Alibaba: wan2.2-t2v-plus API

Wan2.2-t2v-plus (A14B) is a professional-grade text-to-video foundation model featuring a state-of-the-art Mixture-of-Experts (MoE) architecture. By decoupling the denoising process into specialized experts (High-Noise Expert for global layout and Low-Noise Expert for aesthetic refinement), it achieves a total parameter count of 27B while keeping inference costs equivalent to a 14B model. This "Plus" version is trained on a massive dataset with an 83% increase in video volume compared to Wan2.1, resulting in top-tier performance for cinematic lighting, complex human motion (facial expressions, sports), and precise semantic alignment with long, detailed prompts.

• Input: text
• Output: video
• Released: 2025-07-27
• Knowledge cutoff: 2025-05-31