Ant-invested Shengshu Technology has completed tens of millions of yuan in financing and is testing billions of large models internally

Source: Titanium Media

Author: Lin Zhijia

Image source: Generated by Unbounded AI

**Titan Media App exclusively learned that on August 9, **Generative AI company Beijing Shengshu Technology Co., Ltd. (referred to as "Shengshu Technology") announced the completion of tens of millions of yuan in angel + round financing. This round was funded by Jinqiu Fund Investment, funds will be mainly used for algorithm research and development, product development and team expansion.

Earlier in June, Shengshu Technology announced the completion of angel round financing of nearly 100 million yuan from Ant Group, BV Baidu Ventures, and Zhuoyuan Capital, with a valuation of 100 million US dollars after the first round of financing. In less than two months, Shengshu Technology once again completed the second round of investment in a row. (For details, please refer to the previous article of Titanium Media App: "Ant and Baidu jointly invest in AIGC track, Shengshu Technology is valued at US$100 million after the first round of financing")

** Shengshu Technology CEO Tang Jiayu told Titanium Media App that ** company is planning to create a large model with tens of billions of parameters. Compared with other teams, the core barrier of Shengshu lies in the original advanced nature of the multi-modal large-scale model architecture and the continuous innovation ability of the underlying algorithm. The team has developed an open-source large-scale model Unidiffuser.

**"**Currently in the Vincent diagram scene, our effect has reached the effect of the latest version of the basic model of Stable Diffusion, but there is still a certain gap compared with Midjourney, mainly in terms of aesthetics. We expect to reach the latest version of Midjourney by the end of the year effect." Tang Jiayu said.

**It is reported that ** Shengshu Technology was established in March 2023. It is the world's leading provider of generative AI infrastructure and applications. Joint incubation was established, dedicated to creating a controllable multi-modal general-purpose large model. Tang Jiayu previously served as the vice president of Ruilai Wisdom.

In terms of products, In March of this year, the Shengshu Technology team released an open-source multi-modal large model Unidiffuser with 950 million parameters, using the underlying unified network framework U-ViT based on the Transformer architecture to realize the communication between graphic and text modes. Arbitrary cross-modal generation.

Tang Jiayu told Titanium Media App that at present, the industry is mainly focused on solving two core problems: one is to enable the model to understand and align data of various modalities; the other is to enable the model to generate content of various modalities. Most methods still rely on combining multiple models to handle multimodal data, but Unidiffuser leverages the U-ViT architecture and successfully integrates text and images into one framework, and its closed-source version further integrates 3D and video content .

Currently, the research and development core of the Shengshu Technology team mainly has two aspects: The first is to independently develop industrial-level large models with higher parameters based on the Unidiffuser core architecture design, and integrate 3D, For more modal data such as video, at this stage, a general large-scale model internal test with a scale of tens of billions of data and billions of parameters has been realized, supporting functions such as text and image generation, 3D generation, and video element editing;** The second is **Based on the underlying general model capabilities, research and develop an "AI toolset" for content creation scenarios such as image creation and 3D asset creation. The product will be launched before the end of this year.

** In terms of commercialization, Tang Jiayu revealed to Titanium Media App that the commercialization path of Shengshu Technology is divided into two parts: ** On the one hand, it is to develop end-user-oriented applications, such as image creation assistants and 3D asset generation platforms, which are charged in the form of subscriptions, etc. ; On the other hand, it provides invocation model services for B-end organizations.

"We hope to help individual users solve their image creation needs in subdivided scenarios while achieving high-quality results. Currently, one of our internal beta products is positioned as a personal image creation assistant, focusing on more general underlying image creation and editing needs, Truly follow the words, allowing users to create ideal images with extremely simple instructions and interactions. The difficulty involves technological breakthroughs in the controllable generation of the underlying model, as well as the needs and intentions of ordinary users in image creation and editing deep understanding." Tang Jiayu said.

Tang Jiayu believes that compared with language-based large-scale models, domestic and foreign research in the direction of multi-modal large-scale models is in the early stage of exploration, and there is no obvious gap. Issues such as the unification and generalization of large-scale model instructions and fine-tuning of large-scale models are still in the early stage of technical research and verification in the world, so the gap between domestic and foreign countries is not obvious. And in some work directions, China is already ahead of foreign countries.

**It is worth noting that **Shengshu Technology’s investor in this round, “Jinqiu Fund”, has a close relationship with ByteDance. The fund was established in March 2022 by Janna Yang, the head of the original ByteDance financial investment. Most of the founding members came from the original ByteDance investment team, top domestic venture capital institutions and well-known industrial companies. Long-term investment. Earlier, Yang Jie also worked for Sequoia Capital, Shenzhen Venture Capital, Tongchuang Albert, etc.

At present, Jinqiu Fund mainly focuses on AI and going overseas, and has invested in overseas e-commerce service platform Jet Commerce, etc., and Yang Jie has worked in many companies such as Future Robot, Sharkfit, and Zero Rhino Technology.

**Responding to this financing, Zheng Xiaochao, vice president of Jinqiu Fund Investment, said that large models and generative AI are leading the biggest new wave of technology at present, giving birth to new productivity tools and application products, but their core The driving force still comes from the innovation of the underlying model. The Shengshu Technology team not only focuses on underlying technological innovation, but also has full-stack R&D capabilities such as the backbone network of deep generative models, high-speed sampling, and controllable generation, and has the profound technical strength to train the world's largest diffusion model from 0 to 1. At the same time, at the level of commercial product exploration, the Shengshu Technology team also has strong engineering capabilities and product innovation capabilities.

Tang Jiayu said that generative AI technology will allow everyone to better express and create. When imagination is no longer limited, production efficiency will be greatly improved, and traditional production relationships and human-computer interaction models will also be reshaped. Future Shengshu Technology will continue to polish general-purpose large-scale models and application products, and is committed to improving the creativity and productivity of all human beings by continuously liberating imagination. (This article is the first release of Titanium Media App, author｜Lin Zhijia)