NOW LIVESeedance 2.5 is now live on I2V.aiTry it now
i2v.ai
  • Studio
  • Models
  • Seedance 2.5
  • Pricing
模型 / 视频

字节跳动

Seedance 2.5

seedance2-5

支持文字、图片、关键帧与多模态参考输入的可控视频生成模型。

文生视频图生视频首尾帧控制多模态参考视频编辑480p / 720p
通过 API 调用
在线体验价格API案例模型介绍常见问题

Detailed descriptions of scenes, actions, and styles will result in better generation quality

4 seconds5 seconds30 seconds

Loading...

Join Discord community
Content policy notice
Each provider applies its own content policy. Some models may reject unsafe requests or certain real-person reference inputs. If a task is rejected, review the model guidance and try permitted source material.

按模式与输出规格计价

Studio 与 API 共用同一积分余额。以下美元估算基于 $9.90 Mini Pack;购买更大积分包或使用优惠后,实际单次成本可能更低。

模式配置积分预估美元价格
文字或图片输入480p · 5s · 不含音频54$0.69
文字或图片输入480p · 5s · 含音频54$0.69
视频输入示例480p · 5s 输出 · 1 × 5s input · 不含音频33$0.42
视频输入示例480p · 5s 输出 · 1 × 5s input · 含音频33$0.42
文字或图片输入720p · 5s · 不含音频116$1.47
文字或图片输入720p · 5s · 含音频116$1.47
视频输入示例720p · 5s 输出 · 1 × 5s input · 不含音频70$0.89
视频输入示例720p · 5s 输出 · 1 × 5s input · 含音频70$0.89
生成模式
文生视频、图生视频、首尾帧、参考图像、多模态参考、视频编辑
时长
4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30 秒
分辨率
480p, 720p
画面比例
16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive
音频
不支持直接生成音频

通过 Seedance 2.5 构建产品

创建任务后保留返回的 task ID,并通过 recordInfo 轮询结果。页面会展示准确模型价格,方便你在运行前估算产品成本。

获取 API 密钥
POST /api/v1/jobs/createTask
{
  "model": "seedance2-5",
  "inputs": {
    "prompt": "A cinematic tracking shot through a futuristic city at sunset",
    "duration": "5s",
    "aspectRatio": "16:9",
    "resolution": "480p"
  }
}

用镜头语言控制一段场景

30 秒连续叙事镜头

查看模型如何保持镜头运动、材质、光线与角色动作。展开任意案例即可阅读完整英文提示词。

参考素材Images × 8
Reference 1 used for the narrative continuity example.
1/8
生成结果

多参考图生视频 - 连续镜头运动

八张参考图串成穿越六个相连房间的一次连续行走,角色与建筑节奏始终保持一致。

1 / 6

提示词:One continuous take. The camera steadily follows a person in a black coat (reference @image1) walking from left to right through six connected rooms, each with a different colour palette and atmosphere. Every room shares the same structure: white walls, pale herringbone wood flooring, French double-height windows, and sheer white curtains, referencing @image2. The exterior view and interior mood change completely in every room. The subject walks at a constant speed through every open doorway. 0-5s: the first room is an American-comic fight. The subject enters and fights a character from @image3; the character is defeated. 5-10s: the second room feels warm and is made of felt. Outside is a sunflower field from @image4. Warm orange light fills the room while a painter paints sunflowers from @image5. The subject becomes felt as they enter. 10-15s: the third room is sad, rendered as black-and-white comic stop motion. Rain falls outside; the light is cool, grey, and low. A person sits alone on the floor, head down and knees hugged, beside a phone showing an unanswered call. When the subject enters, the lights go out, then immediately come back on in colour as flowers instantly grow across the room. 15-20s: the fourth room is joyful and submerged in the sea, referencing @image6. As the subject swims in, they become transparent; coral reefs and schools of fish surround them. 20-25s: the fifth room is surprising. Outside is a fireworks-filled night sky from @image7, and coloured flashes reflect through the room. A crowd surrounds the subject and pulls them into the celebration. 25-30s: the subject reaches a blank final room, stands in the centre, and snaps their fingers with a matching snap sound. The image cuts to black and the word "seedance" appears in the centre, referencing @image8. Cinematic visual language, premium fashion-film treatment. Let the exterior view determine the light in each room for strong emotional contrast. No other on-screen text.

展开

完整提示词

One continuous take. The camera steadily follows a person in a black coat (reference @image1) walking from left to right through six connected rooms, each with a different colour palette and atmosphere. Every room shares the same structure: white walls, pale herringbone wood flooring, French double-height windows, and sheer white curtains, referencing @image2. The exterior view and interior mood change completely in every room. The subject walks at a constant speed through every open doorway. 0-5s: the first room is an American-comic fight. The subject enters and fights a character from @image3; the character is defeated. 5-10s: the second room feels warm and is made of felt. Outside is a sunflower field from @image4. Warm orange light fills the room while a painter paints sunflowers from @image5. The subject becomes felt as they enter. 10-15s: the third room is sad, rendered as black-and-white comic stop motion. Rain falls outside; the light is cool, grey, and low. A person sits alone on the floor, head down and knees hugged, beside a phone showing an unanswered call. When the subject enters, the lights go out, then immediately come back on in colour as flowers instantly grow across the room. 15-20s: the fourth room is joyful and submerged in the sea, referencing @image6. As the subject swims in, they become transparent; coral reefs and schools of fish surround them. 20-25s: the fifth room is surprising. Outside is a fireworks-filled night sky from @image7, and coloured flashes reflect through the room. A crowd surrounds the subject and pulls them into the celebration. 25-30s: the subject reaches a blank final room, stands in the centre, and snaps their fingers with a matching snap sound. The image cuts to black and the word "seedance" appears in the centre, referencing @image8. Cinematic visual language, premium fashion-film treatment. Let the exterior view determine the light in each room for strong emotional contrast. No other on-screen text.

为什么有效

先固定重复的空间结构和行走节奏,再只在叙事需要的位置改变每个房间。

预期输出

一段连贯的 30 秒运动镜头,在变化的世界中保持同一位角色。

提示

  • 在每个场景段落中使用同一角色参考。
  • 先描述不变元素,再描述每个场景的变化。
  • 需要一镜到底时,明确写出镜头路径。

什么是 Seedance 2.5?

Seedance 2.5 适合需要明确镜头调度、角色连续性与参考素材控制的视频创作。你可以先在 Studio 中验证效果,再通过同一模型 ID 接入 API。

从简单的文字分镜到多图、多视频与音频参考,页面会在提交前展示可用输入、分辨率、时长与积分消耗。

Seedance 2.5 能做什么

连续叙事镜头

通过清晰的时间段、镜头路径和不变元素描述,让长镜头在不同场景之间保持角色与空间连续。

多模态参考控制

将图片、视频和音频参考分配给明确角色,帮助模型稳定人物、环境、动作和节奏。

可控视频编辑

先说明源视频中必须保留的构图与动作,再描述单一替换或移除目标,获得更稳定的编辑结果。

Seedance 2.5 的适用场景

短片与广告分镜

把镜头、动作、场景质感和情绪节奏写成制作简报,快速验证叙事视频或广告片段。

角色与产品连续性

使用参考图或源视频固定人物、产品和环境,再用明确的镜头语言推进场景。

视频特效与内容编辑

针对单个道具、人物或场景进行替换、移除和材质调整,同时尽量保留原始动作与构图。

何时选择 Seedance 2.5

当你的项目依赖镜头控制、参考素材和复杂连续动作时,可选择 Seedance 2.5。先比较 480p 与 720p 的时长和积分,再用 Studio 验证最终参数后接入 API。

Seedance 2.5 常见问题

Seedance 2.5 支持哪些输入方式?+

可用能力以当前表单为准,包括文字、图片、首尾帧、参考素材和兼容的视频编辑输入。不同模式可用的时长与参数会在提交前显示。

Studio 与 API 是否使用同一套积分?+

是。Studio 与 API 使用同一账户、同一积分余额和同一模型 ID。任务失败时会按照平台的失败处理规则释放或返还积分。

如何让长镜头更稳定?+

先写清角色、空间结构、镜头路径和需要保持不变的元素,再按时间段描述场景变化。避免在同一条指令中加入彼此冲突的镜头要求。

继续探索

相关模型

查看全部模型

Seedance 2.0 Mini

发布日期: June 2026

Text, image, and multi-reference video generation with audio

含音频 · 480p · 5s

$0.23

视频输入 · 480p · 5s

$0.28

含音频 · 720p · 5s

$0.48

视频输入 · 720p · 5s

$0.58

打开模型

Seedance 2.0 Fast

发布日期: April 2026

Text, image, reference, and video-to-video generation

标准 · 480p · 5s

$0.37

视频输入 · 480p · 5s

$0.43

标准 · 720p · 5s

$0.77

视频输入 · 720p · 5s

$0.91

打开模型

Seedance 2.0

发布日期: February 2026

Text, image, reference, and video-to-video generation

标准 · 480p · 5s

$0.46

视频输入 · 480p · 5s

$0.56

标准 · 720p · 5s

$0.96

视频输入 · 720p · 5s

$1.18

打开模型
Resources
  • Blog
  • Studio
  • Scenes
  • Works
  • Prompts
  • Image to Prompt
  • Batch Image to Prompt
Company & Legal
  • About
  • Contact
  • Privacy Policy
  • Terms of Service
  • Refund Policy
Image Models
  • Qwen Image 2
  • Seedream 4.5
  • Seedream 5.0
  • Nano Banana Pro
  • Nano Banana Flash
  • Nano Banana 2
Video Models
  • Google Veo 3.1
  • Google Veo 3.1 Lite
  • Google Veo 3.1 Pro
  • Seedance 1.5 Pro
  • Seedance Fast
  • Seedance Quality
  • Seedance 2.0
  • Kling v3.0
  • Kling v3.0 Pro
i2v.ai

I2V AI — image and video creation, with API access

DiscordEmail

This website is an independent third-party platform. We are not affiliated with, endorsed by, or officially representing any AI model providers referenced on this site. All trademarks and brand names belong to their respective owners.

© 2026 I2V AI All Rights Reserved.

Privacy Policyprivacy@i2v.aisupport@i2v.ai