APIMODELSAPIMODELS

Seedance 2.5 对白与口型提示词

Seedance 2.5 视频提示词,每条都配上它生成的那段成片和作者署名。作者公开过原文的照原样收录;没公开的,我们从成片反推一条并明确标注。悬停播放,一键复制。

← 回到全部 95 条 Seedance 2.5 提示词

最近更新: · Seedance 2.5 API 价格与在线体验 · API 文档

一句话摘要

Seedance 2.5 是字节跳动的视频模型,2026 年 6 月 23 日在火山引擎 FORCE 大会发布。已上线的 API 上,它一次生成 4 到 30 秒(也可以传 duration -1 让模型自己决定长度),单次最多接受 50 份参考素材(图片最多 30 张、单张不超过 4K,视频与音频各最多 10 段、各自总时长不超过 30 秒);新增视频编辑与视频延长两类任务,音频与画面在同一潜空间生成。分辨率是 480p 或 720p;FORCE 发布会上展示的原生 4K 10-bit,API 并没有暴露这一档。字节还为它发布了官方提示词指南。本页收录 Seedance 2.5 视频提示词,每条附它生成的成片,可自由复制。2026 年 8 月 8 日起,你可以在 apimodels.app 上以模型名 seedance-2.5 跑 Seedance 2.5,$0.120/秒起;本库这些提示词的作者分别是在即梦(Dreamina)和 Higgsfield 上跑的。

出处:ByteDance Seed(模型方官方站点)

悬停预览 · 点击有声

七镜卧室纪实——猫、来电、门铃

Montage, multi-shot candid observational footage. Do not use a single camera angle or continuous take. Handheld documentary style with the feeling of accidental real-life capture. Slightly imperfect framing, subtle handheld shake, tiny reframing adjustments, gentle exposure breathing, and autofocus that settles half a beat late. Realistic skin texture, soft indoor natural light, film grain, shallow depth of field. Relaxed breathing, natural blinking, restrained and authentic performance. Total of 7 shots. The woman from @ Image 1 is wearing a matching cotton pajama set consisting of a scoop-neck sleeveless top and loose shorts made from the same fabric and design. Barefoot, she lies on her stomach across a lived-in, slightly messy bed in her bedroom, browsing a mobile shopping page on a black smartphone. A house cat jumps onto the bed seeking attention, and she affectionately plays with it. Moments later, a friend calls. She answers, chats, and eventually bursts into laughter. The doorbell rings, so she ends the call and gets up to collect her dinner. Shot 1 (0–2s): She is already moving. One thumb scrolls through a shopping page while she frowns slightly, thinking. She quietly mutters in Korean: "아, 뭐 살랬더라…" ("Ah... what was I going to buy again?"). Lips synchronize naturally. Her bare feet gently sway behind her. A low handheld camera glides along the edge of the mattress, with soft bedding partially obscuring the foreground. Shot 2 (2–4s): The house cat jumps onto the bed, compressing the blanket as it walks toward her forearm and lets out a single meow. The camera gives a slight jolt from the impact, then quickly reframes from the phone to an over-the-shoulder view that includes her face, hands, phone, and cat. Shot 3 (4–6s): She turns warmly toward the cat, stroking it once from its forehead to its shoulders. In a gentle, affectionate tone she says: "우리 애기 왔어?" ("Did my baby come?"). Lip sync is clear and natural. The camera pushes in briefly through the cat's softly blurred foreground toward her face and hand. Shot 4 (6–8s): A friend's ringtone sounds. She glances down, confirms the caller, swipes to answer, and brings the phone to her ear. The camera moves in a loose semicircle from a slightly tilted overhead angle, capturing the entire answering motion. Shot 5 (8–11s): Her friend begins talking. She responds first with confusion, then disbelief: "어, 왜? 진짜 거짓말." ("Huh? Why? No way, you're kidding."). Immediately afterward she naturally bursts into laughter. The cat kneads the blanket beside her. A close handheld side angle keeps both her phone and profile within the same focal plane. Shot 6 (11–13s): A clear doorbell interrupts the laughter. Both she and the cat turn toward the bedroom door. She braces one hand on the mattress to get up. The camera reacts a fraction too late, briefly turning toward the door before naturally correcting back to her rising movement. Shot 7 (13–15s): Still on the phone, she hurriedly says: "어, 나 밥 왔다. 끊어." ("My food's here. I'll hang up."). She ends the call with her thumb, gets off the bed, and walks toward the door while the cat follows across the blankets. The camera slides low beneath her elbow and tilts upward as she stands in one continuous motion. Sound: No music. Only raw production sound: bedding rustling, quiet breathing, finger taps on the phone, the soft impact of the cat jumping onto the bed, one natural meow, faint purring, a single short incoming ringtone, call connection tone, the friend's voice through the phone, the woman's four Korean lines with accurate lip sync, a genuine brief laugh, one clear doorbell, the tap ending the call, and the soft cat footsteps. No subtitles, no on-screen text, no logo, no watermark. Never render a reference sheet or duplicate the subject.

来自 oggii_0看原帖
悬停预览 · 点击有声

两个好友闲聊·日式日常

A cinematic Japanese slice-of-life scene.Two young Japanese women (age 20 and 22) are sitting together and chatting happily. They are close friends having a relaxed conversation. Their lips move naturally with synchronized dialogue (accurate lip sync).Scene: a quiet residential street in Japan in the late afternoon. Warm golden sunlight, gentle breeze, peaceful everyday

来自 tanabe_fragm看原帖
悬停预览 · 点击有声

用参考音色驱动的播客访谈

El entrevistador esta en su podcast entrevistando a @8993a3ce-444a-414b-9800-6798a9c997eb y le pregunta con su voz @0949c9e6-24ae-4895-b567-94aa844e0150: " Buenas mi gente, tenemos un nuevo modelo de IA y hemos traido a un experto en ello, Nabil cuentanos" y entonces el invitado dice con su voz @8993a3ce-444a-414b-9800-6798a9c997eb: "Buenas jordi, la verdad que es una cosa de locos. Gracias a este modelo puedo hacer cosas como esta" y se levanta chasquea los dedos y le sale pelo. Se sienta y dice con su voz @7ac71528-ba91-42ec-92e1-89875ed6e49f: "inténtalo tú" y jordi se levanta y chasquea los dedos y entra un equipo de rugby en el podcast y nabil y jordi empiezan a empujarse con el equipo entero como en una pelea. De repente la camara enfoca a nabil en medio de la pelea y este con su voz @7ac71528-ba91-42ec-92e1-89875ed6e49fdice :" Esto solo es posible con el nuevo modelo de seedance, que loco!" y le da un empujon al que tiene encima forcejeando. De repente jordi en medio de la pelea grita con su voz @0949c9e6-24ae-4895-b567-94aa844e0150: "STOP" y todos los de rugby se convierten en plumas que caen al suelo. Jordi y Nabil se miran y les entra un ataque de risa #seedance2.5 #dreaminacpp @dreamina_ai

来自 Nabilidosoo看原帖
悬停预览 · 点击有声

带同步对白与原创配乐的短片

Create a 30 second cinematic short film titled The Suit Was Still Listening, presented in 16:9 with synchronized dialogue, environmental sound, and original music. The film must feel completely live action, physically grounded, and captured on location, never glossy or synthetic. Visual language: a rain soaked neighborhood laundromat at midnight, warm amber fluorescent tubes inside, cold blue street light and moving car reflections outside, wet pavement, fogged windows, chipped enamel machines, baskets, detergent boxes, loose receipts, realistic skin texture, natural eye moisture, subtle fabric lint, imperfect hair, restrained film grain, gentle halation, lifted shadows, rich but believable color. Wardrobe creates bold editorial color blocking: Mara wears a cobalt trench over a cream blouse, Noah wears a rust knit polo, the teenage attendant wears faded green workwear, and an elderly customer wears a pale gray suit and dark hat. Use expressive medium close shots, tactile inserts, occasional low angles, clean profile compositions, and brief handheld movement that feels operated by a human. Preserve consistent faces, clothing, props, screen direction, reflections, hand anatomy, and spatial continuity. Use realistic speech timing, breathing, blinking, eye focus, and precise lip synchronization. 0 to 3 seconds. Exterior wide shot through falling rain. A lonely laundromat glows on a dark corner. A black sedan turns slowly into the street. Camera makes a subtle forward creep. Music begins with muted upright bass, brushed snare, a low analog pulse, and the rhythmic churn of washers. 3 to 6 seconds. Interior macro shot through a round washer door. A red scarf turns through soap and water. Cut to Mara opening a dryer. A tiny black transmitter falls from the lining of a cream suit and strikes the tile with a sharp metallic click. The score briefly drops out. 6 to 10 seconds. Low medium two shot. Noah freezes beside a folding table. He says quietly, You said clean the suit. Mara studies the transmitter in her palm and replies, I did. You hid a heartbeat in the lining. Keep the delivery controlled and intimate, not theatrical. 10 to 14 seconds. Tight insert on the teenage attendant checking an old signal meter beside the change machine. A green needle jumps. He looks toward the fogged front window and says, It is still transmitting. Camera racks focus from the meter to white headlights sliding across the glass. 14 to 18 seconds. The sedan stops outside. The elderly customer continues folding a shirt without looking up. He says, Then he is already here. Hold on Mara as the moving headlights carve across her face. Let silence, rain, and machine motors carry the tension. 18 to 22 seconds. Noah whispers, Back door. The old man calmly turns the deadbolt on the back exit and answers, Too obvious. Mara notices a delivery rider outside pushing a rolling laundry cart past the window. Use a fast sequence of close shots: her eyes, the transmitter, the cart wheel, the sedan mirror. 22 to 27 seconds. Mara crosses naturally behind a row of machines, slips the transmitter into a sealed laundry bag on the moving cart through an open service hatch, and returns without drawing attention. Noah asks, What did you do. Mara watches the cart continue down the wet street and says, Gave him a cleaner story. 27 to 30 seconds. Exterior telephoto view. The black sedan pulls away and follows the cart. Cut back inside to a symmetrical medium shot of the four characters framed by spinning washer doors. The red scarf circles behind Mara like a slow warning light. She sits opposite Noah and says, Now tell me who owns the suit. End on the transmitter signal fading from the attendant meter as the final bass note lands. Sound design must include rain on glass, distant tires on wet road, washer motors, dryer buttons, fabric movement, the transmitter click, door lock, fluorescent hum, and soft room reflections around every voice. Dialogue must remain clean and naturally mixed above the music. Avoid visual distortion, artificial camera acceleration, floating objects, excessive shallow focus, oversharpening, plastic skin, extra fingers, unreadable signage, random background motion, or dreamlike effects.

来自 bmx_ai13看原帖
悬停预览 · 点击有声

外星指挥官下令(土耳其语)

Image1 komutan ve lider olan bir uzaylıdır. Image2 ise asker uzaylıdır. komutan askere bir savaş alanında emir veriyor. Karşılıklı diyalog sahnesi. Konuşmalar tamamen uydurma bir uzaylı dilinde olmalı Bir blockbuster filminden bir savaş sahnesi olmalı. Sinematik kamera hareketleri ve açıları kullanmalısın.

来自 ozansihay看原帖

怎么写好 Seedance 2.5 提示词

  1. 1按固定顺序写. 主体 → 动作 → 地点 → 运镜 → 光线 → 画面质感。「一个人在走路」只会给你一段素材库循环;同一个主体加上镜头怎么动、光从哪来,才是一个镜头。在 2.5 上真烧过点数的创作者,写法顺序基本都收敛到这一条。
  2. 230 秒是「分阶段」写出来的,不是把句子写长. 官方指南把长片拆成【第一阶段】【第二阶段】【第三阶段】,每段有开场状态、一个主线事件、以及下一段要继承的结束状态。把 10 秒的提示词堆满形容词并不会变成 30 秒;把事件拆成层层继承的状态才会。
  3. 3每个时间段只放一个主要动作,而且只描述一次. 无论你用阶段还是直接写 0—5 秒 / 5—10 秒,一段只给一个主要动作。否则两头坏:五秒里塞三个动作会被一起压扁;而官方明确说过,同一个动作描述两遍反而会劣化。
  4. 4每一份参考素材都要给职责,也要给排除项. 不要只说「这是我的图」,要写「@图1 提供装甲设计与配色关系。不要采用背景。」单次最多可挂 50 份素材,但挂得越多、越不指派职责,稳定性掉得越快。
  5. 5给主体起名字,然后一直用这个名字. 先写一次 <士兵>、<醋饭>、<女生>,之后每次都用这个标记。30 秒里,「她」和「那个男人」会漂到错误的身体上——命名标记是成本最低的身份锁。
  6. 6声音要写成音效清单,不是写气氛. 音频和画面是同一次生成出来的,所以直接列:环境声、具体的拟音、注明语种的对白、以及不要出现什么。本库好几条提示词结尾都是一整份声音清单,这正是它们看起来「完成度高」的一大半原因。
  7. 7主动要求你想要的缺陷. 手持抖动、慢半拍才合上的自动对焦、曝光呼吸、把脸切掉一半的构图。这个模型里的真实感就是一份写明的缺陷清单——不写,你写多少遍「纪实」都还是干净的广告片。
  8. 8结尾放负向清单,参数不要写进提示词. 最后写清不该出现什么:不要字幕、不要水印、不要多出手脚;用参考图时还要加一句——永远不要把参考拼图本身画出来,也不要复制主体。分辨率、时长、画幅是生成设置,不是提示词内容。

提示词公式

主体 + 动作 + 地点 + 运镜 + 光线 + 画面质感 + 声音。超过十秒左右,就把它包进计时的阶段里:每个阶段一个主要动作、一个供下一段继承的结束状态,最后收一段负向清单。

例:<咖啡师> 在打烊的咖啡馆里拉完当晚最后一杯,把杯子放到出品台上。窗上有雨,椅子已经翻上桌面,只有咖啡机上方一盏暖色吊灯亮着,其余部分在暗处。镜头先用低机位中景对着手柄,待杯子落下时升起转为侧脸。写实质感,暖钨丝光对冷窗光,细腻颗粒。声音:磨豆机停转、蒸汽嘶声、雨声、一次杯底磕在木台上的声音,没有配乐。不要出现顾客、字幕或水印。

常见问题

Seedance 2.5 是什么?

Seedance 2.5 是字节跳动的视频生成模型,2026 年 6 月 23 日在火山引擎 FORCE 大会发布。已上线的 API 上:一次生成 4 到 30 秒(也可以传 duration -1 让模型自己定长度)、单次最多 50 份多模态参考素材(图片最多 30 张,视频与音频各最多 10 段)、新增视频编辑与视频延长两类任务(换个产品或背景不必重生成整条),以及与画面在同一潜空间生成的音频。输出分辨率是 480p 或 720p —— FORCE 发布会宣称的原生 4K 10-bit,API 上并没有这一档。字节还为它发布了官方提示词指南,我们整理过完整的英文版。

这些提示词免费吗?

免费。本页每一条都可以直接复制走,拿到任何能调 Seedance 2.5 的地方去跑,不需要登录也不需要付费。

提示词的作者是谁?

分两类,标注不同。作者自己公开过的,按他原本的语言原样收录、带署名并链回原帖,版权归原作者。作者从未公开提示词的成片,我们用视觉模型从成片反推一条,并标记为「反推」:它描述的是**结果**,因此拿不到负向约束、精确台词和参考图工作流,是写法参考,不是作者的原始提示词。如果你是作者、希望撤下或更正署名,发邮件给我们即可。

现在在哪能跑 Seedance 2.5?

**在 apimodels.app 上就能跑**,模型名 seedance-2.5,2026 年 8 月 8 日起——480p $0.120/秒、720p $0.270/秒,仅成功扣费。这是一次状态变更:2026 年 8 月 4 日探我们生产用的火山方舟账户时,所有 Seedance 2.5 的 model id 还都返回 InvalidEndpointOrModel.NotFound,当时我们如实写了「调不到」,而不是给你一个点了生成就报错的页面。同一把 key 还能调 Seedance 2.0(需要 1080p 时仍然用它,$0.092/秒起)、MiniMax H3、Kling、VEO 和 Grok。本库这些提示词的作者,分别是在字节自家的即梦(Dreamina)和 Higgsfield 上跑的。

给 Seedance 2.5 写提示词和 2.0 有什么不同?

手艺一样,结构不一样。4—15 秒时,一段话讲一个动作就够。到 30 秒,你必须把片子拆成阶段,每段一个主要动作 + 一个供下一段继承的结束状态,并且用命名的主体标记代替代词来锁身份——否则模型会把动作均分到全程,身份也会在后半段漂掉。参考素材同理:2.0 最多 9 张图 + 参考视频 + 参考音频,2.5 最多 50 份,素材一多,「逐份指派职责和排除项」就从可选变成必需。

Seedance 2.5 的提示词该写多长?

片子需要多长就多长,不要更长。本库的提示词从约 1700 字到 4600 字不等,最长的都是 30 秒那几条——不是因为话更啰嗦,而是 30 秒要钉死的状态更多。如果你的提示词长是因为把同一个动作换三种说法写了三遍,那就删掉:官方指南明确说过,重复描述同一个动作反而会劣化。

在 apimodels.app 上跑这些提示词

Seedance 2.5 已经上线,和另外 160+ 个图像、视频、语音、大模型共用一个 API、一个 key,按量计费、注册送 $0.10 试用额度。

这份提示词库在 GitHub 上是开源的

我们反推的提示词以 MIT 授权全文开放,作者原文则带署名与原帖链接索引。欢迎提 issue 补充新的提示词与成片。

在 GitHub 上查看

其它提示词库