BEIJING, February 21 (TMTPost)-- ByteDance Ltd. played down its artificial intelligence (AI) tool with the similar feature as Sora, OpenAI’s first AI video generation model.

Credit：Visual China

Prior to release of Sora, ByteDance quietly launched Boximator, a video model product that can generate motions of people or objects in a video through text control, earlier this month, Chinese digital news media outlet Jiemian learned from people familiar with the matter. As a research project on technical methods for controlling object motion in video generation, Boximator is currently not available for application as a completely developed product, and there is still a big gap on picture quality, fidelity, and video length between the project and leading overseas video generation models, a ByteDance insider responded.

ByteDance was reported to start implementing its AI strategy last year and established a new department called Flow in November to focus on businesses-powered by AI innovation. A person close to ByteDance told Jiemian that ByteDance founder and former chief executive Zhang Yiming has devoted all his time to AI over the past year. Zhang’s allocation of energy for AI demonstrates the importance that ByteDance attaches to AI business.

The aforementioned person familiar with the matter revealed that ByteDance is drawing top talent from the entire company to support its AI business. Zhu Jun, the former product manager at TikTok and current vice president of product and strategy at ByteDance, has been appointed as the product manager of Flow. In addition, some core product personnel at TikTok have also been transferred to Flow,, according to the source.

To the date,ByteDance’s Flow department has rolled out three AI conversational products, including Dou Bao, Kou Zi, and Cici. Dou Bao is a chatbot product that can perform various tasks such as question answering, text generation, and language translation. It can also provide personalized services by adapting responses based on user needs and context. Kou Zi is an all-in-one AI Bot development platform that allows users to quickly build various types of question-answering bots based on AI models, regardless of their programming background, handling simple Q&A and complex dialogues.

Compared to other Chinese internet giants like Alibaba and Baidu, ByteDance has taken a relatively low-profile approach in AI and large scale models. Currently, ByteDance’s AI products such as Dou Bao, Kou Zi, and Cici have not been heavily promoted or widely marketed, and ByteDance has not officially disclosed its research direction and roadmap in the field of AI.

ByteDance published a research paper on Boximator in the beginning of this month, calling a new approach for fine-grained motion control. Boximator introduces a simple yet powerful approach for motion specification. Users first select objects in a reference image by drawing boxes around them. They can then define an object's ending position or entire motion path across frames using additional boxes and lines. This visually-grounded technique avoids the need for verbally describing desired motions. Boximator functions as a plug-in for existing video diffusion models.  Its training process preserves the base model's knowledge by freezing the original weights and training only the control module. 

Sora, the model released last Thursday, has redefined the standards for AI video generation models. Sora has increased the video length from five to fifteen seconds to one minute, which can fully meet the needs of creating short videos. According to OpenAI, if necessary, it’s a piece of cake to make videos longer than one minute. It can generate multiple shots and each shot maintains consistency in character roles and visual style. It can generate videos from text prompts, and also support video-to-video editing. It can also generate high-quality images. It can even collage together completely different videos to make them merge into one coherent piece.