Wan 2.5: Alibaba's Wan 2.5: First AI video model with native audio generation (komiko.app)

Alibaba has unveiled Wan 2.5, a multimodal AI video generator that the company says is the first to produce finished videos with native audio synchronization—ambient sound, background music and voiceover—directly in a single pass. Users supply a static image plus a text prompt describing motion and audio, and Wan 2.5 renders cinematic clips at 480p, 720p or 1080p with aligned soundscapes, claiming faster turnaround and lower cost than rivals like Google Veo 3 and Runway. The service runs on Alibaba Cloud’s DashScope platform and targets creators, marketers and educators who need professional-looking, ready-to-share video without separate audio editing. For the AI/ML community, Wan 2.5 signals a step toward tightly integrated audio–visual synthesis and joint multimodal alignment at scale. Key technical implications include end-to-end modeling of temporal visual motion and temporally coherent audio (music, ambisonics, speech), requirements for large paired audio–video datasets, and new evaluation challenges around sync quality and semantic audio relevance. If performance holds up, the model could streamline production pipelines and lower barriers to video creation, but it also raises questions about dataset provenance, copyright, and detecting synthetic media—areas researchers will need to address alongside further model improvements.