Smart things (public account: zhidxcom) Author | Xu Shan Edit | Yunpeng Smart Things reported on June 23 that recently, South Korean artificial intelligence company Pulse 9 launched a Korean pop music girl group, Eternity, which is entirely made of AI. Pulse 9 created 11 AI girl group members through the “Deep Real” technology developed by the company and released the single MV “I’m Real”. The single played on YouTube reached 670,000. ▲Image source Youtube “I’m Real” MV Deep Real refers to the combination of AI-designed character images with human facial features in the database to generate brand new virtual character images. Unlike Deepfake, it does not synthesize with people’s facial images in real life. The virtual character market is on the rise. Behind these virtual character markets is the evolution of technology from generation to generation. 1. The first full-member AI girl group, AI idol is about to enter the film and television industry In December 2020, Pulse 9 launched an “exciting AI (AI heart-throb challenge)” social event. In the virtual character ideal test, people selected the 11 most exciting AI members from 101 virtual characters and formed an AI girl group. Like the domestic talent show “Creation 101”, Pulse 9 has created a “draft girl group” in the AI world. Pulse 9 was established in 2017 as an artificial intelligence technology development company. Initially, Pulse 9 focused on the business of AI painters, but the AI painter business was difficult to expand due to the epidemic. Since then, they have expanded the business of AI image generation from painting and product customization to entertainment business. In April 2021, they launched the Eternity girl group. The group name Eternity (eternal) means “to be the best virtual idol to be with the public forever”. Each AI character has a unique setting. ▲Image source Youtube “I’m Real” MV (Sample~ I don’t know you guys look the same after changing your hair style?) In addition, in December 2020, Pulse 9 also signed a “Memorandum of Understanding for AI Idol’s Entry into Thailand and TV Drama Industry” with Thailand’s Good Drama Company, hoping to open up the AI idol’s Thai market and AI idol film and television industry business. Thailand’s Good Drama company plans to invest 740 million won (approximately 4.2 million yuan) to produce a TV series called “My Dear AI”. This TV series mainly tells the love story between the male protagonist AI idol and human women, and Pulse 9 is responsible for the production of AI male idols. “In Korea, the era of virtual idols talking and interacting with humans is opening,” said Pulse 9 CEO Park Ji-eun. As a creator, she will give AI idols more fantasy and more creative characteristics. At the same time, she also said that unlike real idols, AI idols can express their opinions freely. Because people are more tolerant of AI idols, and AI idols are more tolerant of people’s malicious comments and criticisms. It is worth mentioning that all AI idols produced by Pulse 9 are virtual characters. It is difficult to find the same appearance in the real world, and to a certain extent, they avoid the dispute over portrait rights. 2. Unlike Deepfake, the “face” created by Deep Real has no such person Pulse 9 created 11 AI girl group members through the company’s self-developed Deep Real technology. Deep Real technology can create virtual people, objects, indoor spaces, creatures, and virtual natural environments through artificial intelligence. Most of the existing virtual idols are manually designed by animators for a long time, creating images frame by frame to create AI videos. However, Deep Real technology gets rid of the limitations of face design and movement in existing virtual character creation, such as long time and high operating costs, and uses AI models to generate virtual characters. Pulse 9 said that they can adjust the virtual idol’s eyes, facial expressions, mouth corners and other details through Deep Real technology to make it highly similar to the real person’s appearance and expression. Pulse 9 first created a new virtual character image through AI technology, and then they synthesized the virtual character image with the facial features of the reference character image in the database through the AI model to obtain a brand new virtual character image. The image of the virtual character designed by AI will not overlap with the person in the real world. Pulse 9 emphasizes that Deep Real is different from the existing Deepfake because it creates realistic virtual characters at a reasonable cost instead of synthesizing real faces. They believe that the operation of Deep Real technology will be more complicated. However, the company did not explain its self-developed Deep Real in more detail, and it is difficult for us to find relevant information. And she wanted to emphasize that Deepfake, which is different from it, was once called “banned” by the entire network. Deepfake refers to a type of deeply forged face-changing technology, which changes A’s face to B’s body. Through the exchange of faces, it can make A make actions that he has never done before, say something that he has never said, resulting in a fake effect. Deepfake is based on deep learning technology. By uploading the replaced “original image” and the synthesized “fake image” to the Generative Adversarial Network (GAN) model, when the fidelity of the image is high enough, the image will be output. Then, the AI model uses technologies such as video key frame extraction and face alignment to perfectly “fuse” the human face into the original video. This type of video is highly emulated and deceptive, and it is generally difficult to identify the authenticity with the naked eye. ▲Picture source Github At the end of 2017, a Reddit user named “Deepfake” successfully replaced the face of the heroine of Wonder Woman with other movies. This “realistic” video made a sensation. Subsequently, relevant people released Deepfake’s open source code on Github, and its low technical threshold allows non-professionals to quickly master Deepfake. The emergence of Deepfake, although it will be helpful in sound synthesis and repair in the field of video, audio and video, it has a greater security risk. A fictitious video will not only damage the reputation of the person concerned, but also cause market turmoil and even threaten national security. In November 2019, my country issued the “Regulations on the Management of Network Audio and Video Information Services”, which clearly stated that network audio and video information service providers should deploy relevant technical solutions for the identification of illegal audio and video and non-real audio and video. At the same time, the “battle” against Deepfake has begun all over the world. In September 2019, Facebook announced the Global Deepfake Detection Challenge, which aims to call on researchers to find effective methods to “counterfeit”, improve the technology for identifying fake videos, and maintain a harmonious network environment. Subsequently, Google AI open sourced the Deepfake video detection data set, hoping to help researchers find better ways to identify fake videos. California subsequently issued relevant regulations prohibiting the publication and dissemination of Deepfake videos; it is forbidden to use candidate images to create fake Deepfake videos during the general election. Despite the constant actions of all parties, in fact, the fake videos made by Deepfake are still repeatedly banned on various platforms, causing people to be troubled. The battle with Deepfake has continued. 3. Iteration 3.0 of virtual idols, with a total market of up to 200 billion yuan According to the “2019 Virtual Idol Observation Report” released by iQiyi, the number of users of the second element in China has reached 490 million, and 390 million people are on the road to paying attention to virtual idols. The total output of China’s virtual idol market in 2020 will reach 200 billion yuan. The earliest virtual characters formed a “singing robot” through the superposition of singing software and robot systems, which derives multiple cultures such as electronic music and ghost culture. Subsequently, the 2.0 virtual idol culture broke out. Luo Tianyi, Hatsune Miku and other virtual idols that synthesized software sound sources and anthropomorphic images formed a “phenomenon” out of the circle in the second dimension, and began to participate in various commercial activities as virtual idols. . The virtual character industry is beginning to take shape. ▲From 2020 Hatsune Miku at Station B: There will be you online AR concert in the future At present, virtual idols are ushering in the 3.0 era. Through character 3D modeling and AI synthesis technology, virtual idols are separated from the appearance of previous cartoon characters and cute cartoons, and are gradually endowed with certain learning and interaction capabilities, like a real “People”, Eternity is exactly that. In addition, virtual characters also have their own skills in other fields. During the two sessions this year, the digital virtual editor Xiao C of CCTV network acted as a reporter in the programs of the two sessions and had exclusive dialogues with NPC representatives such as Liang Qianjuan and Ma Huijuan. ▲The picture shows the virtual editor “Little C”, source Baidu AI Behind the virtual editor is Baidu Smart Cloud to provide technical support. Baidu Smart Cloud combines digital human technology with cloud computing technology to support the dual mode of AI-driven and manual supervision of “small C”, hoping to give digital humans “human temperature”. At the same time, Baidu used 4D scanning technology for high-precision facial data collection for the first time, and used AI technology to carry out repeated iterative tuning so that “Little C” could communicate with people more flexibly. Samsung also launched a virtual human project in 2020. The team created Neon through Core R3 technology. Pranav Mistry, the project leader, tweeted that Core R3 technology “can now independently create new expressions, new actions, and new dialogues, which are completely different from the original data.” On November 7, 2018, Sogou and Xinhua News Agency jointly launched an AI virtual anchor known as “the world’s first fully simulated intelligent virtual host”. The AI virtual anchor is the first to record and broadcast the news for 1 hour by the real anchor facing the camera. Through the combination of AI and face key point detection, face reconstruction, lip language recognition and other technologies, combined with voice, image and other information After modeling, the human “clone” is obtained. According to Sogou’s CEO, as long as the audience enters an existing news text on the spot, a virtual Xinhua News anchor will appear on the screen. He will not only use the same voice as a real person to broadcast, but also lip shape and facial expressions. It can also fit perfectly. Although people have always had high expectations for virtual digital human technology, the current virtual digital human still has certain shortcomings. At present, the virtual character industry has high investment costs and low income, and it is not mature enough in terms of simulation technology and content interaction. Most companies invest in virtual characters and find it difficult to make profits. The era of virtual characters has not yet fully arrived. Conclusion: The boundary between true and false is blurred, how can virtual and reality blend in “harmonious” With the continuous improvement of AI and VR/AR technology, the boundary between true and false has gradually blurred. Many software very much hope to collect all kinds of human biological data, but it is difficult to guarantee user privacy. The emergence of Deepfake is a “wake-up call” – no matter what kind of world it is, order and security are essential. If there is a deep forgery technology like Deepfake, it is difficult for people to find an effective solution. Then, the rapid integration of the virtual world and the real world will bring more difficulties, causing people to gradually lose their trust in the real world, thereby further exacerbating social loneliness. How to balance the relationship between virtual characters and real characters in the existing system, and grasp the boundary between reality and virtuality, we still need to explore
You must log in to post a comment.