Wan

More precise identity control points to stronger facial variation and better character consistency across generations. Palette-driven prompting with up to 8 hex codes suggests finer aesthetic control for brand colors and art direction. Stronger text rendering implies better handling of posters, charts, formulas, and other dense layout-heavy outputs. Multi-image composition indicates improved continuity when generating grouped visuals or related asset sets. Reference-aware generation plus localized editing points to a more controllable image workflow, where subjects stay stable and changes can be applied without regenerating the whole scene.

Supports up to 15-second 1080p HD video generation, enabling richer stories, commercials, ads, and cinematic clips beyond earlier short bursts.  Adds video reference support, allowing users to upload a short video snippet to replicate a specific character’s appearance and voice for new scenes.  Introduces multi-shot narrative generation, where the model maintains coherence across multiple scenes in a single video, intelligently segmenting shots based on prompt structure.  Significantly improved audio synchronization with stable multi-person dialogue and natural voice characteristics, supporting richer soundscapes and narration.  First model in the series with role-playing capabilities, enabling creators to appear in AI-generated videos with preserved identity and consistent visuals and voice attributes.  Supports text-to-video, image-to-video, and reference-driven generative workflows, offering broader flexibility for storytelling and creative production.  Designed for social media creators, marketers, educators, and filmmakers, with output suited for campaign videos, narrative storytelling, and dynamic content. 

Higher resolution support: Generates videos in 480p, 720p, and 1080p One-pass audio-video synchronization: Creates fully synchronized videos with audio, voiceover, and lip-sync from a single prompt—no separate recording needed Extended duration: Supports up to 10 seconds of video (compared to shorter durations in earlier versions) Multiple aspect ratios: Offers 6 different aspect/size options (16:9, 9:16, 1:1) for flexibility Custom audio: Allows you to upload your own audio or let the model generate it Strong multilingual support: Reliably processes Chinese and other languages for A/V-synced videos

Introduces Mixture-of-Experts (MoE) architecture for scalable performance.  Cinematic visual quality with enriched training data (significantly larger dataset than 2.1).  Improved motion coherency and prompt adherence versus Wan 2.1.  Supports 720p @ 24 fps and hybrid tasks like text-image-to-video (TI2V).  New variants: Animate-14B (character animation), S2V-14B (speech-to-video), TI2V-5B (compact hybrid).  Integration with tools like Diffusers, ComfyUI, ModelScope & Hugging Face.