Alibaba has introduced the beta version of its Wan3.0 generative AI video model, capable of producing video clips up to 30 seconds long while supporting multimodal reference inputs. The new model is available for public beta testing through Alibaba Cloud’s AI development platform Model Studio and the AI-native cloud platform Qwen Cloud, where users can apply for access.
Wan3.0 Offers Big Upgrade For GenAI Video
Wan3.0 extends the maximum video duration from the 15-second limit of the previous Wan2.7-Video model to 30 seconds, which means it should enable more complex camera movements, continuous shots and extended narrative scenes. The model also includes an intelligent duration recommendation feature that suggests an appropriate video length based on user prompts, along with video extension functions for expanding existing clips.
A key addition in Wan3.0 is its support for multimodal inputs. The model can process text, images, video and audio simultaneously, and also accepts web pages, PDF documents and PowerPoint presentations as reference sources for video generation. According to Alibaba, the system is designed to convert static text-heavy content directly into dynamic video output, reducing the need for separate video production workflows.
Wan3.0 gets visual continuity features intended to minimize the drifting and distortion commonly associated with AI-generated video. Alibaba said the model can render realistic human faces with synchronized micro-expressions, generate multilingual voice output and maintain stable software interface elements and motion graphics. The company also said the model is capable of reproducing detailed reference information, including character appearance, product details, spatial relationships, audio characteristics and visual styles, while preserving layout and voice consistency across generated scenes.
Alibaba said Wan3.0 is intended for a range of applications, including film and video production, short-form drama (“duanju” as it is referred in Chinese-speaking communities) creation, social media content, marketing videos, educational content and simulation video generation for autonomous driving and robotics training.
Pokdepinion: Amidst all the generative AI developments, nobody seems to be clearing the air on the ethics and copyright side of things after all these time.
