Shulou(Shulou.com)11/24 Report--

In early September, many people were shocked by the news that AI's painting "Space Opera House" had won the first prize in the Human Art Competition in the United States. Many people did not expect that AI, who painted sand sculptures, had improved his skills so quickly to surpass human beings. The time of AI painting is from the beginning of the year to the second level, and the image quality is getting higher and higher, almost close to the level of human professional painters. We also see more and more people sharing their works on various AI painting platforms on social media.

AI painting is in the limelight, and the application of using AI to do video is also coming quietly. At the end of September, Meta Xiaoza unveiled his own Make-A-Video AI video production tool. This tool can generate high-quality short videos. Meta AI's video news is not hot yet, and Google is not to be outdone by launching two AI video generation tools: Imagen Video and Phenaki. The former tends to build video quality, while the latter tends to the logic and duration of video. These AI video production tools have their own characteristics.

Only a few months after the AI technology of text-to-image technology became popular, it jumped directly to text to generate dynamic video. From drawing to making video, AI is developing at an amazing speed and makes people look forward to the future of digital media. So what exactly will this leap bring to the future?

AI video is an extension of AI mapping. Before discussing what changes AI will bring to video generation in the future, let's sort out the technical principles and application scenarios of AI video generation.

Let's start with Make-A-Video of the Meta family. In the video released by Xiaoza, we can see a video show produced by AI, in which the teddy bear is painting a self-portrait. Just through the text description, Make-A-Video can generate a video. In the case of the official website, we will find that there are also some flying Superman dogs, drinking horses and so on, all of which are generated by AI.

Google's Phenaki tool is also similar to Make-A-Video, which can generate consistent videos with stories through a series of text prompts. Such as the official website shows riding astronauts, swimming bears and so on.

From AI painting to AI video production, static image creation is transformed into dynamic video interpretation of some simple plot clips, how to rely on technology to achieve?

To put it simply, the principle of AI painting is to connect the image with the text through the neural network model, and then based on the large-scale image and text training set, the text and image features are extracted to match each other, and finally a highly related image is generated.

Compared with AI painting, AI needs the cooperation of multiple AI models to complete video production. The first step of painting and making a video needs to pre-train the text-image model, and first generate a large number of images from the text. There is a big difference in the subsequent steps. AI generates video, and after completing the basic image generation, these pictures need to be connected to become a dynamic, clear and logical video. This requires an additional interpolation model to process the picture into a smooth video action frame by frame, and the super-resolution model is used to improve the pixels of the image. Through the processing of these models, the transition between the front and back frames is smoother, the pixel quality of the picture is higher, and finally the video with high resolution and frame rate is generated.

Compared with AI painting, video can be regarded as a logical and coherent composition of multiple "pictures" from a technical point of view. Video frames are images, and there are picture, logic and other aspects of correlation between each frame. Therefore, there are two levels of difficulty between Youwen Picture and YouWensheng Video. The video generated by AI is the depth extension of the image generated by AI.

Video generated by AI is relatively difficult to achieve. Why do AI researchers pursue creative work in the video field? What is the application value of AI in making video?

What is the value of AI doing video? The prosperity of the mobile Internet has spawned a variety of social and streaming platforms. The rich picture and text and video content in these platforms have become the spiritual food for contemporary people's fragmented time. With the rise of short video platform and live broadcast industry, people's demand for content is becoming more and more exuberant. This has also built a large-scale pan-content industry.

For the creation of content, the core is creativity and efficiency. However, the human-centered creation model seems to be more and more lagging behind in the high-speed iterative content industry. The AIGC model, which applies AI technology to assist content creation, begins to infiltrate into the pan-content field.

From the perspective of video creation, outside the script, finding suitable video material is the core of creation. Although there are a large number of material libraries in the industry, the process of finding material is time-consuming, and video material that matches the content of the script may not be found.

In the face of the need to improve efficiency and fit the content of the script, the AI video generation tool can solve these problems very well. Both Google and Meta's AI video tools can generate videos based on text descriptions.

At present, Make-A-Video can realize three functional scenarios: text to video, picture to video, and video to video. Google Imagen Video can not only generate high-definition videos, but also understand and generate works of different artistic styles. Google Phenaki can now convert text to video and generate longer, coherent works based on text descriptions. Phenaki is targeting the production of long videos.

Whether in the field of short video or long video, for these industries, video generated by AI will give value to the development of the video content industry.

1. Improve video production efficiency and reduce production costs at the same time. Traditional video production requires scripting, material collection, editing and other processes, each of which requires a lot of time and cost. Video generated by AI can generate video through text, or from pictures, videos and other materials, which can reduce the cost of shooting or collecting video materials. AI can correspond to the description of the script text and generate video, which greatly improves the efficiency of video production.

two。 Add a wealth of creativity. The big AI model can traverse and learn all the ideas and styles. In terms of the richness of content, it is beyond the reach of human beings. Through the feeding of different styles and creative materials, AI video generation can create works with a variety of styles to supplement the creativity of human video production.

3. Increase the value of the content industry. The innovation of AI video generation in the field of video content brings new application scenarios and new types of work to the industry. AI painting has given birth to new professional AI painters. Similar to AI drawing, AI video will also give birth to a new career, AI editor, using AI tools to create videos. In the future, video generated by AI will combine with games, film and television, media and other industries, and collide with meta-universe, AR, VR and other scenes to create more scenes and industrial value.

However, at present, the development of video generated by AI is in a very early stage, and it can not completely generate a more perfect video. There are still a lot of problems with the videos we see on Google and Meta. For example, the video action transition is not natural, the understanding angle is strange, the video resolution is not high and so on. The reason for these situations is that the ability of the AI tool model is not high, and there are certain requirements for the quality of the material data fed by the model. If these problems are not well solved, it will also limit the application of some scenes in the future, such as commercial movies and TV dramas with high pixel and logic requirements. On the other hand, short and fast small videos have different effects on quality according to different distribution channels. But in the final analysis, high-quality video content is more likely to be commercialized.

The future business model AI generates video, and the future business model depends on different application scenarios. In the face of some small B-end enterprises that mainly produce short videos, such as media, advertising, e-commerce and other industries. Google, Meta and other AI enterprises will provide AI video production application services for these small B-end enterprises. Business logic similar to AI painting may provide pay-per-view, long-term pay-per-view or production fees based on different functions and needs, helping these industries to improve the efficiency of content creation and increase traffic in online video. However, the development of this business model must be supported on a large scale in order to have the possibility of sustainable development. after all, the video tool development and operation and maintenance costs of AI manufacturers are high.

For the film and television industry, which is mainly distributed by streaming media platform and produces medium-and long-term videos, the frequency and quality requirements are high, so AI manufacturers need to provide solution-based services, or even customized services, to provide their own creative modules, such as special effects, mirror operation, transition and other modular tools. This business model is of high value, but it is a great change for the whole film and television industry as well as the upstream and downstream industry chain. The industry needs to take a long time to transition and adapt.

In addition to film and television companies, the game industry and AI video production are also likely to collide with sparks. Video content development in the game industry can use AI to generate video to improve creativity and efficiency and reduce the cost of development. The business model of the game industry will also be similar to the film and television industry to provide specialized industry solutions.

Of course, in the whole industry, there are some enterprises that do not have a high demand for video generation, but there is not no demand at all. For example, most small businesses need simple corporate promotional videos, or several annual events need video content support. The demand may be two or three times a year, and the frequency is relatively low. These companies do not have professional video production staff and may choose to apply AI video generation tools.

If the perspective shifts from the enterprise to the individual, most individual consumers can also use AI to generate video for entertainment. Just like AI mapping, videos generated by AI will become a new social media topic. Netizens can generate all kinds of videos and exchange ideas by entering text instructions. We may change from the role of being fed to the role of creators sharing ideas and ideas.

The possibility of these business models is based on the premise that the video content is good and the cost is reasonable. In the process of commercializing AI video in the future, we may still face copyright and ethical problems. Whether it is the material library or the style of video generated by AI, it is inseparable from human images, videos and other content. The AI tool requires these human-created image data training iterations. It also means that there is still a gray area of ownership dispute in terms of copyright. On the ethical side, when entering sensitive information such as violence, blood and pornography, the generated content may fall into ethical dilemmas. These problems will be accompanied by video generation for a long time, and better mechanisms and modes need to be set up to reduce the occurrence of such things.

Unlike AI making video, the final content of AI mapping can be abstracted. This kind of image content may be of higher artistic value. But for video, the content must be coherent and logical. This also requires the ability of AI to generate video. It is still unknown whether the AI birth and growth video has logic and can express the story according to the text. Especially for some deep content production, whether AI can create this kind of content needs to be a question mark. And these AI can not reach the field, is the value of human creation.

The creation of content, the creation of art ultimately leads to connection, or to connect wisdom, or to connect the soul. People express empathy through art, and these are fields that AI can't go to. The future, perhaps under the volume of AI, will be the peak of human high-quality content creation.

This article comes from the official account of Wechat: brain Polar body (ID:unity007), by Yan Liang

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.