--- title: "LibTV's actual test: Human users are no longer the entirety of the product" type: "News" locale: "zh-CN" url: "https://longbridge.com/zh-CN/news/279802716.md" description: "LibTV has launched a new AI video creation platform that adopts a node-based canvas format, combining refined creation tools and automated agents, aimed at lowering the barriers to creation and improving efficiency. The platform supports both professional creators and ordinary users, offering various features such as grid card drawing and light and shadow control, helping users better achieve their creative intentions. The design of LibTV ensures that each shot is meticulously crafted, enhancing the overall quality of the video" datetime: "2026-03-19T13:50:40.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/279802716.md) - [en](https://longbridge.com/en/news/279802716.md) - [zh-HK](https://longbridge.com/zh-HK/news/279802716.md) --- > 支持的语言: [English](https://longbridge.com/en/news/279802716.md) | [繁體中文](https://longbridge.com/zh-HK/news/279802716.md) # LibTV's actual test: Human users are no longer the entirety of the product In the past two years, AI video tools have followed a very typical path. At first, it was "conversational"; you input a sentence, and it outputs a video—simple and direct, but the results are uncontrollable. Later, there was the "node-based" approach, which breaks down the creation into scripts, storyboards, visuals, and videos as individual nodes, allowing for segmented progress and frame-by-frame adjustments, but it also raised the requirements for users significantly—you not only need to understand creation but also how to arrange the tools. Each path has its own solutions and limitations. The former is too light, while the latter is too heavy. **On March 18, LiblibAI launched a new product, LibTV, an AI video creation platform in a node-based canvas format. The canvas format itself is not new, but what LibTV incorporates makes it very different.** Firstly, it has packed a large number of refined creation tools into this canvas—grid card drawing, multi-angle three-view, lighting control, image expansion, 5-second extrapolation... These functions point to the same goal: to allow creators to more accurately and conveniently control their creative intentions while utilizing model capabilities. You can feel the weight of this by looking at a user's creation case. This short film, titled "Youth Electric Fantasy Story," has a Japanese film texture, with a montage that is very fragmented but not chaotic—faces of young boys and girls, angles of light entering the carriage, the tremor of wind passing through the wheat fields, the camera movement and scenes, colors and light and shadow maintain a unified tone amidst constant switching. Opening the node diagram reveals how detailed the author has made this... It instantly clarifies why the quality of this short film is so good—every shot has been meticulously designed, not randomly selected. Secondly, LibTV can connect to Agents like OpenClaw, understanding tasks, calling models, and automatically orchestrating workflows through the platform's self-built Skills. This means that this canvas is open to two types of users—professional creators who understand workflows and ordinary users who just want to say a sentence. The refined tools address the creator's sense of control over the results, while the Agents solve the barriers and efficiency of creation. By combining extremely refined tools and automated Agents into one canvas, LibTV has taken a very different path ## The Node-based Canvas Has Been Around for Two Years, but the Gap Between Models and Products Remains To understand what LibTV is doing, we first need to look at the current state of the AI video creation industry. For those creating AI videos, there’s a term that brings a knowing smile—“drawing cards.” It means generating repeatedly, hoping for a satisfactory result. In the early days, people drew cards because the models themselves were unstable, with generation quality fluctuating wildly; now, while model capabilities have improved, the act of drawing cards has not disappeared. The reason is simple: as models improve, creators' demands also rise. Users are defining their “ideal results” with increasing precision, such as wanting a shot with accurate lighting, suitable composition, and appropriate emotion, which raises the difficulty of prompts and the requirements for the models themselves. Thus, to address this issue, node-based tools have emerged one after another. The logic of this product form actually aligns well with the structure of video creation itself—video is not a single image, but a continuous narrative with connections from one frame to the next. Breaking down the video into key nodes for separate processing, while maintaining connections between the nodes, ultimately strings together a complete narrative chain—this transforms video creation from “creating a segment and praying it works” into a process that can be advanced and verified in segments. However, in reality, over the past two years, many canvases have not truly solved the problem of precise control; they have merely reduced the granularity of the issue. This is because controlling a single node essentially fills the gap between models and products, which is not only very challenging but also changes with each model iteration. What’s more troublesome is that node-based tools place extremely high demands on users. You need to possess two abilities simultaneously: the judgment of creative intent and the understanding of tool orchestration. The former is something creators inherently have, while the latter is an entirely new technical language that has almost nothing to do with the creation itself. Both abilities are indispensable, making node-based tools remain the domain of a few professional users even long after their launch. ## LibTV's Solution: More Detailed Tools + Easier Agents, Allowing Creators to Only Focus on “Judgment Questions” The ease of use of tools and the high ceiling of creation is a paradox, yet it is simultaneously needed by users. **This time, LibTV has provided its solution through “people + refined tools” and “Agent + Skill.”** We tested both approaches. First, under the path of people + refined tools, we created image nodes, generated two ancient-style characters, and then selected the “Character Three-view” function to ensure that subsequent character movements and angles could be more stable and accurate. Next, we enter the specific scene creation, where we set three scenes: by the river, under the tree, and in the pavilion. LibTV supports multiple nodes generating simultaneously, which can reduce waiting time. **After completing these two basic steps, the refined tools of LibTV start to ramp up.** For the current image, the tools are generally divided into two categories. One category is for fine control on the original image, such as high definition, image expansion, redrawing, erasing, and cutout, among which the most noteworthy are lighting and multi-angle. In the lighting tool, you can choose the intelligent mode, input natural language for the model to understand by itself, or manually operate the three-dimensional coordinate sphere to control the angle of the light, brightness, and color, with a 0 Prompt threshold. Looking at the actual effect, after adjustments, the texture of the image has significantly improved. We continued our efforts and tried more lighting styles. Subsequently, different lighting styles can be turned into the first and last frames of the video, creating a change in atmosphere. The multi-angle tool does something similar, except that the controlled variable changes to the camera position Users can manually drag the angle ball, use prompts, or pull levers to set the angle. From the left side, top view, bottom view, back view, you select the point, and the AI generates the image directly. Waiting for a single generation is still waiting; we directly let each image generate multiple angles simultaneously, to serve as a reference for subsequent video generation, thereby more accurately controlling the camera movement and enriching the visuals of the video. In addition to fine control on the original image, there is a set of tools that can be called using slash commands. These tools extend the narrative paths that can be taken from this image, such as the multi-angle feature mentioned earlier, where you can use "multi-camera nine-grid" to let the model allocate and generate nine keyframes of different camera positions and shots in one go. There is also a plot development four-grid, generating four different plot directions at once; a 25-grid continuous storyboard, producing 25 frames at once. The generated multi-grid images can be directly sent to new nodes using the grid segmentation function in the image toolbar. We tried it out, and watching the canvas directly produce 25 grids and segment them gave an inexplicable sense of satisfaction. **It is worth mentioning that these functions essentially transform "one image" into "all possibilities that can be extended from this image." This is a very smooth flow, rather than a scattered one. LibTV encapsulates the corresponding AI capabilities according to the structure of the image, covering elements such as light and shadow, angles, shots, and plot development, demonstrating their know-how regarding models and creation.** After completing the image part, we move on to the video segment: connecting one or more image nodes to the video node, selecting the model, duration, ratio, and quality. LibTV has integrated almost all mainstream video models such as Keling and Wan here. Different models have their own applicable scenarios; for example, Keling O3 can synchronize audio and video, while Seedream generates better Chinese and ancient style scenes. Users can switch flexibly according to their needs. We selected some specific frames that match the plot changes from the previously generated 25-grid layout and attempted to connect them all to generate a video. The effect is roughly like this, quite impressive. After all, the generation of keyframes requires high-quality prompts. As an amateur user, I had to rely on the external prompts of large language models, but using the 25-grid function to first batch generate images, then select based on the results, and finally generate the video, this process requires no typing at all; you just need to wait for the results and then make judgments, significantly reducing the difficulty. After the video is generated, you can continue to edit it or use video analysis tools to create a more detailed breakdown—it can decompose the video into a storyboard, marking the shot type, camera movement, and prompts for each frame. If you have a reference video you want to imitate, you can also upload it and use this tool to analyze the shot language first. **At this point, LibTV's material generation and processing are already quite complete, but they haven't stopped.** In addition to image and video nodes, there is also a "script generation" node: you can provide it with a script outline, and it will help you generate the corresponding storyboard script. The key is that after the script is generated, clicking "batch generate storyboards" will produce all the images at once; once the images are out, clicking "batch generate video" will also complete the image-to-video step in bulk. You can make adjustments to individual storyboards or videos, or automate the entire process. Taking this user work as an example, you can intuitively see the level of automation in the entire process: There is a detail worth noting: when the system batch generates scripts, it has already created prompts for each storyboard image and video, so you don't need to write descriptions for each shot individually. You provide direction, and the AI fills in the details, which means the professional threshold for users has been lowered even further. **The batch creation of script nodes hands over the execution details to AI, with humans only responsible for judgment, so the addition of Agents has taken a big step towards lower "human content."** LibTV provides an Access Key in the personal center, and by clicking on Skills in the upper right corner, you can view it, allowing your Agents, including OpenClaw, KimiClaw, etc., to call all its capabilities by installing LibTV Skills The configuration process is very simple; just tell the Agent to install the Skills and send it the Key, and it can be used. When using it, you only need to tell the Agent what you want to generate, and the Agent will call the LibTV Skills by itself, passing your needs to the LibTV backend. The backend Agent will handle the storyboard logic, select models, adjust parameters, generate content, and finally return the results to you, while also setting up the corresponding projects on the canvas, with all nodes connected. From practical tests, we can feel that the smooth functionality orchestration and the addition of the Agent are making this product easier to use. ## Putting People and Agents on the Same Canvas Looking back at the industry, the essence of video creation tools is to bridge the gap between the real needs of creators and the capabilities of models. This gap is not due to the models being insufficiently powerful, but rather the distance between what users want and what models can understand and output. Many non-professional creators cannot accurately express their intentions using model language, leading to phenomena like card drawing and sharing prompts everywhere. What LibTV is doing this time is filling this gap with a set of very pragmatic functions. A refined toolchain breaks down the black box output of the model into adjustable actions that creators can intervene in one by one—lighting can be adjusted, camera positions can be selected, and plots can be developed. The model is responsible for generating materials, while the creator is responsible for making "multiple-choice questions." On the Agent side, there is a more long-term significance. In the future, with the accumulation of Skills and the enhancement of tool invocation and memory capabilities, the Agent can gradually evolve from "executing simple commands" to "understanding creative intentions." A workflow you set up today can be remembered by the Agent tomorrow; a lighting scheme you created today can be reused by the Agent next time; a three-view setting you established for a character today will become the default reference for the Agent to generate that character in the future. What is more meaningful at this stage is that when both are collaborating on the same canvas, you can first let the Agent generate a draft, and then adjust the unsatisfactory nodes one by one, reducing the cost of starting from scratch. Creators only need to focus their energy on the areas that truly require judgment, while handing over the rest This is a gradual evolutionary process: the model provides the underlying generative capability, the Agent is responsible for scheduling and memory, and the creator is responsible for aesthetics and judgment. The relationship among the three is no longer an "human vs tool" opposition, but a closed loop of collaborative evolution. The canvas form is indeed not new, but what LibTV has put into it makes it different ### 相关股票 - [C3.AI (AI.US)](https://longbridge.com/zh-CN/quote/AI.US.md) - [OpenAI (OpenAI.NA)](https://longbridge.com/zh-CN/quote/OpenAI.NA.md) - [Nano Labs (NA.US)](https://longbridge.com/zh-CN/quote/NA.US.md) - [YieldMax AI Option Income Strategy ETF (AIYY.US)](https://longbridge.com/zh-CN/quote/AIYY.US.md) - [iShares Expanded Tech Software Sector ETF (IGV.US)](https://longbridge.com/zh-CN/quote/IGV.US.md) - [SPDR S&P Software (XSW.US)](https://longbridge.com/zh-CN/quote/XSW.US.md) ## 相关资讯与研究 - [08:56 ETAdversa AI Wins "Most Innovative Agentic AI Security" Platform at Global InfoSec Awards During RSA Conference 2026](https://longbridge.com/zh-CN/news/281372271.md) - [08:05 ETPenguin Ai Launches Gwen, a Build-Your-Own AI Platform for Healthcare Operations](https://longbridge.com/zh-CN/news/281363782.md) - [Yupp.ai shuts down less than a year after launching with $33M from a16z crypto’s Chris Dixon](https://longbridge.com/zh-CN/news/281235238.md) - [MedPal AI Wins Strong Shareholder Backing at AGM as It Expands AI Health Platform](https://longbridge.com/zh-CN/news/281501359.md) - [08:16 ETUpstage AI Launches AI-Powered Agentic Workflow Solution for Document-Heavy Processes](https://longbridge.com/zh-CN/news/281366004.md)