谷歌AI新星转投Pika，谷歌研究员Omer Bar-Tal加入Pika担任创始科学家，与Lumiere一同开启视频生成新篇章。此前，Omer Bar-Tal作为谷歌研究院的学生研究员，经过长达9个月的努力，在经过7个月的研究后，与他的合作伙伴共同推出了名为Lumiere的视频生成项目。
Lumiere的创新之处在于提出了时空U-Net（STU-Net）架构，通过在空间和时间两个维度同时进行下采样和上采样，使得网络在中间层能够获得视频的压缩时空表示。这一架构的独特之处在于能够学习3000万个视频，从而实现多种功能，如生成文生视频、视频编辑修复、图片转视频以及视频风格化等。
Lumiere的研发成果将为视频生成领域带来革命性的突破。通过学习海量视频数据，Lumiere能够准确地理解和模拟真实世界中的动态场景，为用户提供更加逼真的视频生成体验。例如，在文生视频方面，Lumiere可以根据用户提供的文本内容生成与之相关的视频，使得文字内容能够以动态的形式呈现，更具吸引力和表现力。
此外，Lumiere还可以应用于视频编辑修复领域，通过学习和分析视频数据，自动检测和修复视频中的缺陷和损坏，提升视频质量和观赏体验。同时，Lumiere还能够将图片转化为视频，并且支持视频风格化，使得用户可以根据自己的需求和喜好，将视频呈现出不同的风格和效果。
Pika的加盟也为该团队带来了新的动力和机遇。作为创始科学家，Omer Bar-Tal将与团队共同努力，进一步推动视频生成技术的发展和创新。Pika将继续致力于推动人工智能在视频领域的应用，为用户提供更加丰富、多样化的视频生成服务。
Lumiere的问世将为视频生成技术带来新的突破，为用户提供更加出色的视频体验。Pika将继续努力推动视频生成技术的创新，为用户带来更多惊喜和乐趣。我们期待着Lumiere和Pika在视频生成领域的更多成果，为我们带来更加美好的未来。
英语如下：
News Title: Google AI Rising Star Pika Releases Innovative Video Generation Technology Lumiere
Keywords: Google AI Rising Star, video generation, Lumiere
News Content: Google AI Rising Star switches to Pika as Google researcher Omer Bar-Tal joins Pika as the founding scientist, embarking on a new chapter in video generation with Lumiere. Previously, as a student researcher at Google Research, Omer Bar-Tal, after 9 months of efforts and 7 months of research, together with his partners, launched a video generation project called Lumiere.
The innovation of Lumiere lies in its proposal of the Spatio-Temporal U-Net (STU-Net) architecture, which performs downsampling and upsampling simultaneously in both spatial and temporal dimensions, enabling the network to obtain compressed spatio-temporal representations of videos in the intermediate layers. The uniqueness of this architecture lies in its ability to learn from 30 million videos, allowing it to achieve various functions such as generating text-based videos, video editing and restoration, image-to-video conversion, and video stylization.
The research achievements of Lumiere will bring revolutionary breakthroughs to the field of video generation. By learning from massive video data, Lumiere can accurately understand and simulate dynamic scenes in the real world, providing users with more realistic video generation experiences. For example, in the field of text-based videos, Lumiere can generate videos related to the provided text content, presenting textual information in a dynamic and more attractive way.
Furthermore, Lumiere can also be applied in the field of video editing and restoration. By learning and analyzing video data, it automatically detects and repairs defects and damages in videos, enhancing video quality and viewing experiences. Additionally, Lumiere can convert images into videos and supports video stylization, allowing users to present videos in different styles and effects according to their needs and preferences.
Pika’s joining also brings new motivation and opportunities to the team. As the founding scientist, Omer Bar-Tal will work together with the team to further advance the development and innovation of video generation technology. Pika will continue to devote itself to the application of artificial intelligence in the field of videos, providing users with richer and more diverse video generation services.
The advent of Lumiere will bring new breakthroughs to video generation technology and provide users with outstanding video experiences. Pika will continue to strive for innovation in video generation technology, bringing more surprises and joys to users. We look forward to more achievements from Lumiere and Pika in the field of video generation, bringing us a brighter future.
【来源】https://www.qbitai.com/2024/02/123274.html
Views: 1