上海的陆家嘴

史上首个实时AI视频生成技术突破:DiT通用,速度提升10.6倍

近日,新加坡国立大学的尤洋团队在AI视频生成领域取得了重大突破,推出了名为Pyramid Attention Broadcast(PAB)的新技术。该技术基于DiT(一种视频生成技术),能够实现实时输出,无需训练即可生成高质量视频。此项技术打破了视频生成速度的纪录,帧率高达21.6 FPS,加速比达到了惊人的10.6倍。

传统的视频生成方法往往面临计算量大、处理时间长的问题,限制了AI视频生成技术的实际应用。而PAB技术的出现,通过减少冗余注意力计算,实现了视频生成的快速和高质量。无论是对基于DiT的现有模型,还是未来可能出现的模型,PAB技术都能提供加速,让它们具备实时生成的能力。

自今年起,OpenAI的Sora和其他基于DiT的视频生成模型在AI领域掀起了一波浪潮。然而,与图像生成相比,视频生成的关注点一直集中在质量上,忽视了模型的推理速度。尤洋团队的这项新技术填补了这一空白,为生成式AI应用开辟了新的道路。

据了解,PAB技术在Open-Sora、Open-Sora-Plan和Latte等流行视频生成模型上的表现尤为出色,无需牺牲质量即可实现高速生成。这一技术的突破性进展将极大地推动AI视频生成技术的发展,对于未来的影视制作、在线视频平台等领域将产生深远影响。

随着技术的不断进步,我们期待AI视频生成技术在未来能够带来更多创新和突破,为观众带来更加丰富多样的视觉体验。

英语如下:

News Title: “PAB Technology Boosts DiT Video Generation Speed by 10.6 Times, Real-Time Lossless Output”

Keywords: 1. Real-Time AI Video Generation Technology

News Content:

A groundbreaking breakthrough in real-time AI video generation technology: DiT achieves a speed increase of 10.6 times

Recently, the You Yang team from the National University of Singapore made a significant breakthrough in the field of AI video generation, introducing a new technology called Pyramid Attention Broadcast (PAB). Based on DiT (a video generation technology), this technology can achieve real-time output and generate high-quality videos without the need for training. This technology has broken the record for video generation speed, with a frame rate of up to 21.6 FPS, and an impressive speed increase of 10.6 times.

Traditional video generation methods often face problems of large computational requirements and long processing times, which limit the practical application of AI video generation technology. The emergence of PAB technology achieves fast and high-quality video generation by reducing redundant attention calculations. PAB technology can provide acceleration for both existing models based on DiT and future models, enabling them to generate videos in real-time.

Since the beginning of this year, OpenAI’s Sora and other video generation models based on DiT have sparked a wave in the AI field. However, compared to image generation, video generation has focused primarily on quality, neglecting the model’s inference speed. The new technology from the You Yang team fills this gap and opens up new avenues for generative AI applications.

It is understood that PAB technology performs particularly well on popular video generation models such as Open-Sora, Open-Sora-Plan, and Latte, achieving high-speed generation without sacrificing quality. This breakthrough in technology will greatly promote the development of AI video generation technology and have far-reaching impacts on future film production, online video platforms, and other fields.

With continuous technological advancements, we look forward to the future of AI video generation technology bringing more innovation and breakthroughs, providing viewers with richer and more diverse visual experiences.

【来源】https://www.jiqizhixin.com/articles/2024-06-28-8

Views: 3

发表回复

您的邮箱地址不会被公开。 必填项已用 * 标注