bytedance/seedance-2.0/reference-to-video

图生视频

Seedance 2.0 Reference-to-Video API by ByteDance

bytedance/seedance-2.0/reference-to-video

Reference-to-video

Multimodal video generation from reference images, videos, and audio. Supports video editing and extension.

输入

正在加载参数配置...

输出

空闲

生成的视频将在这里显示

配置参数后点击运行开始生成

生成的 720p 视频每秒将按 $0.2419/秒 计费。每 1000 tokens 的费用为 $0.0112。tokens 数量按以下公式计算：（输出视频高度 × 输出视频宽度 ×（输入时长 + 输出时长）× 24）/ 1024。若提供视频输入，每 1000 tokens 的费率将降至 $0.00688。在视频输入加 720p 分辨率下，单价为每秒 $0.1486。

你可以继续：

Seedance 2.0 Kling v3 Vidu Wan2.7

参数

代码示例
import requests
import time

# Step 1: Start video generation
generate_url = "https://api.atlascloud.ai/api/v1/model/generateVideo"
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer $ATLASCLOUD_API_KEY"
}
data = {
    "model": "bytedance/seedance-2.0/reference-to-video",
    "prompt": "A beautiful sunset over the ocean with gentle waves",
    "width": 512,
    "height": 512,
    "duration": 3,
    "fps": 24,
}

generate_response = requests.post(generate_url, headers=headers, json=data)
generate_result = generate_response.json()
prediction_id = generate_result["data"]["id"]

# Step 2: Poll for result
poll_url = f"https://api.atlascloud.ai/api/v1/model/prediction/{prediction_id}"

def check_status():
    while True:
        response = requests.get(poll_url, headers={"Authorization": "Bearer $ATLASCLOUD_API_KEY"})
        result = response.json()

        if result["data"]["status"] in ["completed", "succeeded"]:
            print("Generated video:", result["data"]["outputs"][0])
            return result["data"]["outputs"][0]
        elif result["data"]["status"] == "failed":
            raise Exception(result["data"]["error"] or "Generation failed")
        else:
            # Still processing, wait 2 seconds
            time.sleep(2)

video_url = check_status()

安装

安装所需的依赖包。

bash

pip install requests

认证

所有 API 请求需要通过 API Key 进行认证。您可以在 Atlas Cloud 控制台获取 API Key。

bash

export ATLASCLOUD_API_KEY="your-api-key-here"

HTTP 请求头

python

import os

API_KEY = os.environ.get("ATLASCLOUD_API_KEY")
headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {API_KEY}"
}

保护好您的 API Key

切勿在客户端代码或公开仓库中暴露您的 API Key。请使用环境变量或后端代理。

提交请求

import requests

url = "https://api.atlascloud.ai/api/v1/model/generateVideo"
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer $ATLASCLOUD_API_KEY"
}
data = {
    "model": "your-model",
    "prompt": "A beautiful landscape"
}

response = requests.post(url, headers=headers, json=data)
print(response.json())

提交请求

提交一个异步生成请求。API 返回一个 prediction ID，您可以用它来检查状态和获取结果。

POST/api/v1/model/generateVideo

请求体

import requests

url = "https://api.atlascloud.ai/api/v1/model/generateVideo"
headers = {
    "Content-Type": "application/json",
    "Authorization": "Bearer $ATLASCLOUD_API_KEY"
}

data = {
    "model": "bytedance/seedance-2.0/reference-to-video",
    "input": {
        "prompt": "A beautiful sunset over the ocean with gentle waves"
    }
}

response = requests.post(url, headers=headers, json=data)
result = response.json()

print(f"Prediction ID: {result['id']}")
print(f"Status: {result['status']}")

响应

{
  "id": "pred_abc123",
  "status": "processing",
  "model": "model-name",
  "created_at": "2025-01-01T00:00:00Z"
}

检查状态

轮询 prediction 端点以检查请求的当前状态。

GET/api/v1/model/prediction/{prediction_id}

轮询示例

import requests
import time

prediction_id = "pred_abc123"
url = f"https://api.atlascloud.ai/api/v1/model/prediction/{prediction_id}"
headers = { "Authorization": "Bearer $ATLASCLOUD_API_KEY" }

while True:
    response = requests.get(url, headers=headers)
    result = response.json()
    status = result["data"]["status"]
    print(f"Status: {status}")

    if status in ["completed", "succeeded"]:
        output_url = result["data"]["outputs"][0]
        print(f"Output URL: {output_url}")
        break
    elif status == "failed":
        print(f"Error: {result['data'].get('error', 'Unknown')}")
        break

    time.sleep(3)

状态值

processing请求仍在处理中。

completed生成完成，输出可用。

succeeded生成成功，输出可用。

failed生成失败，请检查 error 字段。

完成响应

{
  "data": {
    "id": "pred_abc123",
    "status": "completed",
    "outputs": [
      "https://storage.atlascloud.ai/outputs/result.mp4"
    ],
    "metrics": {
      "predict_time": 45.2
    },
    "created_at": "2025-01-01T00:00:00Z",
    "completed_at": "2025-01-01T00:00:10Z"
  }
}

上传文件

将文件上传到 Atlas Cloud 存储，获取可在 API 请求中使用的 URL。使用 multipart/form-data 上传。

POST/api/v1/model/uploadMedia

上传示例

import requests

url = "https://api.atlascloud.ai/api/v1/model/uploadMedia"
headers = { "Authorization": "Bearer $ATLASCLOUD_API_KEY" }

with open("image.png", "rb") as f:
    files = {"file": ("image.png", f, "image/png")}
    response = requests.post(url, headers=headers, files=files)

result = response.json()
download_url = result["data"]["download_url"]
print(f"File URL: {download_url}")

响应

{
  "data": {
    "download_url": "https://storage.atlascloud.ai/uploads/abc123/image.png",
    "file_name": "image.png",
    "content_type": "image/png",
    "size": 1024000
  }
}

Input Schema

以下参数在请求体中被接受。

总计: 0必填: 0可选: 0

暂无可用参数。

请求体示例

json

{
  "model": "bytedance/seedance-2.0/reference-to-video"
}

Output Schema

API 返回包含生成输出 URL 的 prediction 响应。

idstringrequired

Unique identifier for the prediction.

statusstringrequired

Current status of the prediction.

processingcompletedsucceededfailed

modelstringrequired

The model used for generation.

outputsarray[string]

Array of output URLs. Available when status is "completed".

errorstring

Error message if status is "failed".

metricsobject

Performance metrics.

predict_timenumber

Time taken for video generation in seconds.

created_atstringrequired

ISO 8601 timestamp when the prediction was created.

Format: date-time

completed_atstring

ISO 8601 timestamp when the prediction was completed.

Format: date-time

响应示例

json

{
  "id": "pred_abc123",
  "status": "completed",
  "model": "model-name",
  "outputs": [
    "https://storage.atlascloud.ai/outputs/result.mp4"
  ],
  "metrics": {
    "predict_time": 45.2
  },
  "created_at": "2025-01-01T00:00:00Z",
  "completed_at": "2025-01-01T00:00:10Z"
}

Atlas Cloud Skills

Atlas Cloud Skills 将 300+ AI 模型直接集成到您的 AI 编程助手中。一条命令安装，即可用自然语言生成图像、视频和与 LLM 对话。

支持的客户端

Claude Code

OpenAI Codex

Gemini CLI

Cursor

Windsurf

VS Code

Trae

GitHub Copilot

Cline

Roo Code

Amp

Goose

Replit

40+ 支持的客户端

安装

bash

npx skills add AtlasCloudAI/atlas-cloud-skills

设置 API Key

从 Atlas Cloud 控制台获取 API Key，并将其设置为环境变量。

bash

export ATLASCLOUD_API_KEY="your-api-key-here"

功能

安装后，您可以在 AI 助手中使用自然语言访问所有 Atlas Cloud 模型。

图像生成使用 Nano Banana 2、Z-Image 等模型生成图像。

视频创作使用 Kling、Vidu、Veo 等模型从文本或图像创建视频。

LLM 对话与 Qwen、DeepSeek 等大语言模型对话。

媒体上传上传本地文件用于图像编辑和图生视频工作流。

MCP Server

Atlas Cloud MCP Server 通过 Model Context Protocol 将您的 IDE 与 300+ AI 模型连接。支持任何兼容 MCP 的客户端。

支持的客户端

Cursor

VS Code

Windsurf

Claude Code

OpenAI Codex

Gemini CLI

Cline

Roo Code

100+ 支持的客户端

安装

bash

npx -y atlascloud-mcp

配置

将以下配置添加到您的 IDE 的 MCP 设置文件中。

json

{
  "mcpServers": {
    "atlascloud": {
      "command": "npx",
      "args": [
        "-y",
        "atlascloud-mcp"
      ],
      "env": {
        "ATLASCLOUD_API_KEY": "your-api-key-here"
      }
    }
  }
}

可用工具

atlas_generate_image从文本提示生成图像。

atlas_generate_video从文本或图像创建视频。

atlas_chat与大语言模型对话。

atlas_list_models浏览 300+ 可用 AI 模型。

atlas_quick_generate一步式内容创建，自动选择最佳模型。

atlas_upload_media上传本地文件用于 API 工作流。

了解更多

github.com/AtlasCloudAI/mcp-server

API Schema

Schema 不可用

暂无可用示例

请登录以查看请求历史

您需要登录才能访问模型请求历史记录。

1. Introduction

Seedance 2.0 is a state-of-the-art multimodal generative AI model designed for synchronized video and audio content creation. Developed by ByteDance and integrated into the CapCut/Dreamina platform as of March 2026, this model family advances the field of generative multimedia by combining sophisticated diffusion transformer architectures with physics-informed world modeling for realistic motion and spatial consistency.

Seedance 2.0’s significance lies in its Dual-Branch Diffusion Transformer (DB-DiT) architecture that jointly processes video and audio streams, enabling phoneme-level lip synchronization across multiple languages. Compared to previous iterations, it achieves substantially higher output usability rates and faster generation speeds. The two variants target different workloads: Seedance 2.0 delivers high-fidelity, cinematic-quality renders with enhanced lighting and texture detail, while Seedance 2.0 Fast provides a cost-effective, accelerated pipeline optimized for high throughput and rapid prototyping.

2. Key Features & Innovations

Dual-Branch Diffusion Transformer Architecture: Seedance 2.0 integrates separate yet synchronized diffusion branches for video and audio, enabling tight coupling between visual motion and sound generation. This architecture improves motion realism and audio-visual coherence beyond previous generative models.
World Model with Physics Simulation: The model incorporates a physics-based world modeling approach that simulates realistic object motion and spatial consistency over time. This leads to naturalistic dynamics and stable scene composition across generated video sequences.
Rich Multimodal Input Support: Seedance 2.0 accepts diverse input formats including text prompts, up to 9 images, and up to 3 video or audio clips of 15 seconds each. This flexibility allows nuanced content creation workflows combining static, dynamic, and auditory cues.
Phoneme-Level Lip Synchronization: The native audio generation pipeline supports lip-sync at the phoneme granularity in 8+ languages, ensuring high fidelity mouth movements closely match generated speech or singing.
High Usability and Efficiency: The model achieves an estimated 90% usable output rate compared to an industry average of approximately 20%, reducing post-processing overhead. Additionally, it delivers a 30% inference speed advantage over predecessor systems.
API Variants for Different Use Cases: The Seedance 2.0 endpoint is geared toward high fidelity and cinematic visual effects suitable for final production, while the Seedance 2.0 Fast variant offers roughly 3 times faster generation and approximately 91% cost savings at $0.022 per second of output, ideal for rapid iteration and volume workflows.

3. Model Architecture & Technical Details

Seedance 2.0 is built around the Dual-Branch Diffusion Transformer (DB-DiT), which separately processes video and audio streams via transformer-based denoising diffusion models while synchronizing generation steps to enforce audio-visual alignment. The system leverages a World Model that integrates physics simulation modules, enabling consistent spatial and temporal object behaviors within video sequences.

Training was conducted in multiple stages on large-scale, diverse datasets spanning images, videos, text captions, and audio recordings across multiple languages. Initial large-scale pre-training utilized resolutions spanning from 720p to 1080p, followed by supervised fine-tuning (SFT) to improve text and visual prompt conditioning fidelity. Reinforcement Learning with Human Feedback (RLHF) optimized multi-dimensional reward models that simultaneously assess aesthetics, motion coherence, and audio-visual synchronization quality.

The training pipeline supports multiple aspect ratios including 9:16, 16:9, 1:1, and 4:3, and target output lengths from 4 to 60 seconds. Specialized modules enable the @ reference system for fine-grained control of creative elements based on provided input assets.

4. Performance Highlights

Seedance 2.0 was benchmarked on the comprehensive SeedVideoBench-2.0 suite, which evaluates generative video models across over 50 image-based and 24 video-based benchmarks covering diverse content domains and multi-modal tasks.

Rank	Model	Developer	Score/Metric	Release Date
1	Kling 3.0	External	Competitive	2025
2	Sora 2	External	Competitive	2025
3	Seedance 2.0	ByteDance	High audiovisual sync, motion realism	2026
4	Veo 3.1	External	Strong baseline	2025

Seedance 2.0 matches or exceeds these contemporary models in synchronized video-audio generation, demonstrating especially strong performance in phoneme-level lip synchronization and motion naturalism thanks to the World Model component. Its 30% speed improvement and 90% output usability rate reflect notable efficiency advancements.

5. Intended Use & Applications

Social Media Content Creation: Efficiently generate engaging short videos with synchronized audio and visually rich effects, tailored for platforms like TikTok and Instagram.
E-commerce Product Videos: Automatically produce dynamic product showcases combining text, image, and video inputs with realistic motion and sound to enhance online shopping experiences.
Marketing Campaigns: Craft high-quality cinematic promotional content that integrates brand assets via the @ reference system for tailored storytelling and audience engagement.
Music Videos: Generate synchronized visuals with phoneme-accurate lip-syncing for multilingual vocal tracks to support artist and record label promotional needs.
Short Narrative Films: Create compelling narrative-driven video clips with coherent motion and spatial consistency, supporting indie filmmakers and content creators.
Fashion and Luxury Showcases: Produce visually detailed and aesthetic presentations incorporating texture and lighting refinements for high-end brand communications.