ON OCTOBER 24TH, THE VENTURE CAPITAL AGENCY A16Z RECENTLY PUBLISHED AN ARTICLE SAYING THAT IT IS NOT A GOOD IDEA TO BE ABLE TO DO THISAI video generationIt is not a "one-size-fits-all" "god model" in the field, but rather a phase of specialization and productization。
In its article, A16Z partner Justine Moore states that early industry competition was concentrated on benchmark indicators, such as time, physical effects and authenticity, but that a clear division of labour between different models has now begun。
For example, Veo 3 is better at "Physical Simulation and Synchronization of Audio and Video" for professional creative scenes, while Sora 2 focuses on "Dissemination and Multiple Camera Generation" for entertainment and engraving。
In addition to these two representative products, differentiation programmes such as Grok (Video Generation), Hedra (Long Video Person), Seedance Pro (Multi-photo scene) and Wan (Open Source Model) have emerged in the market. Moore compares this trend to the 17th and 18th century paintings moving from "puzzling" to "styping" that video models are entering an art age of "diversification and specialization"。
The article emphasizes that the greatest opportunity now lies in the product layer。
Despite the growing modelling capacity, ordinary users still need to rely on complex processes, such as Ideogram, nano-banana, Veo 3, Hedra and editing tools. Moore noted that these links could well be consolidated into one-stop products, thus lowering the threshold and unleashing greater market potential。
She also mentioned that the recent completion of the $130 million B round of financing by China AI Enterprise Liblibai had been the largest case of domestic AI application to date. This trend shows that the industry is moving from "roll model performance" to "roll product experience" and is the best time for start-up companies。