ByteDance Is Actively Developing Multiple AI Products
On February 28th, Chinese media Jiemian obtained information from multiple sources indicating that ByteDance is actively engaged in the secret development of various products in the field of AI large models. These products include multimodal digital human offerings, as well as AI-generated image and video solutions. A source revealed that they witnessed a demonstration of ByteDance’s multimodal digital human product late last year.
Moreover, ByteDance’s video editing platform, CapCut, established a secluded team several months ago to discreetly work on AI projects. Currently, this team remains in a highly confidential phase, with their developed products yet to be released.
A knowledgeable individual close to ByteDance noted that founder Zhang Yiming dedicated a significant portion of his focus last year to AI endeavors, underscoring the company’s commitment to AI initiatives.
ByteDance is currently employing a comprehensive approach in the development of AI large model-related products, spanning from the model layer to the application layer and adopting a multifaceted strategy. In the realm of fundamental large models, ByteDance introduced its first large language model, “DouBao,” and the multimodal large model, BuboGPT, in August of last year. Additionally, its Douyin Yunque large model, which was among the first to be filed under the “Interim Measures for the Management of Generative Artificial Intelligence Services,” is now accessible to the public. Recently, ByteDance also unveiled the SDXL-Lightning open model for generating high-quality, high-resolution images in just 2 to 4 steps, significantly boosting the generation speed.
At the AI application layer, ByteDance established the new AI department, Flow, in November of last year, and has since launched three AI conversation products: DouBao, Kouzi, and Cici. Within the fundamental large model layer, ByteDance has strategically positioned itself in both language and image modalities, with both teams reporting to TikTok’s tech lead, Zhu Wenjia.
Another source with insights into ByteDance’s operations mentioned that the company is currently facing notable pressure in its large model strategy due to oscillations between self-research and investment approaches last year.
Initially considering entering the large model arena through investments, ByteDance deliberated investing in companies like MiniMax and Jieyue Xingchen. However, in June of last year, ByteDance opted to discontinue investing in external large model companies and instead focus on internal research and development. The efficacy of ByteDance’s decision to entirely forego investments in favor of in-house research warrants further evaluation, particularly in light of Alibaba‘s recent substantial investment in Moon’s Dark Side.
Despite challenges, individuals familiar with ByteDance’s large model initiatives stress that it is premature to dismiss the company’s positioning in the AI large model sector. Among ByteDance’s product portfolio, CapCut appears to be the prime candidate for advancing ByteDance’s AI large model implementations.
An analysis by one insider highlights CapCut’s role as a video creation tool situated at the forefront of content creation, progressing towards AI applications, particularly in the realm of AI-generated videos. Furthermore, content generated within CapCut seamlessly integrates with Douyin, enabling creators to leverage ByteDance’s AI-generated videos and multimodal digital human products for content creation, presenting substantial creative opportunities.
Before the Chinese New Year, former Douyin Group CEO Zhang Nan stepped down from his position, expressing intentions to concentrate on CapCut’s development moving forward. This move has been interpreted by industry insiders as ByteDance’s strategic push to strengthen its foothold in AI-generated video content through CapCut. An insider noted, “CapCut’s immediate focus lies in addressing the availability of creative materials, encompassing various personalized elements within the video and animation categories.”
Sign in with google
