Li Xingye, Executive Director of the General Business Department of SenseTime's Large Model: How Can Generative AI Empower Brands to Expand Overseas? | 36Kr 2025 Overseas Expansion Conference

On July 25th, guided by the Department of Commerce of Zhejiang Province, the Secretariat of the China Cooperation Center for Special Economic Zones of the BRICS countries, the Hangzhou Bureau of Commerce, and the Qiantang District Bureau of Commerce, and jointly hosted by 36Kr and Qiantang Construction Group, the 2025 "From 'Ingenuity' to the 'World'" Going Global Conference will grandly kick off at the Grand Mercure Hangzhou Qiantang. As a brand - new IP event created by 36Kr focusing on the fields of globalization and going global, the conference will set up a main venue and a sub - venue, the "Investing in the BRICS" - Country Cooperation Matchmaking Meeting. The main venue of the conference will be divided into two major chapters: "Finding Certainty in Uncertainty" and "Doing Business Globally", focusing on popular going - global fields such as consumption, technology, e - commerce, finance, and new energy. It will cover more than 10 keynote speeches, 5 round - table discussions, and the release of the East Forward 2025 Going Global Globalization Innovation List, decoding the certain logic of the coordinated growth of "Product - Technology - Ecosystem" and providing a referable global development path for enterprises to navigate through the fog of globalization and build sustainable going - global capabilities.

On that day, Li Xingye, the Executive Director of the General Department of the Big Model of SenseTime Technology, will bring a keynote sharing titled "Empowering Enterprises' Last Mile of Going Global with AI Big Models".

The following is the content of Li Xingye's speech, compiled and edited by 36Kr:

Good afternoon, ladies and gentlemen! I'm Li Xingye from SenseTime Technology. Today's sharing might be a bit special - we're bringing a technical perspective. In a city like Hangzhou, which is at the forefront of technological innovation, it's most appropriate to talk about AI empowering going global. When I communicated with General Manager Zhang from Jiaogeyou yesterday noon, we also mentioned that many large platforms are now fully embracing AI. Today, I'd like to use about ten minutes to talk through cases and technical demonstrations about what specific empowerment generative AI can bring to the operation, marketing, and service links in going global.

Maybe you don't fully know SenseTime yet. We are the world's first listed pure - AI enterprise (excluding comprehensive companies like BAT). We focus on more vertical industries and develop various industry tools in depth, enabling partners to implement scenarios more conveniently.

Your early perception of SenseTime might be concentrated on the traditional business in the AI 1.0 era, such as face - recognition devices for hotel check - in and face - recognition security check channels at airports and subway stations. These all belong to our basic layout. But today, I'd like to focus on the hot topics in the past two years - big models and how generative AI empowers the operation and marketing of going global.

Next, I'll focus on sharing three core directions:

1. Digital Human Live - streaming Going Global: Reconstructing Live - streaming Efficiency with AI

First, clarify the concept of "digital human": it can be a concrete human - shaped image, a virtual IP like "Little Dolphin" of Bohai Bank, or even an abstract intelligent interactive entity like a smart speaker or Siri.

The core value of digital human live - streaming lies in the balance between efficiency and cost: its interactive effect is close to that of a real - person live - stream, far higher than that of recorded broadcasts and video clips, and it can achieve real - time interaction. For example, when I use my digital human for live - streaming, the big model behind it can respond to questions in the comment section in real - time without relying on a co - host or an assistant to control the situation. Although it can't replace the personal charm of top big V's like Luo Yonghao and Dong Yuhui, it's enough to replace small and medium - traffic anchors and cover more standardized live - streaming scenarios.

Digital human live - streaming has three major advantages in terms of operational capabilities:

- Quick adaptation to multiple scenarios: Just input product keywords (such as "the page - turner is lightweight and has good luster"), and a live - streaming script can be generated, supporting full - platform promotion (we have strategic cooperation with Weibo, Meituan, etc., providing them with digital human technology).

- Full - link marketing closed - loop: For example, when promoting on Weibo, it can not only generate live - streaming copywriting but also automatically capture popular category keywords to generate graphic content of 18 pictures + 120 words. It can also achieve intelligent comment control - whether it's cleaning up its own negative comments or more actively guiding public opinion in the comment section of competitors, it can improve efficiency through AI.

- Replication and scaling: One digital human can start live - streaming on multiple platforms simultaneously and can even "replicate" the styles of stars or KOLs (we call it an "intelligent agent"), achieving low - cost and large - scale operation.

2. Going - Global Content Generation: AIGC Reducing Marketing Costs

In traditional marketing, content production (graphic and video), promotion, and optimization are fragmented, while AIGC can achieve end - to - end full - process empowerment, significantly reducing costs.

- Text - to - video/graphic generation: Input a product description, and an advertising video with 4K precision can be generated, supporting large - screen playback. Based on line drawings or original pictures, any painting style can be generated (such as adding laser effects to shoes or changing the background), and derivative communication materials such as GIFs and H5s can be created, suitable for multi - platform distribution.

- Efficient model replacement and scenario - based creation: The most typical scenario in the e - commerce field is "quickly changing models in outfit pictures". Under the premise of compliance, without repeated shooting, different model images can be replaced for the same piece of clothing, significantly reducing the cost of real - person shooting.

- Full - link integration: From content generation, traffic injection to comment control, AIGC can complete it all in one stop, equivalent to "AI agency operation", which can reduce operating costs by more than 50% compared with the traditional model.

3. AR Product Special Effects: Improving Experience and Conversion with Technology

Our joint laboratory with Zhejiang University has been deeply involved in AR (Augmented Reality) technology and has implemented a large number of cases in scenarios such as the Hangzhou Asian Games and the West Lake intelligent experience. Its core application is in product interaction and experience upgrading:

- Virtual try - on and try - wear: It supports online virtual try - on of makeup and accessories (categories with high rigidity such as shoes, hats, bags, and watches have excellent effects. Clothing has not been fully implemented yet due to the complexity of details such as wrinkles and light spots). For example, internet celebrity anchors don't need to repeatedly apply lipstick for color testing, and users can directly experience the effect of different lipstick shades on their faces in the APP. The fit and efficiency are far higher than the traditional method. Currently, we have reached cooperation with brands such as China Duty Free Group, Henkel, Ralph Lauren, and Dewu, covering both online APPs and offline intelligent fitting mirrors.

- Full - scenario Penetration of Intelligent Agents: The "intelligent agents" extended from AR technology can cover more service scenarios:

l E - commerce shopping guide: for example, realizing 3D shoe and watch try - on in the Dewu APP;

l Supply chain management: providing intelligent planning for the warehouses of Fortune Global 500 companies (such as product placement, customer matching, and contract management);

l Smart home: realizing the function of a "smart housekeeper" through AIoT devices (cameras, speakers), such as inquiring about the home situation when on a business trip ("when did the courier arrive", "did the pet cause trouble"), and reminding the elderly to take medicine.

- Intelligent Public Opinion Management: Through AI, it can capture comments across the whole network in real - time (such as negative information like "a new - energy car caught fire"), quickly generate response words, and even guide public opinion through large - scale positive content distribution. We have cooperated with several 4A companies, covering public opinion management in text, picture, and video forms.

Conclusion: AI is a Tool Revolution and a Productivity Revolution

We always believe that AI is not a gimmick but a tool that can be implemented. Through technologies such as big models, AIGC, and AR, it reconstructs the "Cost - Efficiency - Experience" triangle of going - global marketing. SenseTime's algorithms are still iterating rapidly, and we also look forward to cooperating with more industry partners to explore more scenario values.

If you have ideas about AI applications or are in the upstream or downstream industrial chain of AI, welcome to communicate. Currently, our technology has been connected to platforms such as Amazon, Microsoft, and Alibaba Cloud Overseas, which is sufficient to support the full - scenario needs of Chinese enterprises going global.

Thank you for listening!