Wan 3.0 Reference-to-Video

Wan 3.0 Reference-to-Video is a video generation model developed by Alibaba's Tongyi Wanxiang team, released in August 2026. The model supports multimodal combined input of reference images, reference videos, and reference audio to generate dynamic videos with high-precision physical motion simulation, camera control, and subject consistency. The model supports 480P, 720P, and 1080P resolution output, with a maximum duration of 30 seconds at 30 FPS and native audio-visual joint generation (ambient sound effects, background music, and multilingual lip-sync).

The iCreat platform exposes its reference-to-video capability via the endpoint https://api.icreat.ai/v1/task/submit/aliyun/wan3-0/reference-to-video, operating in an asynchronous three-step mode with 480P at $0.05/second, 720P at $0.10/second, and 1080P at $0.20/second.

Model Positioning

Wan 3.0 Reference-to-Video is positioned as the standard version's reference-to-video model of the Wan 3.0 series, balancing image quality and generation cost.

Compared to image-to-video (I2V) which only supports reference images, reference-to-video (R2V) additionally supports reference video and reference audio input, enabling simultaneous anchoring of visual style, motion patterns, and audio atmosphere. Compared to the Prime accelerated version, the standard version is priced lower, suitable for cost-sensitive scenarios.

On the iCreat platform, the model provides services through an asynchronous three-step workflow where users submit tasks, poll for status, and retrieve results.

Core Capabilities

Multimodal Reference Input

Supports combined input of reference images (reference_image), reference videos (reference_video), and reference audio (reference_audio), simultaneously anchoring visual style, motion patterns, and audio atmosphere for cross-reference consistency retention.

Native Audio-Visual Joint Generation

Natively supports ambient sound effects, background music, and multilingual lip-sync generation, completed in a single inference pass without post-production synthesis.

Multi-Resolution and Long Duration

Supports 480P, 720P, and 1080P resolution tiers, with a maximum single generation of 30 seconds at 30 FPS, meeting creation needs from short clips to complete scenes.

Flexible Aspect Ratio Control

Supports five aspect ratios: 1:1, 9:16, 16:9, 3:4, and 4:3, covering landscape, portrait, square, and special ratio display requirements.

Negative Prompt Support

Supports a negative_prompt parameter, allowing users to describe content that should not appear in the video, enabling precise control over generation results.

Pricing

Total cost = unit price × output video duration.

Note: The above are iCreat platform prices, consistent with Alibaba Cloud Model Studio official pricing (¥0.45/0.9/1.8 per second, 480P/720P/1080P). The iCreat platform price shall prevail.

Application Scenarios

• E-commerce product dynamic visualization and style-consistent advertising video production
• Short drama creation and film VFX segment rapid iteration
• Character consistency-preserving cross-scene video generation
• Audio-visual synchronized commercial advertising and brand content creation
• Creative design concept dynamic visualization

Model Comparison

Comparison Table 1: Wan 3.0 R2V vs T2V vs I2V (iCreat Endpoint)​

Note: R2V, T2V, and I2V are all called via the iCreat endpoint at identical pricing; R2V supports the richest reference input types.

Comparison Table 2: Wan 3.0 Standard vs Prime (iCreat Endpoint)​

Note: Standard is priced at approximately 70% of Prime, with slower generation speed; both versions share the same resolution, duration, and audio capabilities.

Why Choose Wan 3.0 Reference-to-Video?

• Supports reference image + video + audio tri-modal combined input, anchoring visual style, motion, and audio atmosphere
• Standard pricing lower than Prime accelerated version, suitable for cost-sensitive scenarios
• Supports 480P/720P/1080P three resolution tiers, from draft to final output
• Up to 30 seconds single duration, exceeding the limit of most competing models
• Native audio-visual joint generation with ambient sound and lip-sync in one pass
• Negative prompt support for precise content control

Specifications

Architecture

The iCreat platform's Wan 3.0 Reference-to-Video operates using an asynchronous three-step architecture. Users submit tasks via the submission endpoint, and the system returns a task ID. The task status is then polled via the query-status endpoint, and once complete, the final video result is retrieved via the get-result endpoint. Requests must include the API Key in the Authorization header for authentication, along with the X-ICREAT-AI-GROUP header. The request body contains the input object (prompt, negative_prompt, media reference array) and the parameters object (resolution, ratio, duration, watermark) as sibling fields. The media array supports three types: reference_image, reference_video, and reference_audio. The model ID is included in the submission endpoint URL path; the request body does not use a model field.

Notes

• The ratio parameter is required and must be specified when submitting a task
• The duration parameter ranges from 4 to 30 seconds; out-of-range values will be rejected
• Reference media (images/videos/audio) must be publicly accessible HTTPS URLs; private or authenticated addresses may cause task failure
• The model natively supports audio generation; generated videos include audio-visual content by default
• Standard pricing is lower than Prime; consider the Prime version if generation speed is a priority
• Please safeguard your API Key and avoid hardcoding it in client-side code or public repositories

Frequently Asked Questions

How is the billing calculated?​

Billing is calculated as the unit price multiplied by the output video duration. 480P is $0.05/second, 720P is $0.10/second, and 1080P is $0.20/second. For example, generating a 5-second 720P video costs $0.50. The iCreat platform price shall prevail.

What is the difference between R2V, T2V, and I2V?​

R2V (reference-to-video) supports combined input of reference images, reference videos, and reference audio, simultaneously anchoring visual style, motion patterns, and audio atmosphere. T2V (text-to-video) requires only text prompt input. I2V (image-to-video) supports text prompts and reference images. All three are priced identically; choose R2V if you need to use reference videos or audio.

What values does the ratio parameter support?​

Five aspect ratios: 1:1 (square), 9:16 (portrait), 16:9 (landscape), 3:4 (portrait special), and 4:3 (landscape special). This parameter is required.

What is the maximum single duration?​

The video generation duration ranges from 4 to 30 seconds, with a maximum of 30 seconds. The duration parameter must be an integer within this range.

What reference types does the media array support?​

Three reference types: reference_image (reference image), reference_video (reference video), and reference_audio (reference audio). They can be combined to simultaneously anchor visual style, motion patterns, and audio atmosphere.

What is the difference between Standard and Prime versions?​

The Standard version (Wan 3.0) balances quality and cost, priced at approximately 70% of the Prime version. The Prime version offers faster generation speed at approximately 1.4x the standard price. Both versions support the same resolution (480P/720P/1080P), duration (4–30 seconds), frame rate (30 FPS), and native audio capabilities. Choose Prime for speed; choose Standard for cost efficiency.