K 的视频,其中 4 个带声音。每条 25 积分($0.25)起,不用开会员,积分永久有效。 在 Artificial Analysis Text to Video Arena(2026年10月30日 核对)上,Wan 3.0 排第 1(Elo 1,175), MiniMax H3 第 3, Seedance 2.5 第 4。

写下画面里要发生什么,选一个模型,几分钟后就能拿到一条带声音的视频。文字生成视频前就能看到价格,用一次性购买的积分包付款,生成失败自动退回积分。

这段视频用的提示词。可以直接改,也可以自己写。

用这些模型生成的

每个都写了提示词、模型和在这里生成要花多少。选一个,从它开始改。

沙漠里漂浮的博物馆
“Cinematic quality, high-end feel, a golden desert at dawn, a minimalist white art-museum building floating above the dunes, the building's surface carrying fine stone texture and soft reflections. Sunlight passes through the airborne sand to form volumetric light, with distinct layers of distant dunes. The camera starts on an ultra-wide desert panorama, slowly pushing forward, passing through the flying sand grains, into the interior of the floating building. Inside there are suspended sculptures, semi-transparent silk installations, and a character in a white robe, the fabric swaying naturally in the wind. The camera makes a smooth orbiting move around the character, and finally the building walls slowly open to reveal the vast sky and desert. High-end art-advertising style, realistic lighting, refined materials, cinematic composition, elegant motion, surreal yet believable.”Seedance 2.5, 15 秒, 有声, 890 积分, 在 Dreamina 上生成
糕点师,30 秒一镜到底
“One unbroken handheld take, direct-flash 35mm snapshot look, fine grain, white tile and bright window light. 0-5s: from behind, a young pastry cook in a white jacket and a long black apron pipes the last pale-yellow buttercream flowers onto a small white cake on a scrubbed wooden bench in a narrow tiled pantry, steel bowls stacked on the shelves. 5-10s: she puts the piping bag down, turns the cake once on its plate to check it from every side, and straightens one flower with a fingertip. 10-15s: she lifts the cake on its plate, turns and carries it through the cramped kitchen past hanging ladles and a steaming stockpot, the camera following at shoulder height. 15-20s: she shoulders open a back door into a quiet, sunlit courtyard and walks across the flagstones. 20-25s: she sets the cake down on a weathered table laid for one and steps back to look at it. 25-30s: she wipes her hands on the apron and walks back inside, leaving the cake alone on the table in the sun. Sound: piping bag squeeze, a pot lid rattling, the door creak, birds outside.”Kling 4.0, 30 秒, 有声, 1,410 积分, 在 Kling AI 上生成
片场外景,30 秒一镜到底
“One unbroken thirty-second Steadicam move through a working film studio backlot at two in the morning. No cuts anywhere in the take. 0 to 6s: the camera walks out of a scene dock past stacked flats and a rail of period costume, catching a grip taping down a cable run, then pushes through hanging plastic sheeting into open air. 6 to 14s: it crosses a New York street standing set with the rain machines at full pressure, water sheeting off fire escapes and pooling in the gutters, a stunt performer in a soaked overcoat waiting on her mark while a focus puller measures to her face. 14 to 22s: it turns left down a service alley between two sound stages, past a catering truck with its shutter half down and an electrician asleep in a folding chair, the rain noise falling away behind the wall. 22 to 30s: it comes out onto the lot road, cranes up over a parked camera truck and holds on the whole backlot under sodium light, rain still falling on one street and nowhere else. Anamorphic, sodium and mercury practicals, deep blacks, fine film grain. Sound: rain machines roaring then dropping off, a radio crackle, boots on wet asphalt, a generator running under everything, no music.”Wan 3.0, 30 秒, 有声, 1,200 积分, 在 Wan 上生成
面摊老板:一句台词,然后起火
“Medium close-up, 50mm. A night-market noodle vendor in a white apron stands behind a steaming wok under bare hanging bulbs. He looks into the camera and says: "Thirty years. Same pan, same recipe." Then he tosses the noodles high and flames jump up from the wok. Camera: gentle living handheld that barely breathes, slight push-in. Audio: his voice, sizzling oil, crowd chatter, distant scooters. Photorealistic, warm practical light.”MiniMax H3, 10 秒, 有声, 80 积分
机械花绽放广告
“The theme is "mechanical flower bloom". The image needs to highlight the generative video model's strengths in lighting and shadow, art detail, realism, camera movement, and cinematic character feel. The overall style is a high-end tech-brand ad with strong visual impact. The video uses a single-take macro push-in shot, starting from a metal flower bud in darkness, gradually entering the precision mechanical structure inside the petals, and finally climaxing on the image of the mechanical flower fully blooming with light radiating outward. It requires realistic physical lighting, refined metal and glass materials, delicate mechanical motion, stable composition, cinematic color grading, no subtitles.”Seedance 2.5, 15 秒, 有声, 890 积分, 在 Dreamina 上生成
清晨划艇,15 秒长镜头
“One unbroken handheld take on 16mm film, misty grey dawn, a muted palette of wet timber, slate water and fog with one faded-red accent. 0-4s: inside an old timber boathouse on a still lake, a young woman in a faded red rowing vest and grey leggings lifts a pair of long wooden oars down from a wall rack. 4-8s: she carries the oars on her shoulder down the wet wooden slipway, the camera following behind her at shoulder height. 8-12s: she steps carefully into a narrow white single scull tied at the jetty, sits, and slots the oars into their riggers. 12-15s: she pushes off from the jetty with one hand and takes her first stroke, the scull gliding out into the fog, ripples spreading toward the camera. Sound: creaking boards, her footsteps, water lapping, one distant bird call.”Kling 4.0, 15 秒, 有声, 705 积分, 在 Kling AI 上生成
面摊老板:一句台词,然后起火
“Medium close-up, 50mm. A night-market noodle vendor in a white apron stands behind a steaming wok under bare hanging bulbs. He looks into the camera and says: "Thirty years. Same pan, same recipe." Then he tosses the noodles high and flames jump up from the wok. Camera: gentle living handheld that barely breathes, slight push-in. Audio: his voice, sizzling oil, crowd chatter, distant scooters. Photorealistic, warm practical light.”Wan 3.0, 8 秒, 有声, 160 积分, 在 Wan 上生成
厨师对着镜头说一句
“A chef in her forties looks straight into the camera in a bright open kitchen and says, "The secret is you never crowd the pan." She smiles and turns back to the stove. Clean daylight from a window, white tile, flour on her apron, a strand of hair loose. Medium close-up, 50mm, slight handheld. Sound: her voice close and clear, a gentle sizzle, a knife on a board off screen.”MiniMax H3, 5 秒, 有声, 40 积分
珊瑚礁里的水母
“A deep-sea coral-reef scene, a blue-toned tropical underwater world with large expanses of healthy, thriving colorful living coral, including branching coral, brain coral, plate coral, and soft sea-fan coral, with large schools of tropical fish weaving naturally among them. The foreground is crisp and vivid; the distance gradually turns bluer and grayer with reduced contrast. Natural light from above is filtered through the seawater into soft volumetric light, with fine suspended particles and a gentle sense of flow in the water, the soft corals swaying lightly and the fish schools moving smoothly and in coordination. Add a group of semi-transparent pale-lavender glowing jellyfish slowly appearing, 8-12 of them, in varying sizes. Their bell bodies carry a soft bioluminescent effect, tentacles drifting gently with the current, the tentacle tips bearing faint purple light points. The jellyfish are proportioned harmoniously with the corals and fish, the glow soft and not harsh, forming dreamy light spots against the dark seawater background. The jellyfish swarm's movement path is natural, rising in a slow ascending spiral, with some jellyfish swimming past in front of the lens to produce a slight lens-flare effect. Maintain the underwater blue-toned lighting, the jellyfish glow coexisting harmoniously with the faint reflections of the surrounding coral.”Seedance 2.5, 10 秒, 有声, 595 积分, 在 Dreamina 上生成
柔道过腰摔
“Real-time speed, 35mm documentary look, flat daylight from tall paper windows in an old wooden dojo with straw-yellow tatami. Two women judoka, one in a white judogi and one in a deep blue judogi, grip each other's lapels and sleeves. The woman in white steps in, turns her hip under her opponent and throws her in a clean hip throw; the woman in blue rotates over her hip through the air, lands flat on her back with a loud slap of the mat and immediately rolls up to her knees. Camera: low handheld, side-on, following the throw. Sound: bare feet on tatami, a sharp exhale, the slap of the landing.”Kling 4.0, 5 秒, 有声, 235 积分, 在 Kling AI 上生成
毛绒太空追逐
“A 20-second ultra-widescreen plush universe adventure clip, featuring ultra-fine plush textures throughout, cinematic space scale, a contrast between cute appearances and intense action, exaggerated FPV camera movements, and realistic soft-body physics. A spaceship made of plush fabric, buttons, zippers, and stuffing cotton carries three small plush animals — a rabbit, a fox, and a crow — fleeing at high speed through a universe composed of giant yarn planets, fluffy nebulae, and puppet giant beasts. A massive interstellar whale beast covered in deep blue long fur with eyes like two glass buttons chases the ship from within the nebula. The dynamics include the zipper ship launching via catapult, high-speed orbiting around a yarn planet, the giant whale bursting out from behind, navigating through a button asteroid belt, being swallowed by the whale beast, escaping through a golden zipper on its back, the whale slowly deflating as if losing its stuffing cotton, and finally flying through a pink-purple fluffy nebula toward a colorful button sun.”Wan 3.0, 20 秒, 有声, 400 积分, 在 Wan 上生成
黏土动画青蛙
“Stop-motion claymation: a round green frog in a knitted scarf carries a stack of pancakes across a tiny kitchen set, sets it down, then looks at camera and blinks. Visible fingerprints in the clay, felt walls, warm practical lamps, stepped 12fps motion. Sound: playful pizzicato strings, soft clay squeaks, a plate clink.”MiniMax H3, 5 秒, 有声, 40 积分

文字生成视频怎么做:4 步

  1. 1

    选一个模型

    默认选中字节跳动 Seedance 2.5。想要最长 30 秒的一镜到底,换成快手可灵 Kling 4.0 或阿里通义万相 Wan 3.0;想快速、便宜地试稿,选 MiniMax 海螺 H3(海螺 AI 的视频模型)。下面的表格列出了每个模型能做什么、要多少积分。

  2. 2

    写清楚这个镜头

    先写镜头,再写画面里是谁、在做什么、能听到什么。要说的台词放在引号里。下面的提示词写法有详细说明。

  3. 3

    设定时长、尺寸和声音

    选好时长、分辨率、画面比例(TikTok、Reels、Shorts 这类竖屏平台用 9:16),再决定要不要声音。每改一项,价格都会跟着更新,生成前不扣积分。

  4. 4

    生成,再慢慢调

    多数视频几分钟就好,都会留在你的生成记录里。先用短时长或低分辨率试措辞,满意了再用完整画质跑一次。

支持的模型和价格

更新于 2026-10-30

模型时长最高分辨率声音画面比例5 秒一条10 秒一条榜单排名
Seedance 2.54–15s1080p有16:9, 9:16, 1:1, 4:3, 3:4, 21:9300 积分$3.00595 积分$5.95第 4 · Elo 1,157Dreamina Seedance 2.5
Kling 4.05–30s1080p有16:9, 9:16, 1:1, 21:9160 积分$1.60315 积分$3.15还没上榜
Wan 3.05–30s1080p有16:9, 9:16, 1:1, 4:3, 3:4100 积分$1.00200 积分$2.00第 1 · Elo 1,175
MiniMax H35–15s2K有16:9, 9:16, 1:1, 4:3, 3:4, 21:940 积分$0.4080 积分$0.80第 3 · Elo 1,161MiniMax H3 (768p)

每条价格按 720p、带声音计算 (MiniMax H3: 768p),取各模型最便宜的版本;分辨率越高越贵。 榜单排名是盲测投票得出的名次和 Elo 分(AA-Video-T2V v2.1, with audio,2026年10月30日 核对),来源: Artificial Analysis Text to Video Arena

文生视频样片,每条都附提示词

每个都附上原样的提示词、设置和价格。可以直接在上面的生成器里用,也可以先改再用。

陶罐推镜
在 Wan 上生成

Wan 3.0

一个镜头拍完的商品视频

阳光房里的一只陶罐,镜头慢慢推近,一镜到底。提示词来自 Fuser 的万相 3.0 和 Seedance 2.5 对比测试。

  • 文生视频
  • 720p
  • 5 秒
  • 有声
  • 16:9

“Slow dolly-in on a matte terracotta vase standing on a travertine plinth in a sunlit plaster room. Late-afternoon light slides slowly across the wall as dust drifts through the beam. Calm, premium product film, shallow depth of field. Audio: soft room tone and a faint breeze.”

建议:每条视频只写一个镜头运动,并写明速度,例如“slow dolly-in”。

100 积分
曲奇转台展示(竖屏)
在 Dreamina 上生成

Seedance 2.5

社交平台的竖屏广告

选 9:16 时,Seedance 2.5 生成的是真正的竖构图,不是从横屏裁出来的,还自带声音。这条一句话的提示词出自字节跳动自己的教程。

  • 文生视频
  • 720p
  • 6 秒
  • 有声
  • 9:16

“A strawberry sandwich cookie slowly completes one full rotation while the camera moves smoothly around it and keeps the product centered.”

建议:主体放在画面中间,字幕和 App 按钮就挡不到它。

360 积分
厨师对着镜头说一句

MiniMax H3

对着镜头说一句台词

她看着镜头,一字不差地说出写好的台词,周围是厨房的环境声。用的是速度快的 H3 Max Turbo 版本。

  • 文生视频
  • 768p
  • 5 秒
  • 有声
  • 16:9

“A chef in her forties looks straight into the camera in a bright open kitchen and says, "The secret is you never crowd the pan." She smiles and turns back to the stove. Clean daylight from a window, white tile, flour on her apron, a strand of hair loose. Medium close-up, 50mm, slight handheld. Sound: her voice close and clear, a gentle sizzle, a knife on a board off screen.”

建议:台词放进引号,并写明谁在说、怎么说:“looks straight into the camera … and says, “…””。

40 积分
糕点师,30 秒一镜到底
在 Kling AI 上生成

Kling 4.0

一镜到底 30 秒

一个不间断的长镜头里有六个定好时间的动作,每个都落在自己的时间段。提示词来自 Fuser 的可灵 4 教程。和 Fuser 的结果一样,画面是柔和的日光,而不是提示词要的直闪灯效果。

  • 文生视频
  • 1080p
  • 30 秒
  • 有声
  • 16:9

“One unbroken handheld take, direct-flash 35mm snapshot look, fine grain, white tile and bright window light. 0-5s: from behind, a young pastry cook in a white jacket and a long black apron pipes the last pale-yellow buttercream flowers onto a small white cake on a scrubbed wooden bench in a narrow tiled pantry, steel bowls stacked on the shelves. 5-10s: she puts the piping bag down, turns the cake once on its plate to check it from every side, and straightens one flower with a fingertip. 10-15s: she lifts the cake on its plate, turns and carries it through the cramped kitchen past hanging ladles and a steaming stockpot, the camera following at shoulder height. 15-20s: she shoulders open a back door into a quiet, sunlit courtyard and walks across the flagstones. 20-25s: she sets the cake down on a weathered table laid for one and steps back to look at it. 25-30s: she wipes her hands on the apron and walks back inside, leaving the cake alone on the table in the sun. Sound: piping bag squeeze, a pot lid rattling, the door creak, birds outside.”

建议:开头写“One unbroken take”,每 5 秒写一个动作(“0-5s: …, 5-10s: …”)。

1,410 积分
柔道过腰摔
在 Kling AI 上生成

Kling 4.0

两个人之间的快速动作

一个干净的过腰摔,每条胳膊腿都留在对的人身上,落地的拍击声也在音效里。不过镜头是固定的,没有按提示词做低机位手持跟拍。

  • 文生视频
  • 1080p
  • 5 秒
  • 有声
  • 16:9

“Real-time speed, 35mm documentary look, flat daylight from tall paper windows in an old wooden dojo with straw-yellow tatami. Two women judoka, one in a white judogi and one in a deep blue judogi, grip each other's lapels and sleeves. The woman in white steps in, turns her hip under her opponent and throws her in a clean hip throw; the woman in blue rotates over her hip through the air, lands flat on her back with a loud slap of the mat and immediately rolls up to her knees. Camera: low handheld, side-on, following the throw. Sound: bare feet on tatami, a sharp exhale, the slap of the landing.”

建议:用衣服区分谁做什么(“the woman in white … the woman in blue”),模型就不容易把两个人搞混。

235 积分
机械花绽放广告
在 Dreamina 上生成

Seedance 2.5

高端品牌短片

金属和玻璃上的光,是 Seedance 2.5 最拿手的。提示词写了风格、一个微距推镜和材质,视频从一个暗色花苞一直跟到花朵带着光打开。

  • 文生视频
  • 720p
  • 15 秒
  • 有声
  • 16:9

“The theme is "mechanical flower bloom". The image needs to highlight the generative video model's strengths in lighting and shadow, art detail, realism, camera movement, and cinematic character feel. The overall style is a high-end tech-brand ad with strong visual impact. The video uses a single-take macro push-in shot, starting from a metal flower bud in darkness, gradually entering the precision mechanical structure inside the petals, and finally climaxing on the image of the mechanical flower fully blooming with light radiating outward. It requires realistic physical lighting, refined metal and glass materials, delicate mechanical motion, stable composition, cinematic color grading, no subtitles.”

建议:给自己的产品写提示词时,照这个结构来:风格、一个镜头、材质,最后加“no subtitles”。

890 积分
面摊老板:一句台词,然后起火
在 Wan 上生成

Wan 3.0

先说台词,再上高潮

他先按原文说完台词,炒锅才腾起火焰。提示词来自 Krea 的 Seedance 2.5 和万相 3.0 对比测试,那次也是万相 3.0 做出了火焰,Seedance 没有。他身后招牌上的字是模型编出来的。

  • 文生视频
  • 720p
  • 8 秒
  • 有声
  • 16:9

“Medium close-up, 50mm. A night-market noodle vendor in a white apron stands behind a steaming wok under bare hanging bulbs. He looks into the camera and says: "Thirty years. Same pan, same recipe." Then he tosses the noodles high and flames jump up from the wok. Camera: gentle living handheld that barely breathes, slight push-in. Audio: his voice, sizzling oil, crowd chatter, distant scooters. Photorealistic, warm practical light.”

建议:台词放在前面,大动作放在后面(“first …, only then …”)。

160 积分
黏土动画青蛙

MiniMax H3

定格动画风格

黏土质感和一顿一顿的动作从头保持到尾。大约第 4 秒时,它自己切到了一个特写,提示词里并没有要求。

  • 文生视频
  • 768p
  • 5 秒
  • 有声
  • 16:9

“Stop-motion claymation: a round green frog in a knitted scarf carries a stack of pancakes across a tiny kitchen set, sets it down, then looks at camera and blinks. Visible fingerprints in the clay, felt walls, warm practical lamps, stepped 12fps motion. Sound: playful pizzicato strings, soft clay squeaks, a plate clink.”

建议:不想切镜头,就加上“one continuous shot”。

40 积分

文生视频提示词怎么写

提示词里写什么,按模型最好理解的顺序。

先写镜头

开头写景别、镜头焦段或镜头运动,再写主体、动作和光线。好几份提示词教程都用这个顺序,而且构图写在最前面时,模型改写提示词后更能保住它。

“Medium close-up, 50mm, slow push in. A night-market noodle vendor in a white apron stands behind a steaming wok…”

一条视频一个动作,或者给每个动作定时间

5 秒的视频放一个动作最稳。更长的镜头,给每一段写上时间范围:Fuser 测到万相 3.0 的切点和写好的时间只差一帧左右,它的可灵 4 教程也建议大约每 5 秒一个动作。

“0-5s: she pipes the last flower. 5-10s: she steps back and wipes her hands…”

不想切镜头,就写“one continuous shot”

可灵 4.0 默认会生成多镜头视频,万相 3.0 有时也会在中途换角度。两家官方教程的解决办法一样:在第一行就写明。

“One unbroken take, no cuts. The camera follows…”

台词加引号,而且放在前面

要说的话放进引号,写明谁说、用什么语言说,并把这句台词安排在视频前段,让它有时间说完。写在最后一秒的台词经常被截断。

“He looks up and says in English, “Thirty years, same pan, same recipe.” Only then the wok flares.”

写清楚要什么声音、不要什么声音

这里的每个模型都会生成声音,你不写,它就自己配,常常是音乐。字节跳动的教程只在声音和字幕上用否定句:结尾写“No background music”或“No subtitles”。

“Sound: oil hissing, a crowd murmuring, no background music. No subtitles.”

具体的细节比形容词有用

“电影感”“史诗级”这类词,模型抓不住。一种颜色、一种材质、一天里的某个时间、背景里的一件东西,它才抓得住。别指望它数准东西的数量,也别指望小字能拼对。

“A red bicycle leaning on a wet brick wall at dusk, a sodium street lamp flickering behind it.”

文生视频模型哪个好?四个模型怎么选

四个模型都能根据文字生成带声音的视频,但脾气各不相同。下面的说法来自 Fuser、Krea、Atlas Cloud 的独立同提示词测试,模型厂商自己的提示词教程,以及我们自己跑的样片。排名和价格见上面的表格。

阿里通义万相 Wan 3.0 最听提示词的话。在 Krea 的测试里,它给出了“对提示词最字面、最有戏剧性的版本”;在 Fuser 用同样输入跑的五轮测试里,它抓住了 Seedance 2.5 漏掉的细节。它一次能生成最长 30 秒,每秒价格比 Seedance 2.5 低。缺点是有时会在中途换角度,声音混得偏响(Krea、Fuser)。

字节跳动 Seedance 2.5 画面最精致、最有电影感:金属、玻璃和皮肤上的光都很好看,混音更安静,对白支持十多种语言。海外版即梦 Dreamina 用的也是 Seedance 2.5。它是四个模型里每秒最贵的,720p 下特写的脸会偏糊,快速动作可能变形。字节跳动的教程建议每个镜头只用一种镜头运动(ByteDance)。

快手可灵 Kling 4.0 适合动作戏和长镜头:每位测试者的 15 到 30 秒一镜到底都撑住了;在我们的柔道样片里,每条胳膊腿都留在对的人身上;招牌和字幕也都写对了。不特别说明的话,它会加剪辑点和音乐,也会忽略一些风格词,比如某种胶片质感(Fuser)。

MiniMax 海螺 H3 在这里有两个版本。H3 Max Turbo 是本页最快、最便宜的选择,适合在花更多积分之前先试措辞。官方版 H3 多了 2K。它的 768p 输出是 1344×768,不是 720p;它还会改写太短的提示词,可能因此多出一个剪辑点(Krea)。

文生视频还是图生视频?

文生视频的一切都从你的文字凭空生成,所以同一条提示词跑两次,会得到两条不同的视频。如果产品、人物或构图必须和你已有的东西一致,就从一张图开始:图生视频会把你的图片当作第一帧。常见的做法是先用 AI 生图做出静态图,再让它动起来。

时长、分辨率和画面比例怎么选

选能装下动作的最短时长。5 秒放一个动作;10 秒放两三个定好时间的动作;20 到 30 秒适合在可灵 4.0 或万相 3.0 上拍一个长镜头。分辨率对价格的影响最大,而且用更高分辨率重跑是一条新视频,不是把草稿放大,所以 720p 只用来试措辞。画面比例要在生成前选好:16:9 用于 YouTube 和网站,9:16 用于手机,1:1 用于信息流;想要电影宽银幕效果,可灵 4.0、Seedance 2.5 和海螺 H3 还有 21:9。

文生视频怎么收费

每条视频都有一个积分价格,取决于模型、时长、分辨率和是否带声音,生成前就能看到。积分来自一次性购买的积分包,永久有效、不清零,没有要取消的包月。视频失败、超时或被拒,积分自动退回你的余额。

本页样片是怎么做的

每条样片都标明了用的提示词、模型和设置,以及它今天在 zzo.ai 上的价格。多数提示词来自公开发表的测试和教程,出处写在每张卡片上,你可以拿我们的结果和他们的对比。有些样片是在模型厂商自己的 App 上,用同样的模型、提示词和设置生成的,卡片上会注明。所有视频都没有剪辑过;哪条没做到提示词里的某一部分,卡片上也会写出来。

结果不对怎么办

常见的问题、原因和改法。

视频自己切到了另一个角度

原因: 可灵 4.0 默认开着多镜头;测试者也见过万相 3.0 在明确说了不要切的情况下,还是在中途切了镜头。

改法: 提示词开头写“One continuous shot, no cuts”,动作留在一个地方。还是会切,就再跑一次,或者换 Seedance 2.5。

出现了不想要的音乐,或者声音太大

原因: 模型会用自己的配乐填满安静的地方。独立测试发现,可灵 4.0 和万相 3.0 的混音比 Seedance 2.5 响。

改法: 写清楚你要的声音,结尾加“No background music”。音量可以在剪辑软件里调小;用 Seedance 2.5 和万相 3.0 时也可以直接关掉声音。

招牌、标签上的字写错了

原因: 画面里的文字对视频模型来说仍然很难,字节跳动和阿里在各自的说明里都承认这一点。

改法: 文字写短、放进引号,或者在剪辑软件里加标题。可灵 4.0 把我们样片里的招牌和字幕都写对了。

动作一快,手、四肢或物体就变形

原因: 又快又大的动作是所有模型最难处理的部分;字节跳动的教程建议用缓慢、连续的动作。

改法: 把动作放慢,或者一次只写一个动作。打斗、运动、舞蹈这类画面,独立测试里可灵 4.0 做得最好。

台词没说完就断了

原因: 写在视频末尾的台词,没剩多少时间让它说完。

改法: 把台词挪到开头、写短一点,或者选更长的时长。

生成被拒绝或失败了

原因: 模型自己的审核、或者我们的内容检查拦下了提示词或结果,也可能是模型超时。在 Seedance 2.5 上,品牌名、胶片型号和导演名字也可能被拒。

改法: 不扣积分:积分会自动退回。去掉真人和品牌的名字再写一次,也可以看看《可接受使用政策》。

常见问题

更新于 2026-10-30

文生视频是什么意思?

就是用文字生成视频:你写一段话,描述画面里有什么、发生什么、能听到什么,AI 模型据此生成一条视频。这里的四个模型都会把声音和画面一起生成。如果你手上已经有一张图,想从这张图开始动,那叫图生视频。

文生视频模型哪个好?现在排名第一的是哪个?

看你要做什么。在 Artificial Analysis 的文生视频排行榜上(由盲测投票得出),万相 3.0 排第一,海螺 H3 和 Seedance 2.5 紧随其后(具体名次和我们核对的日期在本页顶部)。可灵 4.0 还没有上榜;独立测试者认为它在快速动作和长镜头上表现最好。

文生视频的 AI 工具有哪些?我该先用哪个?

本页提供四个:做精致的广告和带对白的视频,先用 Seedance 2.5;要严格按提示词来、做长镜头、价格低一些,用万相 3.0;人物之间的动作和画面里的文字,用可灵 4.0;想快速、便宜地试稿,用海螺 H3(H3 Max Turbo)。四个模型都在 zzo.ai 网页上直接用,一个账号、一份积分余额。

文生视频多少钱一条?

取决于模型、时长、分辨率和是否带声音。本页的表格列出了每个模型 5 秒和 10 秒的价格,生成器在你开始前会显示准确价格。每条视频用积分付款,不用开会员。

文字能生成多长的视频?

可灵 4.0 和万相 3.0 一次最长 30 秒一镜到底,Seedance 2.5 和海螺 H3 最长 15 秒。要做更长的片子,每个镜头生成一条,再在剪辑软件里拼起来。

文字生成视频能带声音吗?有台词、音效和音乐吗?

能。这里的每个模型都在同一次生成里做出画面和声音。把声音和台词写进提示词就行。Seedance 2.5 和万相 3.0 可以关掉声音;可灵 4.0 和海螺 H3 一直带声音。

能用中文写提示词,或者让人物说中文吗?

本页样片的提示词都是英文原文,大多来自公开测试和官方教程,我们核实过的也是英文的效果。阿里的官方文档写明万相 3.0 的提示词支持中文和英文。对白方面,快手公布的可灵 4.0 支持语言里有中文,MiniMax 也把中文列为海螺 H3 稳定支持的语言之一;可灵官方教程的示例就是用英文写画面,把中文台词放在引号里。

文生视频提示词怎么写?

按这个顺序写:镜头(景别、运动)、主体、动作、光线、声音。一条视频写一个动作,长镜头给每段定时间;台词加引号、放在前面;不想切镜头就写“One continuous shot”;不想要配乐就写“No background music”。上面的提示词写法有每一条的示例。

能做 TikTok、Reels、Shorts 用的竖屏视频吗?

能。在生成器里选 9:16。四个模型的文生视频都支持 9:16,表格里列出了每个模型支持的全部画面比例。

AI 能在视频里写字或加字幕吗?

有时可以。可灵 4.0 把我们样片里的招牌和字幕都写对了,MiniMax 也说海螺 H3 能准确生成文字;Seedance 2.5 和万相 3.0 经常把画面里的字写错。必须一字不差的文字,在剪辑软件里加。

为什么视频会自己切到别的镜头?

有些模型会自己加剪辑点,尤其是可灵 4.0 和万相 3.0。在提示词开头写“One continuous shot, no cuts”,要求一镜到底。

生成一条视频要多久?

一般 1 到 5 分钟;1080p 的长镜头可能要 10 分钟以上。生成好的视频留在你的生成记录里,随时可以下载。

生成失败或被拦下了,积分怎么办?

这条视频的积分会自动退回你的余额,不管是模型失败、超时、拒绝了提示词,还是结果被我们的内容检查拦下。你只为拿到手的视频付积分。

能生成真人或明星的视频吗?

公众人物不行,没有本人同意的任何人也不行。在提示词里描述的虚构人物没问题。《可接受使用政策》列出了哪些内容不允许。

生成的视频可以商用吗?

可以。你生成的视频归你使用,包括广告和给客户做的项目,前提是遵守我们的条款,并且不照搬别人的品牌、角色或肖像。

要开会员吗?

不用。zzo.ai 只卖一次性的积分包,用多少买多少。积分永久有效、不清零,所有模型都能用,网站和 API 通用。

用文生视频还是图生视频?

从一个想法开始,用文生视频。画面风格、产品或人物必须和你手上的某张图一致,用图生视频:视频会从你的图片开始。

更多工具