ByteDance's Latest Advances in AI Large Model Domain: Multimodal Digital Humans and AI-generated Graphics, Video Products Emerge

2024-02-29

According to multiple sources, ByteDance is secretly developing products in the field of AI large models, covering various aspects such as multimodal digital humans, AI-generated images, and AI-generated videos. This news has attracted widespread attention in the industry. It is reported that in the second half of last year, insiders witnessed the demo of ByteDance's multimodal digital human product and gave it high praise. At the same time, ByteDance's editing team has secretly formed a closed team dedicated to AI product development. Currently, the team is still in the confidential stage and has not announced any related products. Media outlets have sought confirmation from ByteDance regarding the above news, but as of the time of publication, no response has been received. However, a source close to ByteDance revealed that last year, ByteDance founder Zhang Yiming focused most of his energy on AI, demonstrating the company's emphasis on AI business. In the development of AI large model products, ByteDance has adopted a comprehensive layout strategy, exploring from the model layer to the application layer. In the field of basic large models, in August last year, ByteDance launched the first large language model "DouBao" and the multimodal large model BuboGPT. In addition, its TikTok Yunque large model has also successfully passed the filing of the first batch of "Interim Measures for the Administration of Generative Artificial Intelligence Services" and is open to the public. It is worth mentioning that a few days ago, ByteDance also released the SDXL-Lightning open model for generating images, which can generate high-quality and high-resolution images in a short period of time, increasing the generation speed tenfold. This innovative technology undoubtedly provides strong support for the development of AI-generated image products. At the AI application layer, ByteDance established a new AI department called Flow in November last year and has already launched three AI dialogue products, including DouBao, Koushi, and Cici. In the field of basic large models, ByteDance has made layouts in both language and image modalities, with both teams reporting to TikTok's technical leader, Zhu Wenjia. Although ByteDance faces certain pressure in the layout of large models, several insiders who are familiar with the situation have expressed that it is still too early to completely deny its layout in the field of AI large models. Among them, the editing team is considered one of the most promising products for ByteDance's AI large model implementation. As a video creation tool, the editing team is positioned in the upstream of content creation, moving towards AI-generated videos. In addition, the videos created by the editing team are followed by Douyin, and creators can use ByteDance's AI-generated videos and multimodal digital human products to create content, which provides great imagination space. Before the Spring Festival this year, the former CEO of Douyin Group, Zhang Nan, resigned from the CEO position, stating that he would focus on the development of the editing team in the future. This move has also been interpreted by industry insiders as ByteDance's effort in the direction of AI-generated videos. However, insiders have revealed that ByteDance faced strategic swings in the layout of large models. Initially, the company planned to enter the field of large models through investment and even considered investing in large model companies MiniMax and Jieyue Xingchen. However, in June last year, ByteDance decided to abandon investment in external large model companies and turn to self-research. Whether this decision is correct remains to be further observed. Overall, ByteDance's layout in the field of AI large models has gradually emerged. In the future, with the continuous development of technology and the changing market, we look forward to seeing more innovations and breakthroughs.