Wan 2.6

Wan 2.6 is Alibaba's latest generation of visual generation models released in December 2025, comprehensively upgraded from the open-source Wan 2.2 architecture. The model uses a Mixture-of-Experts (MoE) design with 14 billion parameters, trained on 1.5 billion videos and 10 billion images.

The series' core breakthrough is its role-play capability — through the Wan2.6-R2V (reference-to-video) model, users can upload a character reference video capturing both appearance and voice, then generate new scenes starring that same character through text prompts. It supports single-person, multi-person, or character-object combinations, making it China's first video model with reference-to-video generation. The series also introduces intelligent multi-shot storytelling, automatically converting simple prompts into multi-shot scripts while maintaining character, scene, and key element consistency across shots.

Key Capabilities

• Reference-to-Video (R2V): Upload a character reference video — AI precisely extracts appearance and voice features to generate new scenes starring that character via text prompts, supporting people, animals, objects, or multi-subject combinations.
• Text-to-Video (T2V): Supports text and audio input to generate cinematic multi-shot videos, with deep understanding of lengthy Chinese and English prompts capturing nuance and artistic intent.
• Image-to-Video (I2V): Generate cinematic shots from a single image with first-frame and first-last-frame control for precise start and end positioning.
• Intelligent Multi-Shot Narrative: Automatically converts prompts into professional storyboard scripts, supporting panoramic, close-up, and tracking shots with smooth transitions while maintaining character and scene consistency.
• Audio-Visual Synchronization: Native audio-visual sync with precise lip-sync for multi-person dialogue scenes, more natural human voice expression, and enhanced music and song generation quality.
• Up to 15-Second Output: Video duration extended from the previous 10 seconds to 15 seconds (10 seconds for reference-based generation), providing more room for story development.
• Video Editing: Supports localized editing (subject or clothing replacement, bystander removal), video extension, frame expansion, and multi-image reference fusion.

Technical Strengths

Feature	Benefit
Role-play pioneer	China's first video model supporting reference-to-video generation, enabling users to insert themselves or any character into AI-generated scenes — a new paradigm for personalized video creation
Multi-shot narrative engine	Automatically plans and executes multi-shot sequences, maintaining character identity, lighting, and scene logic consistency within a single generation
14B parameter MoE architecture	Large-scale Mixture-of-Experts architecture combined with massive training data (1.5B videos + 10B images) ensures visual fidelity and instruction-following precision
Native audio-visual sync	Natural and fluent voice expression in multi-person dialogue scenes with precise lip-sync, sound effects, and background music matching visual rhythm
Open-source ecosystem	Built on the open-source Wan architecture, deployable and integrable via Alibaba Cloud Model Studio API, official website, and ComfyUI
1080p / 24fps	Output quality reaches 1080p at 24fps with realistic portrait texture and lighting, delivering content at professional production standards