騰訊的李玉濤談世界盃直播：AI大規模製作的首秀以及雲提供商向多模態領域的推進

Tencent Cloud served as the technical backbone for the recently concluded FIFA World Cup across the US, Canada, and Mexico, powering official broadcasts in 17 countries and regions. This coverage reached two-thirds of all authorized platforms in Asia-Pacific and domestic markets. Li Yutao, Vice President of Tencent Cloud and Head of International Product Technology, highlighted in a recent interview that AI was deployed at an unprecedented scale for live broadcast production. Key processes—including image enhancement, intelligent directing, automated editing, smart horizontal-to-vertical conversion, quality inspection, and traceability—were fully automated. This support for a single World Cup event marks just one facet of AIGC’s transition into large-scale production.

Li Yutao describes Tencent Cloud’s approach in the multimodal space as "Harness." Rather than focusing solely on how models generate content, the strategy centers on the entire supply chain from generation to stable user delivery. This encompasses preprocessing, quality inspection, stitching, encoding, distribution, and format adaptation.

He noted that generative AI has introduced multiple challenges for multimedia applications, particularly in live streaming, on-demand viewing, media creation, and interactive scenarios. These are constrained by network transmission, video quality, user experience, and the integration of various features, making it difficult for a single model to address all issues. Many vendors attempt to connect every new multimodal large model as soon as it emerges, leading to integration problems and inevitable technical hurdles. When most companies face similar challenges, a common demand emerges from customer feedback: there is a need for a provider to handle underlying tasks, allowing businesses to focus on innovation and upper-level applications. Identifying user needs, selecting appropriate models, fixing inherent flaws, and delivering final services to users are the core challenges a video and audio PaaS cloud provider must solve.

On June 5, at the Tencent Cloud AI Industrial Application Conference, the company launched its AI brand "Tencent Cloud WAND." This initiative includes six self-developed media-specific models and over 60 AI capabilities covering the entire pipeline of content generation, understanding, processing, and encoding. These tools are specifically trained for vertical scenarios such as e-commerce, short dramas, education, and sports live streaming. Li Yutao explained that the team integrates all models, including large language models, multimodal models, and small-parameter models for image enhancement or specific applications. Work on the multimodal Harness began last year, coinciding with the rise of generative video models and a growing industry consensus on the Harness concept.

Video generation is widely regarded as the second AI transformation scenario to achieve a commercial closed loop, following AI coding. Since early this year, competition in large model video scenes has accelerated, significantly altering market structures. Domestic vendors are advancing rapidly in commercialization, with ByteDance’s Seedance and Kuaishou’s KeLing AI forming a duopoly in the AI video generation market. Public data indicates that by March, KeLing AI’s ARR approached $500 million, representing fourfold growth in a year. Other reports suggested that by mid-year, Seedance 2.0’s ARR reached $2 billion. Although ByteDance denied the revenue figures as "excessively high and inconsistent with reality," it confirmed that Seedance 2.0 has crossed the "productivity turning point." Meanwhile, in overseas markets, OpenAI announced in March that it would discontinue its independent app, API interface, and ChatGPT-integrated video functions for Sora, effectively exiting the consumer-level AI video generation market.

When developing To B products and solutions, ecological cooperation is Li Yutao’s top strategic priority. He acknowledged that his team has debated whether the role of Harness might be ignored or bypassed as the link between generation and interaction shortens. His assessment, based on six months of observation, is that this will not happen; rather, the importance of Harness may increase. If large models represent backend productivity, their efficiency gains will significantly stimulate demand for front-end applications. The breadth of demand growth will always outpace productivity improvements, meaning large models can never fully meet end-user needs. This persistent "supply-demand gap" underscores the necessity of Harness.

Li Yutao identifies two major opportunities for Tencent Cloud in the multimodal Harness space. First, many video and audio scenarios require rethinking. Second, as more powerful and comprehensive models emerge, the challenge lies in quickly applying them to enterprise-level production services and delivering the latest large model experience to end users. Continuously integrating the latest models for customers and managing the entire production process and toolchain are critical objectives of the multimodal Harness.

Since last year, AI computing power has shifted from training to inference. Many companies now spend more on inference than on training, indicating that numerous AIGC applications are entering large-scale production. In China, short dramas and animated series are typical scenarios where multimodal technologies are widely used for production. In the Asia-Pacific region, video and audio solutions serve as the "frontline" of Tencent Cloud’s international expansion, leading in revenue, market share, and industry awareness. Over the past five years, Tencent Cloud has prioritized internationalization, investing heavily in business resources and infrastructure abroad. The company disclosed that its overseas business has maintained double-digit growth for the past three years.

Li Zhicheng, General Manager of Tencent Cloud Video and Audio, noted that whether for short dramas, animated series, e-commerce, or tool-based scenarios, clients initially prioritize effectiveness and influence but later focus on ROI—specifically, whether the technology generates profit and creates value. The most frequent inquiries now concern short dramas and e-commerce tools, which are expected to become widespread in the next 1-2 years. Many users and enterprises will adopt these technologies, similar to large language models, which were once confined to vertical industries but are now ubiquitous and integral to daily work.

With the arrival of the Agent era, Tencent Cloud’s capabilities will integrate with video and audio intelligents, AI hardware, embodied robots, and cloud phones. Leveraging global infrastructure and edge inference, these solutions will support the rapid global deployment of AI applications. Li Yutao mentioned that the team has collaborated with several embodied intelligence and autonomous cleaning vehicle companies, such as Zhuji, to develop TRRO remote real-time control (a remote digital human solution). This addresses core issues like low-latency stable communication and audio-video data transmission during manual intervention in weak network conditions.

Looking ahead, Li Yutao outlines three key directions for competition among cloud providers in the multimodal field. First is multimodal training and computing power. Unlike language model training, a significant portion of investment in the multimodal field goes toward understanding and generating content. Future end-to-end voice interactions and video generation will require substantial computing power and infrastructure investment from cloud providers. Second is storage and data accumulation. Multimedia demands higher storage capacity than text, with distinct requirements for stability, data cleaning, and acquisition. Large volumes of video data must be cleaned and processed before pre-training, and combining different modalities involves processing steps that differ significantly from text-based tasks. Third is the application production side. Cloud providers offering end-user-oriented inference services and comprehensive processing capabilities are key factors in customer choice. Effectively meeting customer demands driven by operational innovation trends is equally critical for Tencent Cloud.