TikTok parent company launches technology to challenge Google

ByteDance 's new tool confidently surpasses all competitors. Photo: ByteDance .

ByteDance - the parent company of TikTok and Douyin, has just officially introduced Seedance 1.0, an artificial intelligence (AI) tool capable of creating videos from text and images. This is considered an important step forward for the Chinese technology group in the race to develop an AI-based content creation platform.

The company recently released a detailed research paper on Seedance 1.0, which is designed to convert simple instructions into high-quality videos without requiring detailed scripts or complex commands. The tool not only handles individual shots, but also combines multiple camera angles, smooth transitions, and ensures character consistency throughout the video.

“We have found a way to separate the spatial and temporal information in videos. This technology uses a unique method to ‘encode’ the location, allowing AI to learn to generate videos from both text and images in the same model. As a result, AI can automatically generate videos with different scenes smoothly,” ByteDance said in the research paper.

ByteDance confidently claims that Seedance 1.0 is superior to existing AI video creation tools on the market, especially in its ability to closely follow user ideas, image sharpness, and naturalness in character movements.

According to Artificial Analysis - a platform specializing in analyzing and evaluating the performance of AI models, Seedance 1.0 has surpassed other video-generating AI tools such as Google's Veo 3, Kuaishou's Kling 2.0 or OpenAI's Sora. This tool shows outstanding performance in both text-to-video and image-to-video tasks.

The company also revealed that Seedance 1.0 was trained on a massive video dataset, collected from publicly available and licensed sources. The training videos went through a rigorous filtering process to remove violent or sensitive content.

Many opinions say that the data source mainly comes from TikTok and Douyin, two platforms operated by ByteDance itself.

Seedance 1.0 training process is divided into several stages: initially learning from rich image and video data, then continuing to learn deep scene transition techniques in different styles.

Humans also play a key role in the training process, as engineers select high-quality videos for the model to learn from. The training loop continues until Seedance 1.0 can pick the optimal result from a large number of videos generated on demand.

Currently, Seedance 1.0 limits the maximum video length to 5 seconds (compared to 8 seconds of Veo 3). However, the outstanding advantage is the fast processing speed: it only takes 41 seconds to create a Full HD video. One downside of Seedance 1.0 is that it does not support automatic audio dubbing like its competitor from Google.

ByteDance plans to soon release this tool for both regular users and professional content creators, serving the needs of producing promotional videos or short content on social networks.

Before Seedance 1.0, ByteDance had developed AI video creation tools such as OmniHuman, Goku, and Jimeng AI. However, Seedance 1.0 is the first product that the company confidently claims can surpass its competitors in terms of AI video creation capabilities.

Source: https://znews.vn/cong-ty-me-tiktok-ra-mat-cong-nghe-thach-thuc-google-post1562025.html