Vidu Q4 Preview Reference-to-Video API

Vidu Q4 Preview Reference-to-Video is a multimodal video generation model developed by Vidu (Shengshu Technology), designed for high-consistency multi-subject reference generation and flexible first-frame/first-last-frame image-to-video generation. It supports up to 15 reference images and 3 reference audio tracks simultaneously, ensuring precise control over character identity, style, and environmental consistency across generated video sequences.

On the iCreat platform, this model is accessible via a unified asynchronous endpoint at https://api.icreat.ai/v1/task/submit/vidu/q4-preview. It supports two generation modes (reference and image) sharing the same endpoint, offering five resolution tiers from 540p to 4K, native audio synthesis, and multiple aspect ratios. Pricing is calculated per second of generated video, starting at $0.045 per second for 540P.

Model Positioning

Vidu Q4 Preview Reference-to-Video is positioned as a high-fidelity, multi-subject video generation model tailored for complex creative workflows. By unifying Reference-to-Video and Image-to-Video capabilities under a single endpoint, it eliminates the need for model switching, allowing users to seamlessly transition between strict multi-subject reference adherence and flexible frame-based transitions. It serves as the standard version on the iCreat platform, operating under standard content policies while delivering top-tier visual and audio fidelity.

Core Capabilities

Multi-Subject Reference Generation

The reference mode supports up to 15 reference images and 3 reference audio tracks simultaneously, enabling precise control over multiple characters, objects, and environments within a single generation. This allows for the creation of complex, narrative-driven video content where consistency across subjects is strictly maintained.

Flexible Image-to-Video Transitions

The image mode accommodates 1 to 2 frame images, supporting both first-frame generation and first-last-frame transition generation. This provides users with robust tools for animating static images or creating smooth, dynamic transitions between a defined starting and ending visual state.

Ultra-High Resolution and Native Audio

The model supports output resolutions up to 4K, ensuring exceptional visual detail and clarity. Additionally, it features native audio synthesis, automatically generating synchronized audio tracks that match the visual context, with the option to explicitly disable audio output if not required.

Unified Asynchronous API Architecture

Utilizing a standard two-step asynchronous flow (submit task and poll for result), the API ensures efficient handling of high-resolution video generation tasks. The request body uses a flat JSON structure, simplifying integration and reducing payload complexity.

Pricing

Total cost = Unit price × Output video duration.

Application Scenarios

• Short film and animation production with consistent multi-character scenes
• Marketing and advertising videos requiring high visual fidelity and brand consistency
• Music video generation with synchronized native audio and reference tracks
• Dynamic transition effects between keyframe illustrations
• High-resolution content generation for digital media and creative agencies

Model Comparison

Comparison Table 1: Vidu Q4 Preview Standard vs Global Version

Comparison Table 2: Same-Tier Multi-Resolution Video Model Comparison

Note: Vidu Q4 Preview offers the lowest unit price across all available resolution tiers among the compared models. The 540P tier is a unique low-cost entry point provided exclusively by Vidu.

Why Choose Vidu Q4 Preview Reference-to-Video?

• Unified endpoint for both reference-to-video and image-to-video generation, simplifying API integration
• Industry-leading multi-subject reference capacity with up to 15 images and 3 audio tracks
• High-fidelity output up to 4K resolution with native audio synthesis
• Flexible aspect ratio options (16:9, 9:16, 3:4, 4:3, 1:1) for diverse platform requirements
• Cost-effective pricing structure with a low-entry 540P tier starting at $0.045/second
• Standard flat JSON request structure for streamlined development

Specifications

Architecture

The model operates on a two-step asynchronous architecture. A client submits a video generation task via a POST request to the submission endpoint using a flat JSON payload. The server responds with a task_id. The client then polls the task result endpoint with this task_id until the status reaches a terminal state (SUCCEEDED, FAILED, or NOT_FOUND). During processing, the result array remains empty. Upon success, the response includes the video URL, download URL, and costUSD.

Notes

• The same task_id is used for both submission and polling; during processing, the result is an empty array, and polling should continue.
• FAILED and NOT_FOUND are terminal states; do not continue polling. Check your request parameters or media URLs.
• Generated media URLs are temporary and should be downloaded and persisted immediately. Do not hardcode temporary URLs.
• The request body must be a flat JSON object. Do not wrap parameters in an input or parameters object.
• The mode parameter is case-insensitive. In reference mode, the prompt is required and must not be empty; in image mode, the prompt is optional.
• sounds is only supported in reference mode (up to 3 tracks). Passing a non-empty sounds array in image mode will result in an HTTP 400 error.
• aspectRatio is only effective in reference mode. Do not pass it in image mode, as the aspect ratio automatically follows the first frame.
• isRec is only available in image mode. Do not pass it in reference mode.
• audio defaults to true. Set it to false to explicitly disable audio output.
• seed can be specified to fix the generation seed for reproducibility.

Frequently Asked Questions

How do I call Vidu Q4 Preview Reference-to-Video?

The API uses a two-step asynchronous flow. First, submit a task via a POST request to https://api.icreat.ai/v1/task/submit/vidu/q4-preview with Bearer authentication and a flat JSON body. Then, poll https://api.icreat.ai/v1/task/result using the returned task_id until the status is SUCCEEDED.

How is the cost calculated?

The total cost is calculated as the unit price multiplied by the output video duration, billed according to the selected resolution tier. For example, generating a 5-second video at 720P costs $0.095 × 5 = $0.475. The costUSD field is returned in the response upon successful task completion.

What is the difference between reference mode and image mode?

Reference mode supports 1–15 images and up to 3 audio tracks, requires a non-empty prompt, and allows the aspectRatio parameter. Image mode supports 1–2 frame images, makes the prompt optional, automatically matches the aspect ratio to the first frame, and enables the isRec parameter. Passing a sounds array in image mode will result in an HTTP 400 error.

What resolutions and aspect ratios are supported?

The model supports 540p, 720p, 1080p, 2K, and 4K resolutions. Supported aspect ratios include 16:9, 9:16, 3:4, 4:3, and 1:1. Aspect ratio customization is only available in reference mode.

What are the limitations for reference materials?

In reference mode, you can provide up to 15 images and 3 audio tracks. In image mode, you can provide 1–2 frame images. All provided media URLs must be publicly accessible.

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

The standard version and the global version (-global endpoint) differ in their endpoint paths and content policies (standard vs. global). Their core capabilities and pricing are completely identical. The standard version supports a higher limit for sounds (up to 3 tracks).