Wan 2.6 Reference-to-Video Spicy API

Wan 2.6 Reference-to-Video Spicy is Alibaba Cloud's next-generation reference-guided video generation model, released with the Wan 2.6 series on December 16, 2025. According to Alibaba Cloud's official announcement, it is the first reference-to-video generation model in China: users upload character reference media containing appearance and voice, then use a text prompt to place the same character in entirely new scenes. Each run delivers HD output up to 15 seconds at 720P or 1080P, with consistent identity preservation and preserved character voice. As the Spicy variant, it targets bolder, more expressive creative direction with relaxed content restrictions, suited to mature-themed creative production.

The model is available through the iCreat platform as an API using a two-step async task flow: submit a task to get a task_id, then poll the result endpoint until completion. It supports 720P and 1080P resolutions (four size options across landscape and portrait) with duration-based billing: $0.10 per second at 720P and $0.15 per second at 1080P, with reference inputs not billed.

Model Positioning

Wan 2.6 Reference-to-Video Spicy is positioned for character-consistency-first video production. Unlike text-to-video generation that starts from scratch, it anchors the subject's appearance and voice with reference media, then drives the performance with a prompt — ideal for continuous content where the same character appears repeatedly. Within the Wan 2.6 family, it complements text-to-video and image-to-video: T2V handles free creation from text, I2V animates still images, and reference-to-video brings existing characters into new scenes. Typical needs include character consistency, style transfer, e-commerce showcases, and creative visual production.

Core Capabilities

Reference-Guided Generation

Generates videos conditioned on external reference media such as reference video clips, character identity references, and scene elements. The prompt describes the target scene, character actions, and visual details, while the reference media anchors subject appearance and style.

Identity and Voice Consistency

Transfers the reference subject's appearance and voice into newly generated scenes. People, animals, and objects can all take the starring role, and multiple subjects can co-star while maintaining their distinct identities, with realistic motion dynamics.

HD Output and Flexible Duration

Supports 720P and 1080P resolutions across four landscape and portrait sizes — 1280*720, 720*1280, 1920*1080, and 1080*1920 — with 4–15 seconds per run, covering the mainstream short-video duration range.

Precise Action Alignment

Aligns actions with the prompt precisely: character performances and camera work stay consistent with the described intent, keeping the narrative expression accurate.

Pricing

Total Cost = Unit Price × Output Video Duration. Only the output video duration is billed; reference inputs are not charged separately. The costUSD field in the response returns the actual cost once the task succeeds.

Application Scenarios

• Character-consistent content: IP characters, creators, and virtual personas appearing stably across scenes
• Short dramas and narrative work: role play, multi-subject co-starring, and serialized storylines
• E-commerce showcases: product and model showcase videos built from reference media
• Style transfer: carrying the appearance and style of reference media into new scenes
• Creative visual production: rapid production of ads, promos, and social media videos

Model Comparison

Same-Series Comparison

Cross-Model Comparison

Same-series data is from the official Alibaba Cloud release; this model's specifications are per the iCreat channel documentation.

Why Choose Wan 2.6 Reference-to-Video Spicy?

• Reference media anchors character identity, so you never re-describe character details in every prompt
• Dual consistency of appearance and voice — the character not only looks right, the voice travels with them
• Transparent billing: charged by output duration only, with reference inputs free
• A two-step async API with clear status semantics — status and costUSD make progress and cost visible at a glance
• HD output at 720P / 1080P and 4–15 second durations covering mainstream short-video scenarios

Specifications

Architecture

The Wan 2.6 series was released in December 2025, covering five model types: text-to-video, image-to-video, reference-to-video, image generation, and text-to-image. Reference-to-video uses a reference-guided generation architecture: the model conditions on external reference media (reference video clips, character identity references, scene elements), transfers the reference subject's appearance and voice into the new scene, and generates dynamic footage driven by the text prompt while preserving identity across multiple co-starring subjects. At the series level, it supports multi-shot narrative and native audio-video sync, with improved instruction following and visual quality over the previous generation.

Notes

• The submit and query steps must use the same task_id
• While processing, result is [] — keep polling, at a recommended interval of 2–5 seconds
• FAILED is terminal; check request parameters and reference URLs, then resubmit
• URLs in reference_urls must be HTTPS
• input and parameters are sibling top-level fields; do not nest them
• Both prompt and reference_urls are required; omitting either fails submission
• costUSD is returned only on success; failed tasks carry no cost field

FAQ

How do I get started with the Wan 2.6 Reference-to-Video Spicy API?

Register on the iCreat platform and obtain an API Key from the console, submit a generation request to the submit endpoint, and poll the result endpoint with the returned task_id. All requests are authenticated via the Authorization: Bearer header. Keep your API Key safe and never expose it in client code or public repositories.

How is billing calculated?

Total Cost = Unit Price × Output Video Duration. 720P is $0.10 per second and 1080P is $0.15 per second, billed on output duration only; reference inputs are not charged separately. For example, a 720P, 5-second video costs $0.50. The costUSD field in the response returns the actual cost once the task succeeds.

What reference media is supported?

Reference videos and reference images are supported, passed as an array of HTTPS URLs via reference_urls to guide character appearance, motion, or style. In the prompt, use names such as Character1 and Character2 to refer to different reference subjects and describe their actions and interactions.

What resolutions and aspect ratios are supported?

parameters.size accepts 1280*720, 720*1280, 1920*1080, and 1080*1920, corresponding to landscape and portrait frames at 720P and 1080P. Choose the size that fits your target platform.

How do I query the task result?

Send a POST request to https://api.icreat.ai/v1/task/result with task_id in the body. A status of SUBMITTED means processing — result is [] and you should keep polling. When status is SUCCEEDED, result returns the video resource array; read url or download_url to retrieve the video.

What if the task fails?

FAILED is a terminal state. Check in order: whether the prompt is empty, whether the reference URLs are publicly accessible, whether size is one of the four allowed values, and whether duration is within 4–15. Fix the issue and resubmit. The cost field is returned only when the task succeeds.