In 2024, the technology sector witnessed a historic moment as OpenAI officially launched its video generation large model, Sora, internationally; domestically, Kuaishou took the lead in introducing its self-developed video generation large model, "KeLing." This innovation signifies a new phase in AI technology, transitioning from text generation to video generation, and turning "video generation" from a laboratory concept into a powerful tool capable of simulating the real world. This development foreshadows an accelerated period of benefits in the AIGC (Artificial Intelligence Generated Content) field.
Technological Advancements: From Trials to Leaps
Driven by the digital wave, the channels for information transmission and reception have become more diversified than ever before. Videos, with their unique ability to present multi-dimensional information, rich visual expressions, and vivid dynamics, have emerged as a central force in the field of information dissemination. Compared to mere text descriptions and static images, videos can seamlessly integrate textual narratives, intuitive images, lively sounds, and fine visual effects, weaving a multi-layered information network within each frame to provide viewers with an in-depth experience and immersive enjoyment. This expressive method, which combines various sensory stimuli, greatly enhances the effectiveness and appeal of information transmission.
Advertisement
China's vast internet user base and rich resources of short video content provide a massive and high-quality foundational data support for the training and development of large video generation models, giving it a natural advantage in this area. Numerous products have emerged, such as Kuaishou's "KeLing," SenseTime's "Vimi," Zhipu's "QingYing," and Aisi Technology's "Pixverse." As a short video live streaming platform, Kuaishou's "KeLing" has unique advantages, not only performing well in hard indicators such as generation speed, video duration, coherence, aesthetics, and realism but also accurately understanding and creatively executing complex instructions. This allows AI-generated video content to approach human creative levels.
Empowerment Applications: From Imagination to Reality
The release of large model products like Sora and "KeLing" not only highlights the immense potential of AI in the field of video generation but also greatly stimulates people's imagination about its wide range of application scenarios. In the field of education, video generation technology makes virtual experiments and historical scene reenactments possible, greatly enriching the diversity and personalization of educational resources. In game development, AI-generated game scenes, characters, and plots lower the development threshold and enhance the playability and diversity of games. In the medical field, video generation technology assists in simulating surgeries and displaying lesions, providing doctors with intuitive diagnostic evidence. In the entertainment industry, AI not only assists in screenwriting but can also automatically generate scenes, characters, and special effects, significantly shortening production cycles and reducing costs. Currently, nine directors, including Li Shaohong, Jia Zhangke, Ye Jingtian, Xue Xiaolu, Yu Bai, Dong Runian, Zhang Chiyu, Wang Zi Chuan, and Wang Mao Mao, are relying on the "KeLing" large model to produce short film clips entirely generated by AIGC, showcasing "film-level" capabilities.
Industry Implementation: From Concept to Practice
In just half a year, large video generation models have quickly completed the transition from technological innovation to exploration and implementation, and then to widespread commercial application. For example, in the e-commerce sector, video large models have achieved substantial "empowerment" and tangible "quality and efficiency enhancement." Merchants can use AI to quickly generate product demonstration videos, showcasing the features and advantages of products to consumers in a more intuitive and vivid manner. This not only enhances the shopping experience but also promotes product sales conversion. Taking Kuaishou as an example, functions such as video script generation based on the "KuaiYi" large model, live broadcast real-time script generation, and advertising lead customer service, combined with digital human technology, help advertisers generate high-quality content at a low cost, improving lead conversion efficiency. This has resulted in an eightfold increase in Kuaishou's AIGC monthly active customer numbers since the beginning of the year, a 64-fold increase in monthly GMV scale, and a 12-fold increase in platform AIGC advertising revenue scale. AIGC marketing materials reached a daily consumption peak of over 20 million in June. To date, "KeLing" AI has served more than 2.6 million users, generating 53 million images and 27 million videos, demonstrating the huge potential of large models in commercial scenarios.
Future Outlook: Opportunities and Challenges Coexist
China places great emphasis on the development of generative artificial intelligence. The "Decision of the Central Committee of the Communist Party of China on Further Comprehensively Deepening Reform and Promoting Chinese-style Modernization" explicitly proposes to "improve the development and management mechanisms of generative artificial intelligence." Governments at all levels are actively promoting the implementation of relevant policies, continuously deploying around algorithms, computing power, data, scenarios, and talent in generative artificial intelligence, and supporting the development of generative artificial intelligence through multiple channels and means.At the same time, the widespread application of large models for video generation has also brought a series of challenges and issues, such as the transformation of creative paradigms. The creation methods of "text-to-video" and "image-to-video" are completely different from the original film and video creation, and it is necessary to study related tutorials as soon as possible to help large models better empower creation. In addition, due to the continuous improvement of the realism of generated content and the continuous reduction of thresholds, how to ensure the authenticity of generated content, avoid the spread of false information, and protect intellectual property rights, these issues need to be faced and resolved by us together. Recently, the Cyberspace Administration of China has issued the "Regulations on the Identification of Artificial Intelligence-Generated Synthetic Content (Draft for Comments)". Under the guidance of policies, the video generation industry will develop more standardized and healthily, contributing greater strength to social progress and prosperity.
Large models have entered the period of large-scale application, and the dividends of video generation are accelerating. This is not only another leap in AI technology but also opens up a new world full of infinite possibilities for us. I believe that with the joint efforts of all parties, AIGC led by video generation will accelerate its implementation, empower various industries, and promote society to move towards a better future.