news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Text | Yu Jinglin

Editor |

Sora, which suddenly appeared out of nowhere, shocked the world with its impressive video effects, but also exposed some limitations.

One prominent problem is that Sora can generate up to 60 seconds of video at a time, but its disadvantage is that it cannot guarantee the continuity between multiple generated clips.

Take the work "air head" by Shy Kids displayed on the OpenAI official website as an example. First of all, the specific scene contains some unnecessary elements, such as the unexpected appearance of human face patterns on the balloon.

Second, the properties of the same object (such as the color and type of the balloon) change inconsistently between clips, affecting the coherence and narrative fluency of the video. The model may also generate some redundant information, such as generating additional ropes at the balloon tether.

Source: Studio Huanjing

These problems show that Sora lacks control over the generated content, which affects the narrative logic and coherence of the video.

"Huanjing Studio" hopes to solve these problems. The founder believes that video generation tools such as Sora have expanded the boundaries of creation, demonstrated the possibility of producing a large amount of high-quality content, and then led to the transformation of downstream industry tools. In other words, the Sora-like model has reshaped the video content ecosystem industry, and the production chain will be redefined as the two core stages of intelligent production of raw materials and refined post-processing. The material production is completed by generation tools such as Sora, and the refined post-processing is completed by Huanjing Studio.

Source: Studio Huanjing

Talking about the background of the establishment of "Huanjing Studio", the founder told 36Kr: "At present, most video editing products on the market have high operating thresholds. Although the official tutorials are provided, they are relatively simple. Creators still need to explore on their own if they want to be proficient in using them."

At the beginning of 2023, the team began to conceive and launch a simple and easy-to-use video special effects production software, and completed the technical productization route planning half a year later. In October of the same year, it obtained a seed round of financing of 3 million yuan invested by Yuntian Angel Fund.

Taking character replacement as an example, Studio first analyzes the materials uploaded by users, extracts the character movements, camera trajectories and 3D scene reconstruction, so as to achieve layer-like separation of video elements.

This type of control is different from general video generation technology. Instead, it performs parameter-level control on elements in the video in 3D space to ensure the content and logical coherence of the video clips.

Source: Studio Huanjing

In terms of color grading, "Studio Changing" realizes automatic ambient light matching. Through the editor of Studio Changing, users can easily select any character in the video, and use AI technology to realize automatic recognition, tracking, removal and background filling of characters, eliminating a lot of repetitive manual operations, improving the efficiency of video secondary creation, and allowing people to focus more on creative work.

Video creators are the main target group of Huanjing Studio. The team said that the typical target users of Huanjing Studio will be C-end users with certain creative ability and experience on platforms such as YouTube, Douyin and Bilibili, as well as small B-end customers.