Seedance 2.5 Spicy API

Seedance 2.5 Spicy is a multimodal video generation model developed by ByteDance. As the Spicy global version, it is tailored for a global user base with a global content policy. It supports multimodal inputs including reference images, reference videos, and reference audio, enabling single video generation up to 30 seconds. The model also supports video editing, video extension, and the web_search tool, with output resolutions of 480p or 720p and durations ranging from 4 to 30 seconds, or -1 for automatic selection.

On the iCreat platform, Seedance 2.5 Spicy is accessed via an asynchronous two-step API. The submission endpoint is https://api.icreat.ai/v1/task/submit/bytedance/seedance-2-5-global, and results are polled via https://api.icreat.ai/v1/task/result using the same task_id. Pricing is based on resolution and whether a reference video is included, ranging from $0.084 to $0.606 per second.

Model Positioning

Seedance 2.5 Spicy is positioned as a high-end, long-duration multimodal video generation model. Compared to the standard Text-to-Video endpoint, the Spicy global version provides a unified multimodal endpoint accepting text, image, video, and audio inputs. It is designed for creators and developers requiring complex video generation, editing, and extension capabilities with global content accessibility.

Core Capabilities

Multimodal Input Support

Seedance 2.5 Spicy accepts text prompts, reference images, reference videos, and reference audio. Users can define specific roles for media inputs such as reference_image, reference_video, reference_audio, first_frame, and last_frame to precisely control the generated content.

Long-Duration Video Generation

The model supports generating videos up to 30 seconds in a single request. Users can specify an exact duration between 4 and 30 seconds or set the duration to -1 to let the model automatically determine the optimal length.

Video Editing and Extension

Beyond standard text-to-video generation, the model natively supports video editing and video extension workflows, allowing users to modify existing footage or seamlessly lengthen previously generated clips.

Web Search Tool Integration

The model supports a web_search tool, enabling it to leverage online information during the generation process to enhance the accuracy and relevance of the output based on real-world data.

Multilingual Prompt Support

The model understands prompts in Chinese, English, Japanese, Indonesian, Spanish, and Portuguese. Chinese prompts support up to 2,000 characters, while English prompts support up to 2,000 words.

Pricing

Total Cost = Unit Price × (Output Video Duration + Reference Video Duration). When a reference video is included, a lower unit price tier applies.

Note: The pricing table includes a 1080p tier, which is a reserved price tier on the platform. The current submission endpoint only supports 480p and 720p resolutions.

Application Scenarios

• Generating long-form, high-quality promotional or cinematic videos up to 30 seconds.
• Creating video content from multimodal references, such as driving a video generation using a specific character image and a reference audio track.
• Extending existing video clips to create longer, continuous narratives.
• Editing and modifying existing videos to update content or change specific elements.
• Producing localized content using multilingual prompt support for global audiences.

Model Comparison

Comparison Table 1: Seedance 2.5 Series

Note: The Spicy global version is a unified multimodal endpoint with the most comprehensive capabilities. Prices reflect iCreat platform pricing.

Comparison Table 2: Cross-Vendor Long-Duration Video Model Comparison

Note: Seedance 2.5 Spicy is the only model among the four supporting 30-second single-generation and video editing/extension capabilities.

Why Choose Seedance 2.5 Spicy?

• Unified Multimodal Endpoint: Accepts text, images, videos, and audio in a single request, simplifying complex generation workflows.
• Extended Duration: Capable of generating up to 30 seconds of video in one request, reducing the need for multiple generations and stitching.
• Advanced Video Manipulation: Native support for video editing and extension provides greater flexibility for iterative content creation.
• Global Content Policy: The Spicy global version is designed for a global user base, ensuring broad accessibility.
• Multilingual Support: Understands prompts in six languages, facilitating localized content creation.

Specifications

Architecture

Seedance 2.5 Spicy utilizes an asynchronous two-step API architecture. Users submit a video generation task via the submission endpoint with a flat request body structure containing the content array and parameters. The API returns a task_id. Users must then poll the task result endpoint using the same task_id. While the task is processing, the result array will be empty. Once the task succeeds, the result array will contain the generated video URL and download URL, along with the costUSD.

Notes

• Both steps of the API use the same task_id. While processing, the result array will be empty ([]); continue polling at 2-5 second intervals.
• A FAILED status is a terminal state. Check your request parameters and reference media if a task fails.
• If reference content contains human faces or copyrighted IP, the need_review parameter must be set to true, otherwise generation will likely fail.
• The role parameter is required for media types. Supported values are reference_image, reference_video, reference_audio, first_frame, and last_frame.
• When using last_frame, you should also provide first_frame following first-last frame control conventions.
• Reference media limits: Images must be 300-6000px with an aspect ratio of 0.4-2.5 and under 30MB. Videos must be MP4 or MOV format, 480p or 720p resolution, 2-15 seconds long, and under 200MB. Audio must be WAV or MP3 format, 2-15 seconds long, and under 15MB.
• Chinese prompts are limited to 2,000 characters, and English prompts are limited to 2,000 words.
• When a reference video is included, a lower unit price tier applies. Total Cost = Unit Price × (Output Duration + Reference Video Duration).
• The pricing table includes a 1080p tier, but the current submission endpoint only supports 480p and 720p. 1080p is a reserved price tier.

Frequently Asked Questions

How do I call Seedance 2.5 Spicy?

The API uses an asynchronous two-step process. First, submit your task to the endpoint https://api.icreat.ai/v1/task/submit/bytedance/seedance-2-5-global with your Bearer token. Then, poll the result endpoint https://api.icreat.ai/v1/task/result using the returned task_id until the status is SUCCEEDED or FAILED.

How is the cost calculated?

The total cost is calculated as Unit Price × (Output Video Duration + Reference Video Duration). Including a reference video applies a lower unit price tier. For example, generating a 10-second 720p video with a 10-second reference video costs $0.181 × 20 = $3.62.

What is the difference between the global version and the standard version?

The global version uses a different endpoint path (-global) and features a content policy tailored for a global user base. The model capabilities and pricing remain consistent between the versions.

When should I use need_review?

You must set need_review to true when your reference content contains human faces or copyrighted IP. If you do not, the generation will most likely fail.

What reference media and roles are supported?

The model supports three types of reference media: images, videos, and audio. The role parameter supports reference_image, reference_video, reference_audio, first_frame, and last_frame. Each media type has specific format, duration, and size limits.

What is the difference between Seedance 2.5 Spicy and Seedance 2.0 Spicy?

Seedance 2.5 supports longer durations (4-30 seconds), video editing and extension, the web_search tool, audio_url references, and multilingual prompts. However, its supported resolutions have converged to 480p and 720p, whereas 2.0 supported up to 4K.