Preview of the Chengdu Big Model｜Dialogue with Haiyi Interactive Entertainment: Pursuing rapid iteration and finding a balance between technology and market demand
2024-08-14
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
Preview Chengdu Large ModelThe development of artificial intelligence is in full swing, integrating with thousands of industries and generating new technological power. As the key to the competition of artificial intelligence, big models will enter the first year of monetization this year.Red Star Capital noted that in this industry competition, the compound growth rate of Chengdu's artificial intelligence industry has exceeded 40% in the past three years, and its "competitiveness of artificial intelligence technology industry" has ranked sixth in the country. Sichuan Province has even listed artificial intelligence as the "No. 1 Innovation Project" this year.Recently, Red Star Capital interviewed several AI large model manufacturers in Chengdu and discussed with them the "Chengdu power" in the artificial intelligence trend.
In May this year, the large models of three companies in Chengdu successfully passed the national generative artificial intelligence service registration, and "Haiyi AI" or "Haiyi Painting Large Model" was one of them.
In April this year, Haiyi AI, less than a year after its establishment, had over 130 million model calls, a library of over 1 billion materials, over 5 million registered users worldwide, and 2 million monthly active users. According to data from website traffic monitoring service provider Similarweb, Haiyi AI's visits in March exceeded 10 million, ranking fourth on the global AI raw image list.
Recently, Red Star Capital Bureau interviewed Zhang Yi, an operational product expert at Haiyi Interactive Entertainment. He believes that large-scale model manufacturers should find a balance between technology and market demand. While investing in technology, they also need to control market needs and user acceptance of AI.
Zhang Yi
Large model developmentat the same timeControl market demand
Red Star Capital Bureau: Only one year after its establishment, Haiyi’s products ranked 4th in the global AI raw image list. How did it achieve such “speed”?
Zhang Yi:I think "fast" is mainly reflected in version iteration. We have a strong R&D team internally, and the internal iteration speed is very fast.
In terms of ease of use, we have simplified the platform information or optimized parameter prediction to make the product entry level lower; in the process of AI generation, such as text-generated images, image-generated images, or conditional-generated images, the sections will generate workflows or work processes that suit different users' needs. The entire workflow is called "AI workflow". We are also continuously optimizing in this regard, constantly enriching and improving the workflow.
From the launch of the project in April last year to April this year, we have been launching new AI-expanded products every month, including AI photo or simplified creative flow, which are suitable for UGC (user-generated content) application scenarios, as well as applications such as drawing boards, model training, and comfyui that are more suitable for PGC (professionally produced content).
What we have done is equivalent to making a scalable ecological product. As our ecosystem continues to expand and there are more and more users, it will be a bit like a snowball. As the number accumulates, we will have enough competitiveness.
Part of our team actually came from the game industry. This means that the requirements for R&D are very high. They switched to the AI field very quickly, and it can even be considered a "dimensionality reduction". We went through about two months of relatively intensive development iterations, and the product quickly exploded after it was launched.
Red Star Capital Bureau: The difficulty for many large-scale model companies lies in integrating with the industry. Is Haiyi Interactive Entertainment's logic to find the scenario first and then solve the technology?
Zhang Yi:For us, we only need to focus on technical research and don’t have to worry about finding specific landing scenarios. In fact, our users create the scenarios, and then we solve the problems for them.
Because the platform we provide is designed to empower users, users can use the platform functions to create their own application scenarios based on their careers or needs, which is equivalent to everyone having their own unique application scenarios.
This ecosystem makes the Haiyi platform a combination of "multiple practical AI workflows". Haiyi only needs to play the role of "main beauty", and the rest of the functions are implemented by open source models. The advantage of this is that Haiyi has now accumulated tens of millions or even hundreds of millions of high-quality data or pictures, which can be used to feed back, build, and optimize our own self-developed models.
On the one hand, we invest in the research and development of our own models, and on the other hand, we also control the degree of current market demand and the degree of user acceptance of AI in the current market. In this way, the products we produce will have relatively better market acceptance.
HopefulThe video field model will be launched by the end of this year
Red Star Capital Bureau: How intelligent can the company's big model be now? What state do you hope to achieve in future iterations?
Zhang Yi:Our self-developed large model has a relatively high level of intelligence. For example, if I upload a picture as a novice, I don’t need to type in the prompt words myself. The platform will automatically analyze some characteristics of the picture in AI semantic understanding, such as its color matching and perspective. The platform will automatically fill in for me. I only need to modify it on this basis and finally click to generate it. It is equivalent to automatically matching users and finally finding pictures that meet their imaginations.
Haiyi AI
In the future, we hope to be able to describe a scene or character through voice. The Haiyi platform can not only achieve this, but also further mutate the image, or modify the details, etc.
After completing some basic functions in the field of images, many companies did not continue to delve deeper into the field. In our opinion, this shows that the foundation is not stable enough.
In the video field, we are currently developing the underlying large model with Huawei and other platforms, aiming to create diversified AI content generation, and we are expected to launch models in the fields of text-generated video and image-generated video by the end of this year or early next year. However, we will not blindly transform, we will make a certain product bigger and better before taking on the second product, and "video is just a singularity in the AI universe, not the end."
Red Star Capital Bureau: What challenges still exist at present?
Zhang Yi:There will still be some pressure on computing power. With more computing power, we can accelerate our expansion and realize more functions.
The models on our platform are very rich. In addition to self-developed models, there are many excellent models that can be connected to the ecosystem. We have cooperated with some of the world's most famous model websites. That is to say, users do not need to download and install them. They can be used directly on our platform. This is our advantage, but it also means a huge consumption of computing power.
In addition, our platform has many free trial functions and a large number of users, which also require more computing power support.
High-end talents can provide more development support for enterprises
Red Star Capital Bureau: Companies choose to develop in Chengdu. What advantages do you think it has in artificial intelligence?
Zhang Yi:I think that in the past two years, the Chengdu government has been increasing its support for high-tech industries and innovation projects, which can accelerate our R&D cycle and provide a good innovation environment for enterprises.
Chengdu also has great urban attributes. It not only has a long history, but also has strong development in internationalization, entertainment, tourism, etc. At the same time, as an important science and technology and industrial center in western China, Chengdu has formed a relatively complete AI industry chain, with upstream chip design, data storage, sensors, etc., midstream companies like AI model training or algorithm development, and downstream intelligent manufacturing, smart cities, medical health, and even entertainment industries. This provides good support for the development of AI companies.
In terms of talent, Chengdu also has universities such as Sichuan University and University of Electronic Science and Technology of China that can continuously supply professional talents. At present, we are still eager for some high-end talents, especially experts at the doctoral or tutor level. Their professional knowledge and research capabilities can help enterprises make new breakthroughs in technological innovation and market competition, and can optimize or support products, model algorithms and even academics.
Red Star News reporter Wang Tian
Editor: Deng Lingyao
(Download Red Star News and get a reward for reporting!)
