Wu Haibo, General Manager of WeShop: The era of AI entrepreneurship is no longer about "shell - wrapped applications" | 2025 AI Partner Conference

The year 2025 marks the beginning of the explosive era of AI applications. As the global AI competition enters the "China Moment," a profound technological revolution is quietly reshaping the industrial landscape. At this critical juncture, the industry faces core questions: How can we bridge the gap between AI technology and large - scale applications? Where will the next disruptive AI super - application be born?

On April 18th, the 2025 AI Partner Conference, hosted by 36Kr, grandly kicked off at the Modu Space in Shanghai. The theme of this conference is "The Super App is Coming," focusing on the disruptive transformation of AI applications in all industries. The conference is divided into two major chapters: "The Super App is Coming" and "Who Will Be the Next Super App," covering seven major topics such as "Growing Up in the AI World" and "In 2025, Focus on Super Apps in AI." It includes more than 10 keynote speeches, 3 round - table discussions, and two sessions for releasing the lists of outstanding AI case enterprises, deeply analyzing how AI technology reconstructs business logic and reshapes the industrial landscape, and exploring the infinite possibilities brought by AI super - applications.

On that day, Wu Haibo, the general manager of WeShop, gave a keynote speech titled "WeShop's Globalization Journey: Evolution or Elimination."

The following is the content of Wu Haibo's speech, edited by 36Kr:

Thank you for the invitation from 36Kr! Good afternoon, everyone. I'm Wu Haibo from WeShop. Today, I'd like to share some observations and thoughts from our practical experiences in the past two years of WeShop's entrepreneurship, hoping to inspire entrepreneurs in the AI wave.

Wu Haibo, General Manager of WeShop

Let's take a minute to watch a video to get an intuitive understanding of WeShop's business.

To put it simply, WeShop specializes in providing AI commercial photography solutions for e - commerce enterprises. In traditional e - commerce operations, merchants need to spend a lot of money on hiring models, renting venues, and hiring photographers to complete product photography. However, with our self - developed AI tools, we help merchants change models and backgrounds with one click and efficiently generate product display images. As the first team in China and one of the first globally to launch AI commercial photography tools, WeShop inherits the fashion e - commerce genes from Mogujie. Currently, we mainly target overseas customers and charge in the SaaS subscription model. I have been deeply involved in product development and search recommendation algorithms at Mogujie for many years. Since 2021, I have been engaged in large - model research and development. Now, I'm responsible for product development and cooperation with major customers at WeShop.

The entrepreneurial experience in the past two years has made me deeply understand the two sides of the large - model wave. I'm used to sharing the "bad news" first and then the "good news" - after all, only by facing challenges can we seize opportunities.

Bad news: Models are applications, and startup companies face a survival crisis. The cruel reality in the era of large models is that "models are applications" is becoming the mainstream. Don't fantasize about using a simple "shell - wrapping" approach to deal with competition. OpenAI once pre - charged developers with $200 in computing power. Although it seems to support the ecosystem, it may instantly disrupt businesses that rely on its interfaces. This situation of "getting the computing power but losing the business" has repeated itself in the past two years. For entrepreneurs, if they can't deeply integrate large - model capabilities into their businesses, it's like building a building on the beach, which may be washed away by the technological wave at any time.

Good news: The SOTA curse and the open - source dividend. However, from another perspective, the "chaos" in the large - model field is precisely an opportunity for entrepreneurs.

The SOTA (State - of - the - Art) curse: In the fields of images and videos, no model can maintain the technological peak for a long time. Take the model ranking list on Hugging Face as an example. The once - popular Midjourney is no longer at the top; the HiDream model launched by the JD team has jumped from obscurity to the second place in just a few months. The competition in the language - model field is even more intense, and the speed of technological iteration far exceeds our imagination.

The rise of the open - source ecosystem: When tech giants are betting on closed - source models, Meta's open - sourcing of Llama has shaken the industry and spawned a huge ecosystem. In the image field, some teams found it difficult to surpass leading players by simply competing in models, so they chose to open - source their models to attract global developers to optimize them together. Although the initial performance of open - source models may be weaker than that of closed - source models, their progress speed is amazing with the power of the community.

Key point: Entrepreneurs should learn to leverage open - source resources. For any "black technology" of a closed - source model, there will probably be an open - source alternative within 2 - 3 months. This rapid technological diffusion actually provides an opportunity for small and medium - sized teams to overtake on the curve.

Take WeShop as an example. Our technological iteration has almost entirely benefited from the promotion of the open - source ecosystem. From 2023 to 2024: The product images generated by the initial version impressed customers, but there were obvious flaws in details - unnatural clothing wrinkles, a strong "AI feeling" on people's faces, and a rigid background synthesis. In 2025: With the latest open - source models and fine - tuning technologies, we have achieved a qualitative leap. Now, the product images can not only accurately restore the details of clothing but also simulate the lighting effects of real shooting. At 2K resolution, the texture of the generated images is almost indistinguishable from that of real - shot images.

Let's look at some real - life cases:

Scenario 1: Product display images in shopping malls. The 2024 version could only achieve basic replacement, while the 2025 model can generate more realistic lighting and more natural product placement angles.

Scenario 2: Wedding dress photography. Compared with the results generated by GPT - 4o, WeShop's images are superior in terms of clothing texture and wrinkle details - which is crucial for e - commerce because product images must match the physical products 100%.

Core logic: A startup team of less than 20 people could not have achieved such rapid technological iteration in a short time without the help of open - source models.

How can startup companies avoid being "submerged" by the impact of large models? I've summarized two key strategies:

1. Choose scenarios with "strategic depth"

Not all scenarios are suitable for startup teams to enter. If a business can be completed by hiring an employee with a monthly salary of 5,000 yuan, it is likely to become the "main track" for large models and will be quickly monopolized by giants. On the contrary, if a scenario is highly complex and requires strong professional capabilities (for example, it needs talents with a monthly salary of 20,000 yuan or even higher), it means a high technological threshold and great value, and startup companies have a better chance to build barriers.

Case: The product images generated by GPT - 4o are "good enough" for ordinary C - end users, but far from sufficient for e - commerce merchants. Merchants need to accurately restore product details to ensure that the physical products received by users match the images. This professional requirement is WeShop's "strategic depth."

2. Build a "model - friendly" business

Don't confront large models head - on. The core of a startup team is not to self - develop large models but to make models a "beta" to boost the business. For example, when GPT - 4o emerged, we didn't panic. Instead, we deeply analyzed its technical architecture (such as the defects of the DR + deficient architecture in detail processing) and predicted the possible breakthrough directions of the open - source community. Once a new open - source model emerges, we can quickly integrate and optimize it, continuously widening the gap with general models in high - resolution scenarios such as 2K and 4K.

Key understanding: Entrepreneurs must understand AI, deeply study the model structure and the direction of technological evolution, and even read a large number of papers. Only by understanding the ability boundaries of large models can we predict technological trends and make early arrangements.

In terms of user growth strategies, the AI era is fundamentally different from the traditional Internet era:

Internet era: Product homogenization is serious, and the user migration cost is low. Enterprises have to invest huge budgets to compete for traffic. "Users are always more expensive now than in the future."

AI era: Technological iteration is extremely fast, and new products emerge in an endless stream. Users' loyalty to AI tools is almost zero. Instead of spending resources to compete for existing users, it's better to focus on refining products and wait for the explosion of "killer applications." I believe that the cost of acquiring AI users will be lower in the future - because there is no real monopoly in the market yet, and new players may attract users through technological innovation at any time.

Finally, I'd like to give a piece of advice to entrepreneurs: In the AI wave, the most important thing is to "stay at the table." The arrival of AGI (Artificial General Intelligence) will completely reshape all industries, and our current imagination may just be the tip of the iceberg. As a startup team, we don't need to pursue "big and comprehensive." We can focus on niche scenarios like WeShop, look for opportunities outside the "range" of giants, and create more possibilities for the future through continuous innovation and technological iteration.

That's all for my sharing today. Thank you!