search
HomeTechnology peripheralsAIWhen Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Recently, the field of AI video technology has attracted much attention, especially the Sora video generation large model launched by OpenAI, which has caused widespread discussion. At the same time, in the field of video editing, large-scale AI models such as Agent have also shown strong strength.

Although natural language is used to handle video editing tasks, users can directly express their intentions without manual operations. However, most current video editing tools still require a lot of manual operations and lack personalized contextual support. This results in users needing to solve complex video editing problems on their own.

The key is how to design a video editing tool that can act as a collaborator and continuously assist users during the editing process? In this article, researchers from the University of Toronto, Meta (Reality Labs Research), and the University of California, San Diego propose to use the multi-functional language capabilities of large language models (LLM) for video editing, and explore the future video editing paradigm, thereby Reduce frustration with the manual video editing process.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

  • Paper title: LAVE: LLM-Powered Agent Assistance and Language Augmentation for Video Editing
  • Paper address: https://arxiv.org/pdf/2402.10294.pdf

The researcher developed a video editing tool called LAVE, which integrates Multiple language enhancements provided by LLM. LAVE introduces an intelligent planning and execution system based on LLM, which can interpret the user's free-form language instructions, plan and execute related operations to achieve the user's video editing goals. This intelligent system provides conceptual assistance, such as creative brainstorming and video footage overviews, as well as operational assistance, including semantic-based video retrieval, storyboarding, and clip trimming.

In order to smoothly operate these agents, LAVE uses a visual language model (VLM) to automatically generate a language description of video visual effects. These visual narratives allow LLM to understand the video content and use their language capabilities to assist users in editing. In addition, LAVE provides two modes of interactive video editing, namely agent assistance and direct operation. This dual mode provides users with greater flexibility to improve the agent's operation as needed.

As for the editing effect of LAVE? The researchers conducted a user study with 8 participants, including novice and experienced editors, and the results showed that participants could use LAVE to create satisfactory AI collaborative videos.

It is worth noting that 5 of the six authors of this study are Chinese, including the first author, Bryan Wang, a doctoral student in computer science at the University of Toronto, Meta research scientists Yuliang Li, Zhaoyang Lv and Yan Xu and Haijun Xia, assistant professor at the University of California, San Diego.

LAVE User Interface (UI)

Let’s first look at the system design of LAVE, as shown in Figure 1 below.

LAVE's user interface consists of three main components, as follows:

  • Language enhanced video library, displayed with automatic generation Video clips described in the language;
  • Video clipping timeline, including the main timeline for editing;
  • Video clipping agent, Enables users to interact with a conversational agent and get help.

#The design logic is this: when the user interacts with the agent, the message exchange will be displayed in the chat UI. When doing so, the agent makes changes to the video library and clip timeline. In addition, users can directly operate the video library and timeline using the cursor, similar to traditional editing interfaces.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Language enhancement video library

The functions of the language enhancement video library are as follows As shown in Figure 3.

Like traditional tools, this feature allows clip playback but provides visual narrative, i.e. automatically generated text descriptions for each video, including semantic titles and summaries. The titles help understand and index the clips, and the summaries provide an overview of each clip's visual content, helping users form the storyline of their editing project. A title and duration appear below each video.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Additionally, LAVE enables users to search for videos using semantic language queries, and the retrieved videos are displayed in a video library and sorted by relevance. This function must be performed by the Clip Agent.

Video Clip Timeline

After selecting a video from the video library and adding it to the Clip Timeline , they will be displayed on the video clip timeline at the bottom of the interface, as shown in Figure 2 below. Each clip on the timeline is represented by a box and displays three thumbnail frames: the start frame, the middle frame, and the end frame.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

In the LAVE system, each thumbnail frame represents one second of material in the clip. As with the video gallery, a title and description are provided for each clip. The clip timeline in LAVE has two key features, clip sorting and trimming.

Sequencing clips on the timeline is a common task in video editing and is important for creating a coherent narrative. LAVE supports two sorting methods. One is LLM-based sorting, which uses the storyboard function of the video clip agent. The other is manual sorting, which is sorted by direct user operation. Drag and drop each video box to set the order in which clips appear.

Trimming is also important in video editing to highlight key segments and remove excess content. While trimming, the user double-clicks on the clip in the timeline, which opens a pop-up window showing one-second frames, as shown in Figure 4 below.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Video Clip Agent

LAVE The Video Clip Agent is a chat-based component that facilitates interaction between users and LLM-based agents. Unlike command line tools, users can interact with agents using free-form language. The agent leverages LLM's linguistic intelligence to provide video editing assistance and provide specific responses to guide and assist the user throughout the editing process. LAVE's agent assistance functionality is provided through agent operations, each of which involves performing a system-supported editing function.

Overall, LAVE offers features that cover the entire workflow from ideation and pre-planning to actual editing operations, but the system does not mandate a strict workflow. Users have the flexibility to leverage subsets of functionality that match their editing goals. For example, users with a clear editorial vision and a clear storyline may bypass the ideation phase and jump straight into editing.

Back-end system

This study uses OpenAI’s GPT-4 to illustrate the design of the LAVE back-end system, which mainly includes agent design, Implement two aspects of editing functions driven by LLM.

Agent Design

This research leverages the multi-language capabilities of LLM (i.e. GPT-4) (including Reasoning, planning, and storytelling) builds the LAVE agent.

LAVE agent has two states: planning and execution. This setup has two main benefits:

  • Allows the user to set high-level goals that contain multiple actions, eliminating the need to detail each individual action like traditional command line tools .
  • Before execution, the agent will present the plan to the user, providing opportunities for modification and ensuring that the user has full control over the operation of the agent. The research team designed a back-end pipeline to complete the planning and execution process.

#As shown in Figure 6 below, the pipeline first creates an action plan based on user input. The plan is then converted from a textual description into function calls, and the corresponding functions are then executed.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Implement LLM driven editing function

In order to help users complete the video For editing tasks, LAVE mainly supports five functions driven by LLM, including:

  • Material Overview
  • Creative Brainstorming
  • Video Retrieval
  • Storyboard
  • Clip Trim

The first four of them can be accessed through the agent (Figure 5), while the clip The trim feature is available by double-clicking on a clip in the timeline, which opens a pop-up window showing one-second frames (Figure 4).

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Among them, language-based video retrieval is implemented through the vector storage database, and the rest is implemented through LLM prompt engineering. All features are built on automatically generated verbal descriptions of the original footage, including titles and summaries for each clip in the video library (Figure 3). The research team calls the text descriptions of these videos visual narration.

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors

Interested readers can read the original text of the paper to learn more about the research content.

The above is the detailed content of When Sora detonated the video generation, Meta began to use Agent to automatically cut the video, led by Chinese authors. For more information, please follow other related articles on the PHP Chinese website!

Statement
This article is reproduced at:51CTO.COM. If there is any infringement, please contact admin@php.cn delete
ai合并图层的快捷键是什么ai合并图层的快捷键是什么Jan 07, 2021 am 10:59 AM

ai合并图层的快捷键是“Ctrl+Shift+E”,它的作用是把目前所有处在显示状态的图层合并,在隐藏状态的图层则不作变动。也可以选中要合并的图层,在菜单栏中依次点击“窗口”-“路径查找器”,点击“合并”按钮。

ai橡皮擦擦不掉东西怎么办ai橡皮擦擦不掉东西怎么办Jan 13, 2021 am 10:23 AM

ai橡皮擦擦不掉东西是因为AI是矢量图软件,用橡皮擦不能擦位图的,其解决办法就是用蒙板工具以及钢笔勾好路径再建立蒙板即可实现擦掉东西。

谷歌超强AI超算碾压英伟达A100!TPU v4性能提升10倍,细节首次公开谷歌超强AI超算碾压英伟达A100!TPU v4性能提升10倍,细节首次公开Apr 07, 2023 pm 02:54 PM

虽然谷歌早在2020年,就在自家的数据中心上部署了当时最强的AI芯片——TPU v4。但直到今年的4月4日,谷歌才首次公布了这台AI超算的技术细节。论文地址:https://arxiv.org/abs/2304.01433相比于TPU v3,TPU v4的性能要高出2.1倍,而在整合4096个芯片之后,超算的性能更是提升了10倍。另外,谷歌还声称,自家芯片要比英伟达A100更快、更节能。与A100对打,速度快1.7倍论文中,谷歌表示,对于规模相当的系统,TPU v4可以提供比英伟达A100强1.

ai可以转成psd格式吗ai可以转成psd格式吗Feb 22, 2023 pm 05:56 PM

ai可以转成psd格式。转换方法:1、打开Adobe Illustrator软件,依次点击顶部菜单栏的“文件”-“打开”,选择所需的ai文件;2、点击右侧功能面板中的“图层”,点击三杠图标,在弹出的选项中选择“释放到图层(顺序)”;3、依次点击顶部菜单栏的“文件”-“导出”-“导出为”;4、在弹出的“导出”对话框中,将“保存类型”设置为“PSD格式”,点击“导出”即可;

ai顶部属性栏不见了怎么办ai顶部属性栏不见了怎么办Feb 22, 2023 pm 05:27 PM

ai顶部属性栏不见了的解决办法:1、开启Ai新建画布,进入绘图页面;2、在Ai顶部菜单栏中点击“窗口”;3、在系统弹出的窗口菜单页面中点击“控制”,然后开启“控制”窗口即可显示出属性栏。

GPT-4的研究路径没有前途?Yann LeCun给自回归判了死刑GPT-4的研究路径没有前途?Yann LeCun给自回归判了死刑Apr 04, 2023 am 11:55 AM

Yann LeCun 这个观点的确有些大胆。 「从现在起 5 年内,没有哪个头脑正常的人会使用自回归模型。」最近,图灵奖得主 Yann LeCun 给一场辩论做了个特别的开场。而他口中的自回归,正是当前爆红的 GPT 家族模型所依赖的学习范式。当然,被 Yann LeCun 指出问题的不只是自回归模型。在他看来,当前整个的机器学习领域都面临巨大挑战。这场辩论的主题为「Do large language models need sensory grounding for meaning and u

强化学习再登Nature封面,自动驾驶安全验证新范式大幅减少测试里程强化学习再登Nature封面,自动驾驶安全验证新范式大幅减少测试里程Mar 31, 2023 pm 10:38 PM

引入密集强化学习,用 AI 验证 AI。 自动驾驶汽车 (AV) 技术的快速发展,使得我们正处于交通革命的风口浪尖,其规模是自一个世纪前汽车问世以来从未见过的。自动驾驶技术具有显着提高交通安全性、机动性和可持续性的潜力,因此引起了工业界、政府机构、专业组织和学术机构的共同关注。过去 20 年里,自动驾驶汽车的发展取得了长足的进步,尤其是随着深度学习的出现更是如此。到 2015 年,开始有公司宣布他们将在 2020 之前量产 AV。不过到目前为止,并且没有 level 4 级别的 AV 可以在市场

ai移动不了东西了怎么办ai移动不了东西了怎么办Mar 07, 2023 am 10:03 AM

ai移动不了东西的解决办法:1、打开ai软件,打开空白文档;2、选择矩形工具,在文档中绘制矩形;3、点击选择工具,移动文档中的矩形;4、点击图层按钮,弹出图层面板对话框,解锁图层;5、点击选择工具,移动矩形即可。

See all articles

Hot AI Tools

Undresser.AI Undress

Undresser.AI Undress

AI-powered app for creating realistic nude photos

AI Clothes Remover

AI Clothes Remover

Online AI tool for removing clothes from photos.

Undress AI Tool

Undress AI Tool

Undress images for free

Clothoff.io

Clothoff.io

AI clothes remover

AI Hentai Generator

AI Hentai Generator

Generate AI Hentai for free.

Hot Article

R.E.P.O. Energy Crystals Explained and What They Do (Yellow Crystal)
2 weeks agoBy尊渡假赌尊渡假赌尊渡假赌
R.E.P.O. Best Graphic Settings
2 weeks agoBy尊渡假赌尊渡假赌尊渡假赌
R.E.P.O. How to Fix Audio if You Can't Hear Anyone
2 weeks agoBy尊渡假赌尊渡假赌尊渡假赌

Hot Tools

Safe Exam Browser

Safe Exam Browser

Safe Exam Browser is a secure browser environment for taking online exams securely. This software turns any computer into a secure workstation. It controls access to any utility and prevents students from using unauthorized resources.

SublimeText3 Mac version

SublimeText3 Mac version

God-level code editing software (SublimeText3)

Atom editor mac version download

Atom editor mac version download

The most popular open source editor

PhpStorm Mac version

PhpStorm Mac version

The latest (2018.2.1) professional PHP integrated development tool

VSCode Windows 64-bit Download

VSCode Windows 64-bit Download

A free and powerful IDE editor launched by Microsoft