UltraVideo: High-Quality UHD Video Dataset with Comprehensive Captions
Abstract
A high-quality UHD-4K text-to-video dataset, UltraVideo, and expanded UltraWan models, are introduced to support advanced video generation research.
The quality of the video dataset (image quality, resolution, and fine-grained caption) greatly influences the performance of the video generation model. The growing demand for video applications sets higher requirements for high-quality video generation models. For example, the generation of movie-level Ultra-High Definition (UHD) videos and the creation of 4K short video content. However, the existing public datasets cannot support related research and applications. In this paper, we first propose a high-quality open-sourced UHD-4K (22.4\% of which are 8K) text-to-video dataset named UltraVideo, which contains a wide range of topics (more than 100 kinds), and each video has 9 structured captions with one summarized caption (average of 824 words). Specifically, we carefully design a highly automated curation process with four stages to obtain the final high-quality dataset: i) collection of diverse and high-quality video clips. ii) statistical data filtering. iii) model-based data purification. iv) generation of comprehensive, structured captions. In addition, we expand Wan to UltraWan-1K/-4K, which can natively generate high-quality 1K/4K videos with more consistent text controllability, demonstrating the effectiveness of our data curation.We believe that this work can make a significant contribution to future research on UHD video generation. UltraVideo dataset and UltraWan models are available at https://xzc-zju.github.io/projects/UltraVideo.
Community
GLOBAL SETTINGS
Cinematography & Motion: Handheld shot, natural shallow depth of field. Camera starts static, then smoothly descends, followed by a rapid pull-back and a slow, steady horizontal panoramic pan. Slow-paced, continuous takes with no immediate cuts until the scene change.
Style & Ambiance: Bright sunny daylight. Documentary-style, real-world aesthetic. Clear vibrant blue sky with sparse white clouds.
Subject Details: A hand wearing a clean white glove holds US currency. The currency is a thick stack of twenty-dollar bills, secured by two yellow rubber bands.
Ambient Noise: No specific discernible ambient noise.
CHRONOLOGICAL TIMELINE
[00:00-00:05.5] Action: A hand in a white glove holds a thick stack of twenty-dollar US bills, secured by two rubber bands. The hand slowly lowers the money downwards. Context: The immediate background is blurry, showing dry grass and dark ground cover. A TikTok watermark "@nycmoneyhunt" is present in the bottom left.
[00:05.5-00:08] Action: The money is gently placed and partially hidden amidst tall, dry green and brown grass. The camera follows the money into the grass, becoming slightly obscured by the blades. Context: A close-up view of the base of the grass, showing soil and small debris on the ground.
[00:08-00:10] Action: The camera rapidly pulls back and pans smoothly upward and to the right, revealing a wider outdoor scene. Context: A paved pathway is in the foreground, flanked by tall grass. In the midground, commercial buildings with prominent "TOP POT" and "WINK" signs are visible, alongside a Burger King sign. A tall street lamp stands. Parked cars are visible in a lot. The sky is clear and blue.
[00:10-00:12] Action: The camera continues its slow, steady panoramic pan to the right. Context: More of the parking lot, additional commercial buildings including one with a "MACY'S" sign in the distance, and trees displaying early autumn foliage are revealed.
[00:12-00:16] Action: The camera slowly pans back to the left, completing a broad panoramic sweep of the urban commercial area. Context: The paved pathway, patches of grass, various commercial buildings, a street with moving cars, and the expansive blue sky remain in view.
Get this paper in your agent:
hf papers read 2506.13691 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 2
Datasets citing this paper 4
Spaces citing this paper 4
Collections including this paper 0
No Collection including this paper