Alibaba (09988) Video Generation Big Model Wan 3.0 Starts Open Beta

The Zhitong Finance App learned that on August 6, Alibaba's (09988) video generation model Wan 3.0 began public testing. Wan3.0 has been fully upgraded in terms of generation time, versatile creation, all-round reference, and realism: it can generate a 30-second video at a time to fully express creative intent; in addition to the four basic modes of text, image, audio, and video, it supports document format input such as doc, xls, ppt, pdf, MD for the first time; strives to accurately restore the real world — thousands of people, and every frame is realistic and believable.

Wan3.0 supports single-segment 30-second video generation, and the narrative rhythm is easier. Complex camera languages such as continuous motion and one shot to the bottom can be fully expressed. AI video is moving from “generating one lens” to “telling a complete story.” It also supports an intelligent duration function, which can automatically recommend an appropriate length of time based on prompts; with a video extension function, it can also easily extend the story direction and plot development, so that creativity can grow naturally.

The portraits in Wan3.0 generated videos strive to be “a thousand people, a thousand faces”, and more keenly portray facial features and skin details. The characters express natural restraint, and their facial expressions are linked to each other; even in group portraits, they can delicately present the complex emotions of different characters.

In all-round reference tasks, Wan3.0 can accurately reproduce reference details to maintain consistency in key dimensions such as characters, props, sounds, spatial relationships, and style. For characters, fine-grained characteristics such as facial features, hair color, body, clothing, and accessories can be maintained stably to avoid acting; for props, the multi-angle appearance, hardware structure, logo, and material details are stable; for scenes, the relationship between character position and position perspective is accurately handled; for style, the tone of film and television shooting is also not easy to cause style confusion when creating in multiple styles.

For digital information, Wan3.0 enhances the aesthetic level of images, not only focusing on the harmonious combination of colors, but also maintaining the smooth expression of information presentation, making charts, data, and interface content equally textured. Wan3.0's video editing capabilities have also been greatly improved, and it is possible to modify images, plot, and lines. Wan3.0 still has room for improvement in sound quality and text accuracy.