本文来源:https://mp.weixin.qq.com/s/xVh0rHVh2a5onDynnqUXdA
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
你好,我是郭震昨天看到几个博主分享Agent制作短视频,我看得手痒。今天我也体验了一把,分别生成了类似老友记那样情景短剧,汽车TVC短片,电商商品宣传,以及AI短剧。Wow,这个效果我没想到,它们已经达到直接发布级水准。很兴奋,于是有了这篇实测文章,感兴趣的可以看看。1 作品展示我创建这些作品,使用的是LibTV,这个视频Agent内置了100+视频Skill,拿来即用:
每个 Skill 都可以理解为一位专业 AI 导演,如下图选中它,一句提示词:
就生成下面的情景喜剧短视频,因为我平时喜欢看情景喜剧学口语,很多地道表达放进具体场景里更容易记住:
除了做了双人电梯对话,我还遥控LibTV当导演,制作了类似老友记那样的四人客厅群戏:
做完看的我挺爽,制作过程中,我还重点观察了下目前先进的 Agent 能不能处理好口语对白、人物反应和连续笑点,结果让我信服了。
过了把制作情景喜剧短视频的瘾后,我还玩了把AI短剧,使用LibTV里的A24 美学Skill,生成了如下精致而克制的短片:
视频时长1分左右,这里面的人物表演、镜头氛围与情绪推进,还带有一些小惊悚,过瘾!
试过真人创作类短视频后,我又玩了把使用卡通形象宣传面包,最近电商场景很多人都在制作此类短视频:
选择LibTV内置的皮克斯动画广告Skill,加载一张喜欢的卡通人物图,如图所示一句简单的提示词,过一会儿就生成如下喜人的卡通短视频,把门店制作面包过程展示出来:
接下来我想试试商业视频里更考验运镜、速度感和产品质感的汽车TVC。对品牌团队来说,这类内容很常用,我想看看只给一句提示词,它能不能完成一条有高级感的汽车广告:
几条片子做下来,我觉得 LibTV 的成片质量确实挺高:选一个 Skill,再给一句需求,它就能交付接近发布状态的完整视频。
接下来,我把背后的制作流程摊开,看看 Agent 具体做了什么,以及遇到视频节奏问题时,我是怎么让它继续修改的。
2 从零制作短视频这节主要介绍LibTV使用的核心逻辑,看完这节你也能做出上面的视频。首先进入 liblib.tv 后,选择开始创作:
进入后选择情景喜剧Skill,一句如下提示词:
大家做短视频最头疼的问题之一就是制作分镜,这个视频Agent都能自动创建:
就连视频中的多个核心锚点也都搞定,如下自动创建4个人物角色三视图:
之后逐步生成每个分镜的视频:最后合并成片,声音和字幕也都包括进去了,详见文章开头的第二个视频。大家看到这就是它家的这个Agent,从任务规划,再到文本,图片,声音,视频,字幕一连串生成。实话讲,我一直都很期待这样的Agent,能端到端成片!熟悉我的读者知道,我在做本地知识库Agent,需要和优秀的产品学习,它家的下面右侧智能体工作面板+搭配左侧AI画布,尤其这个AI画布就对我带来了灵感:智能体边工作,边在左侧自动创建AI工作流,这种可观测性就比较流畅,同时还可以编辑每个节点,这为二次修改提供很大便利。比如点击工作流中的某个分片节点,上面弹出 剪辑,裁剪,高清,解析等二次编辑按钮:
比如在生成这条视频时,我发现人物静止时间稍微长了1-2秒钟,那我就可以使用剪辑按钮,直接cut掉。
除了工作流界面,还支持故事版,可以按照文本和视频顺序逐个预览和二次编辑:
我的体感,感觉工作流+故事版,这种双模式为视频创作后期带来了很大便利。
以上就是LibTV的主要使用逻辑,哦对了,还有一个支持自定义Skill,从而更便捷制作专属你的视频。接下来我来说下怎么做。
3 自定义Skill生成连续AI短剧
还记得上高中那回看好莱坞大片,挺让人震撼的,但没想到现在我也能通过Agent来做了。
如下就是我自己创建的电影级科幻连续短剧导演Skill:
你可以去LibTV,搜索我写的这个skill,直接引用它:
就能像我一样,开启多集AI短剧创作之路。
所以今天我就直接把编写的提示词难度,直接打到零,如下所示,只需要at我写的这个skill,生成科幻短剧这几个字,过一会儿第一集《建立信任》就出来了:
一个45秒的视频,核心人物在这一集里都对齐了。确定好第二集的核心构思后,它会再次按照分镜脚本,人物图片,分镜视频,合成分镜,一连串搞定,第二集视频《共同战斗升级》:
同样的思路,生成第三集《最终决战》:
我仔细看了下三集剧情,故事叙事,人物主角展开的都很顺畅,衔接都很丝滑,给我最大的感受是,LibTV 就是一个全能视频Agent!它能自己拆剧本、定人物、做分镜、生成视频,再把整条流程串起来。最后交付的是一集接近可发布状态的完整视频,还能继续修改,并顺着剧情继续往下拍。最后总结一下
这次我用 LibTV 做了情景喜剧、商品动画、汽车TVC、电影短片和连续短剧。题材跨度挺大,但它都能把剧本、人物、分镜、声音和字幕串起来,最后交付完整视频,确实像一个能把活做完的全能视频 Agent。
100+ 专业 Skill 覆盖不同创作领域,选中一个,就像请来一位对应方向的 AI 导演。自己的经验也能保存成专属 Skill,这正是 LibTV 打造全球最大专业视频 Skill Hub 的价值。
成片出来后,我还能通过故事板和节点工作流继续调整镜头、时长与节奏。哪里有问题就改哪里,用起来既省事,也保留了足够的控制空间。
Source of this article: https://mp.weixin.qq.com/s/xVh0rHVh2a5onDynnqUXdA. For copyright issues, please email: 13951431913#139.com or submit a comment on the right.
Hello, I am Guo Zhen. Yesterday, I saw a few bloggers sharing about Agents making short videos, and it made me eager to try. Today, I gave it a go and generated a sitcom-style short drama like Friends, a car TVC, an e-commerce product promotion, and an AI short drama. Wow, I didn't expect the results to be this good; they have reached a level suitable for direct publishing. Excited by this, I wrote this hands-on review. If you're interested, feel free to read it. 1. Work Showcase I created these works using LibTV, a video Agent with 100+ built-in video Skills ready to use:
Each Skill can be understood as a professional AI director. As shown in the figure below, select it and enter a prompt:
It then generates the following sitcom short video. Since I usually enjoy watching sitcoms to learn spoken English, many authentic expressions are easier to remember when placed in specific scenes:
Video Details
Besides creating a two-person elevator dialogue, I remotely directed LibTV to produce a four-person living room scene similar to Friends:
Watching the finished product felt great. During the production process, I closely observed whether current advanced Agents could handle spoken dialogue, character reactions, and continuous punchlines. The results convinced me.
After satisfying my urge to make sitcom short videos, I also tried an AI short drama. Using the A24 Aesthetic Skill in LibTV, I generated the following exquisite and restrained short film:
Video Details
The video is about 1 minute long. The character performances, camera atmosphere, and emotional progression even carry a touch of mild horror. It's thrilling!
After trying short videos with real people, I also experimented with promoting bread using cartoon characters, a type of short video many are currently creating for e-commerce scenarios:
I selected the Pixar Animation Ad Skill built into LibTV, loaded a favorite cartoon character image, and as shown in the figure, entered a simple prompt. Soon, the following delightful cartoon short video was generated, showcasing the bread-making process in the store:
Next, I wanted to try car TVCs, which are more demanding in terms of camera movement, sense of speed, and product texture in commercial videos. For brand teams, this type of content is very common. I wanted to see if providing just one prompt could result in a high-end car advertisement:
After producing several clips, I feel that LibTV's final video quality is indeed quite high: select a Skill, provide a requirement, and it delivers a complete video close to a publish-ready state.
Next, I will lay out the production process behind the scenes to see exactly what the Agent did and how I got it to continue modifying when encountering video pacing issues.
2. Creating Short Videos from Scratch This section mainly introduces the core logic used by LibTV. After reading this section, you can also create the videos above. First, enter liblib.tv and select "Start Creating": After entering, select the Sitcom Skill and enter the following prompt: One of the most headache-inducing problems when making short videos is creating storyboards, but this video Agent can automatically create them: It even handles multiple core anchors in the video, such as automatically creating three-view diagrams for four character roles as shown below: Then it gradually generates the video for each storyboard: Finally, it merges them into a complete film, including sound and subtitles. See the second video at the beginning of the article for details. You can see that this Agent handles everything from task planning to text, images, sound, video, and subtitles in a series. Honestly, I've always been looking forward to such an Agent that can produce end-to-end videos! Readers familiar with me know that I work on local knowledge base Agents and need to learn from excellent products. Their interface with the intelligent agent work panel on the bottom right paired with the AI canvas on the left, especially this AI canvas, gave me inspiration: The agent works while automatically creating an AI workflow on the left. This observability is quite smooth, and each node can also be edited, providing great convenience for secondary modifications. For example, click a specific slice node in the workflow, and buttons for secondary editing such as Clip, Crop, HD, and Parse will pop up at the top:



For instance, when generating this video, I noticed the character remained still for about 1-2 seconds too long. I could then use the Clip button to directly cut it out.
Besides the workflow interface, it also supports storyboards, allowing you to preview and edit each item one by one according to the text and video order:
My feeling is that this dual mode of workflow + storyboard brings great convenience to the post-production of video creation.
The above is the main usage logic of LibTV. Oh, by the way, there is also support for customizing Skills to more conveniently create videos exclusive to you. Next, I will explain how to do it.
3. Custom Skills to Generate Continuous AI Short Dramas
Do you remember that time in high school when I watched a Hollywood blockbuster? It was quite shocking. But I didn't expect that now I could also do it through an Agent.
Below is the film-grade sci-fi continuous short drama director Skill I created myself:
You can go to LibTV, search for the skill I wrote, and directly reference it:
Then you can, just like me, embark on the journey of creating multi-episode AI short dramas.
So today, I directly set the difficulty of the written prompt to zero, as shown below. You just need to @ the skill I wrote and enter the words "generate sci-fi short drama". After a while, the first episode "Building Trust" will appear:
Video Details
A 45-second video where the core characters are aligned in this episode. After determining the core concept for the second episode, it will once again handle the storyboard script, character images, storyboard videos, and merged storyboards in a series. Here is the second episode video "Joint Battle Upgrade":
Video Details
Using the same logic, generate the third episode "Final Showdown":
I carefully reviewed the plots of the three episodes. The story narrative and the development of the protagonists are very smooth, and the transitions are silky. My biggest takeaway is that LibTV is an all-around video Agent! It can break down the script itself, define characters, create storyboards, generate videos, and string the entire process together. Finally, it delivers a complete video close to a publish-ready state for each episode, which can still be modified and continued shooting along the storyline. Finally, a summary:
This time, I used LibTV to create sitcoms, product animations, car TVCs, film shorts, and continuous short dramas. The range of themes is quite large, but it can string together the script, characters, storyboards, sound, and subtitles, finally delivering a complete video. It truly acts like an all-around video Agent that can get the job done.
100+ professional Skills cover different creative fields. Selecting one is like hiring an AI director for that corresponding direction. Your own experience can also be saved as an exclusive Skill. This is precisely the value of LibTV building the world's largest professional video Skill Hub.
After the final video is produced, I can continue to adjust shots, duration, and pacing through the storyboard and node workflow. If there is a problem somewhere, just fix it there. It is both time-saving and retains sufficient control space.

