
Vidu API 为开发者提供对 Shengshu AI 和清华大学视频模型的访问权限,采用 U-ViT 架构构建,该架构统一了 Diffusion 和 Transformer 设计。输入 1-4 张参考图像,Vidu 可保持主体在各镜头间的一致性,具有智能摄像头切换和流畅的电影级运动。Atlas Cloud 为整个产品线提供 Day-0 模型访问权限和一个 OpenAI 兼容密钥。立即开始构建。
Atlas Cloud 为您提供最新的行业领先创意模型。
Scan durations, resolutions, reference image limits, and audio options across all 25 Vidu API video generation endpoints to match the right model to your workload.
| Modality | Description |
|---|---|
| Vidu Q3-Pro T2V API (Text to Video) | Describe a scene in plain text and Q3-Pro renders it as video up to 16 seconds long, with five aspect ratios plus general and anime styles. Super resolution output reaches 1440p, and audio with background music generates in the same pass. It suits cinematic ads and narrative shots where fidelity matters most. |
| Vidu Q3-Turbo T2V API (Text to Video) | Need volume more than polish? This turbo tier converts text prompts into clips of 1 to 16 seconds below Pro tier pricing, while keeping the same 1440p super resolution ceiling, style presets, and optional audio track. Teams use it for rapid storyboarding and testing creative directions. |
| Vidu Q3-Pro I2V API (Image to Video) | Upload one still image, describe the motion you want, and the model animates it with smooth, cinematic movement for up to 16 seconds. Output scales from 540p to 1440p super resolution with generated audio and background music. Product showcases and photo animation pipelines are its natural home. |
| Vidu Q3-Turbo I2V API (Image to Video) | When turnaround matters, the lowest priced Q3 tier brings photos to life from a single reference image and a motion prompt. It reaches 1440p super resolution and attaches audio automatically unless disabled. Social clips and bulk catalog animation run economically on it. |
| Vidu Q3-Pro SE2V API (Start-End to Video) | Give it a first frame and a last frame, and Q3-Pro interpolates a coherent transition between them lasting up to 16 seconds. The two frames must keep their aspect ratios within 0.8 to 1.25 of each other, and audio plus background music are generated on request. Storyboards, morph effects, and scene bridges all fit here. |
| Vidu Q3-Turbo SE2V API (Start-End to Video) | Fast frame bridging is the turbo variant's specialty. It connects a start image and an end image into one continuous shot, supports 540p through 1440p super resolution output, and toggles audio with a single flag. Iterate transitions quickly here before committing to a Pro render. |
| Vidu Q3 R2V API (Reference to Video) | Consistency across camera angles sets this endpoint apart. Feed it 1 to 4 reference images and it keeps subjects stable through intelligent camera switching, producing clips of 3 to 16 seconds at up to 1440p super resolution with audio. Character driven ads and episodic content benefit most. |
| Vidu Q3-Mix R2V API (Reference to Video) | Blend up to 4 reference images into one scene with intelligent transitions and smooth dynamic effects. The Q3-Mix endpoint holds subject identity while settings shift, renders clips of 1 to 16 seconds in five aspect ratios, and adds generated audio by default. Multi-shot brand stories and mashup creative are built on it. |
| Vidu Q2 T2V API (Text to Video) | Text goes in, and a finished clip of 1 to 8 seconds comes out at up to 1080p. General and anime styles, five aspect ratios, and automatic audio with background music are all controlled through simple parameters. It makes a solid default for everyday prompt to video work. |
| Vidu Q2-Pro I2V API (Image to Video) | Animating a single photo with prompts of up to 5000 characters, this Pro endpoint outputs 540p to 1080p clips running as long as 10 seconds. Movement amplitude can be pinned to small, medium, or large for precise motion control. Explainers and marketing stills convert well here. |
| Vidu Q2-Pro-Fast I2V API (Image to Video) | The Fast variant keeps 720p and 1080p output and the same price as the standard Pro tier while shortening generation queues for image animation jobs. Durations stretch from 1 to 10 seconds with optional audio and background music. Pick it when pipelines demand quick iteration. |
| Vidu Q2-Turbo I2V API (Image to Video) | At the lowest price in the entire Vidu family, this endpoint animates one image into clips of up to 10 seconds at 540p, 720p, or 1080p. Audio and background music generate automatically unless switched off. High volume UGC tools and batch jobs are where it earns its keep. |
| Vidu Q2-Pro SE2V API (Start-End to Video) | Two keyframes become one fluid shot. Q2-Pro fills the gap between a start image and an end image over 1 to 8 seconds at up to 1080p, with generated audio and background music available. Keep the frame aspect ratios within 0.8 to 1.25 of each other for a valid request. |
| Vidu Q2-Pro-Fast SE2V API (Start-End to Video) | Prefer speed for frame interpolation? This variant renders start to end transitions at 720p or 1080p in clips up to 8 seconds, with audio and background music toggles included. It shares standard Pro pricing while tightening iteration loops for editors. |
| Vidu Q2-Turbo SE2V API (Start-End to Video) | The most economical start-end option in the family lives here. Supply first and last frames and receive a 1 to 8 second clip at up to 1080p, with optional audio in the same call. Prototyping transitions before a Pro render is a common workflow. |
| Vidu Q2 R2V API (Reference to Video) | Up to 7 distinct subjects, each defined by 1 to 3 reference images, can appear together in a single generated scene. Clips run 1 to 10 seconds at up to 1080p, and an audio type switch isolates speech or sound effects. Multi-character storytelling is the headline use case. |
| Vidu Q2-Pro R2V API (Reference to Video) | Casting several consistent characters into premium footage is what the Pro reference tier does best. It accepts up to 7 subjects with 3 images each, prompts up to 1500 characters, and outputs clips of up to 10 seconds at 1080p with speech only or sound effect only audio modes. Serialized content and virtual character work fit naturally. |
| Vidu Q1 T2V API (Text to Video) | The first generation text endpoint outputs fixed 5 second clips locked at 1080p, with 16:9, 9:16, and 1:1 aspect ratios plus general and anime styles. Audio and background music remain switchable. Choose it when predictable, uniform clip specs simplify a pipeline. |
| Vidu Q1 I2V API (Image to Video) | Simplicity defines this endpoint: one image, one prompt of up to 5000 characters, one 5 second 1080p clip. Movement amplitude and audio toggles are the only knobs to tune. Integrations that standardized on Q1 output keep running unchanged. |
| Vidu Q1 SE2V API (Start-End to Video) | Interpolating between two uploaded frames, Q1 produces a 5 second 1080p transition with adjustable motion amplitude and optional audio plus background music. Frame pairs must keep their aspect ratios within 0.8 to 1.25 of each other. It remains a stable choice for fixed length transition effects. |
| Vidu Q1 R2V API (Reference to Video) | Structured casting is the pattern here: up to 7 subjects, each backed by 1 to 3 images, render into a fixed 5 second 1080p clip. Audio narrows to speech only or sound effects only when needed. Multi-character scenes with locked output specs suit it well. |
| Vidu 2.0 I2V API (Image to Video) | Would a short silent clip do the job? Vidu 2.0 turns one image into a 4 or 8 second video with adjustable movement amplitude and seed control for reproducible results. Thumbnails, previews, and lightweight animation tasks are typical fits. |
| Vidu 2.0 SE2V API (Start-End to Video) | Exactly two images, a prompt of up to 1500 characters, and a choice of 4 or 8 second duration define this compact endpoint. Motion intensity runs from small to large under one movement amplitude parameter. It handles quick morphs and simple frame pair transitions. |
| Vidu 2.0 R2V API (Reference to Video) | From 1 to 3 reference images, this endpoint composes videos in 16:9, 9:16, or 1:1 while preserving the referenced subjects. Seed and movement amplitude round out a deliberately small parameter set. Consistent character clips for social formats are the common case. |
| Vidu R2V Q1 API (Multi-Image Reference to Video) | Broader flat input distinguishes this reference endpoint, which takes a single list of 1 to 7 images in 16:9, 9:16, or 1:1. Prompts cap at 1500 characters and each image can reach 50MB. It works well when subjects arrive as loose image sets rather than structured casts. |
从四个参考图像一致性到原生音频和16秒的1080p片段,Atlas Cloud上的Vidu API将Shengshu的整个Q系列放在一个按需付费的端点后面。
Vidu API接受1到4张参考图像,并在每一帧中保持角色、产品和场景的视觉一致性。这种可靠性使得序列化讲述和品牌安全的产品视频在规模上变得实际可行。
上传开始帧和结束帧,模型自动构建它们之间的平滑过渡。动作设计师使用这个功能进行变形、揭示和精确计时的场景变更。
为什么要渲染无声片段并稍后为其配分?每个Q3端点都可以在同一次处理中生成同步的音效和背景音乐,所以对话场景和广告都能以准备好发布的状态到达。
如果场景需要喘息的空间,运行最多16秒的片段,分辨率从540p到1080p,并使用超分辨率放大到1440p。更少的缝合镜头意味着故事更流畅,后期制作更轻。
运动振幅设置将输出从细微的电影漂移转换为大的戏剧性动作,而文本到视频端点在一般和动画风格之间切换。导演无需重写提示即可获得镜头级控制。
从Q1到Q3的每一代Vidu模型都在OpenAI兼容密钥后面运行,在Atlas Cloud上具有透明的按需付费定价。在Turbo上起草,然后在Pro上重新运行相同的调用以进行最终渲染。
The same prompt, generated by Vidu and other leading video models
一名特技驾驶员独自驾驶一辆哑光黑色肌肉车在雨中穿过霓虹灯照亮的夜市。以低角度追踪镜头开场,滑过湿润的沥青,汽车绕过一个角落时,火花从护栏飞出。快速摇镜到摊贩急忙收起摊位,纸灯笼疯狂摇晃。切换到无人机镜头螺旋上升,汽车在两辆公交车之间仅相隔几英寸处通过,随后硬切到驾驶舱内部:手紧抓方向盘,眼睛瞥向一面充满闪烁警灯的镜子。以慢动作180度转向一条冒烟的小巷结尾,引擎轰鸣和雨声充满了配乐。光影逼真、电影般的灯光、10秒、16:9宽高比。
Vidu Q3
Seedance 2.0
Vidu Q2 Pro
动画风格:年轻女剑士在破碎的浮空神社上方面对巨大风暴龙。以快速推移镜头聚焦她的眼睛开场,闪电划破天空,随后快速转向跟随她冲过崩塌的石桥,碎石在她脚边扬起。当她跃起时摄像机环绕,她的剑刃被蓝色火焰点燃,而龙的颌骨在她一帧之前站立的位置坠落。拉回到史诗般的宽镜头,当她的一击将闪电劈成两半,冲击波在云层中荡起涟漪。清晰的二维单线填充阴影、动态速度线、音轨中的雷鸣和高涨的合唱团。8秒、16:9宽高比。
Vidu Q3 Pro
Seedance 2.0
Vidu Q2 Pro
从角色一致的品牌视频到起始和结束帧的运动设计,Vidu API 在 Atlas Cloud 上的单个按使用量付费端点后整合了 text-to-video、image-to-video 和 reference-to-video 工作流。
将 1 到 4 张参考图像输入到 Vidu Q3-Mix 或 Q3 reference-to-video,对象在场景转换中保持一致。短剧团队和 IP 创作者在每一集保持相同的面部特征。
产品发布和品牌广告通过 Vidu API 呈现高达 1080p 的平滑动态效果和可选音频而栩栩如生。广告公司无需预订任何拍摄日期就能迭代广告概念。
需要在两个关键帧之间进行精确过渡?Q3-Pro 和 Q2 start-end-to-video 模型在第一帧和最后一帧之间插值平滑运动,支持标志显示、场景变形和产品转换。
作者通过 Vidu API text-to-video 端点将文本提示转换为具有多种视觉风格的 1080p 素材。可选的音频生成在一次传递中为故事版和动画预览提供完成效果。
当静止产品照片或插图需要动作时,Q1 至 Q3-Turbo 的 image-to-video 模型使用平滑、自然的运动为其制作动画。电子商务团队和社交媒体编辑每天发布令人瞩目的视频片段。
Vidu 的 U-ViT 架构为长形式、高度一致的输出而构建,是动画设计和连续剧的自然适配。工作室通过 API 调用而不是手动关键帧来制作故事板、动画化和修改整个序列。
并排比较视频长度、分辨率上限、音频输出和秒级速率,你能清楚地看到 Vidu API 在视频处理管道中的每一块钱能为你做什么
| 模型 | 输入模式 | 最大视频长度 | 最高分辨率 | 原生音频 |
|---|---|---|---|---|
| Vidu Q3-Mix | 文本加 1 到 4 张参考图像 | 16s | 1080p 原生,1440p 超分辨率 | √ |
| Vidu Q3 (Pro & Turbo) | 文本、图像、开始-结束帧、参考图像 | 16s | 最高 1080p,1440p 上采样 | √ |
| Vidu Q2 (Pro & Turbo) | 文本、图像、开始-结束帧或参考 | 10s | 1080p | √ |
| Kling 2.5 Turbo | 文本或单张图像加提示词 | 10s | 1080p(Pro 级别) | - |
| Google Veo 3.1 | 文本、图像、最多 3 张参考图像 | 每个视频 8 秒,可扩展 | 1080p,预览中为 4K | √ |
| OpenAI Sora 2 | 文本提示 | 12s | 720p(Sora 2 Pro 上为 1080p) | √ |
几分钟即可上手 — 按照以下简单步骤,通过 Atlas Cloud 平台集成和部署模型。
在 atlascloud.ai 注册并完成验证。新用户可获得免费额度,用于探索平台和测试模型。
将先进的 Vidu 模型与 Atlas Cloud 的 GPU 加速平台相结合,提供无与伦比的性能、可扩展性和开发体验。
低延迟:
GPU 优化推理,实现实时响应。
统一 API:
一次集成,畅用 Vidu、GPT、Gemini 和 DeepSeek。
透明定价:
按 Token 计费,支持 Serverless 模式。
开发者体验:
SDK、数据分析、微调工具和模板一应俱全。
可靠性:
99.99% 可用性、RBAC 权限控制、合规日志。
安全与合规:
SOC 2 Type II 认证、HIPAA 合规、美国数据主权。
按生成的视频秒数计费,无需订阅。Q2-Turbo端点从每秒$0.026开始,Q3-Turbo每秒$0.034,Q3-Pro和Q3参考转视频每秒$0.042,Q3-Mix参考转视频每秒$0.106。在透明的按调用定价模式下,您只需为生成的内容付费。
Q3端点支持长达16秒的视频片段,分辨率选项为540p、720p和1080p,以及高达1440p的超分辨率输出。纵横比包括16:9、9:16、4:3、3:4和1:1,因此相同的端点可以针对横向、纵向或正方形格式。
是的。Q3模型包括一个默认启用的音频切换,文本转视频端点添加了单独的背景音乐开关。当需要用于您自己的声音设计的无声素材时,请关闭两者。
上传一到四张角色、产品或场景的参考图像,模型在生成新的运动和相机角度时保持该主体的身份。Q3参考转视频添加了智能摄像机切换,在多个摄像机位置上具有更好的一致性,适合于情节式和多镜头叙述。
是的。文本转视频端点公开了一个具有常规和动画选项的样式参数,该系列以风格化的角色动画而闻名。将动画样式与参考图像配对,以在整个系列中保持相同的角色。
创建Atlas Cloud API密钥,选择Vidu端点,并使用您的提示和参数发送REST请求。每个模型都有一个playground和已发布的参数架构,因此您可以在编写代码之前验证设置。定价是按使用次数计费,新的Vidu版本在第0天获得访问权限。今天开始构建。
It does. Text-to-video endpoints expose a style parameter with general and anime options, and the family is known for stylized character animation. Pair the anime style with reference images to keep the same character across an entire series.
Create an Atlas Cloud API key, pick a Vidu endpoint, and send a REST request with your prompt and parameters. Every model has a playground and a published parameter schema, so you can validate settings before writing code. Pricing is pay-as-you-go per call, and new Vidu releases arrive with Day-0 access. Start building today.
Join the Discord community for the latest model updates, prompts, and support.