■ Fu Xinxin, reporter of this newspaper
Use AI to allow everyone to pursue beauty cheaply, quickly and safely, and have their own AI photographer. Recently, a mini program called "Miao Duck Camera" took this as a slogan and quickly became popular on social platforms.
In the WeChat mini program Miaoduck camera, upload 21 photos, spend 9.9 yuan discount, you can have an exclusive digital avatar, choose your favorite from more than 30 templates, and generate texture photos comparable to innocence blue and hippocampal photo studio, and there are high-definition, more like me and other refined photos available for download. From the 33 templates published by the WeChat public account, there are 27 female portrait templates, including Qinghuan, luxury photosensitive, urban formal wear, etc., and only 6 kinds of male templates, such as white clothes, formal wear, silent films, etc. On July 7, Miaoduck Camera announced on Sina Weibo: 21 new templates, including 3 for girls and 2 for boys.
9.9 yuan can't buy a loss? There is a queue of 7-8 hours during peak hours
"Come out a week earlier, I don't have to spend 299 to take image photos" "My mother said that this picture is a good looking me" "I have recommended it to all my friends around me, I can't stop at all"... Fancy praise on the Internet made "Duck Camera" quickly out of the circle.
Yesterday, after a night of waiting for the "'Digital Twin' Production Progress Bar", the reporter also completed the "first experience" of the Mini Program:
To make a photo, you first need to make a digital avatar. The operation prompt of the WeChat public account of Miaoduck Camera reads "upload photos with multiple angles, multiple expressions, and multiple backgrounds and initiate generation", but in fact it is not as simple as a sentence. The first step is to upload a frontal photo, requiring "clear facial features", "not frontal", "occluded" and "too blurry" will be rejected by the system. During the operation, the reporter tried to upload a photo of the two, but was also rejected. In the second step, add at least 20 photos with the same requirements. It's time to test the stock of photos, and it will take some time to find more than 20 "frontal" and "unobstructed" personal photos at once.
To find all the photos, you need to exit the mini program and enter the official account to pay, which is currently charged at 9.9 yuan. After paying, the real "embarrassment" comes: the system prompts that 5446 people are making in front of it, and the estimated queue time is 8 hours... When you wake up, the digital avatar appears, and three sets of photos are generated by default, in addition, there are other themes to choose from, including workplace, idol, costume, etc.
At the same time, several colleagues tried it out. Among them, a colleague did not use the account himself, but used the photos of his relatives, and also completed the production. A colleague felt that the mini program was like implementing the "face change" or "head change" technique, "some are not so similar, but the photos produced by the traditional photo studio through various PS modifications are actually not so similar to me, they are all 'fake'." Another colleague complained, "It's too difficult to turn over more than 20 front-face photos at once." ”
Xiaohongshu user @Xiaobi-chanshukubi described the details: "It doesn't feel as realistic as it is said on the Internet, but it does have its own shadow, and I feel that only a few templates are more close to themselves... It's very interesting to upload photos, uploading photos that friends say they look like me is rejected in a second, and if it is my own words, the second will pass. Technology self-media @BadCommentaryJun described the experience of small changes in colleagues' classes: "After seeing the results, I feel that 7 hours is worth it. By default, three sets of photos are generated, most of which are similar to Banban! She felt like it herself. However, it also has times when it pulls its crotch. For example, if you feel that the photo is not like yourself enough, click 'more like me' to regenerate, and the resulting face is directly broken. ”
While the praise continues, there are naturally criticisms. In the WeChat mini program, there were 641 ratings, and Miaoduck Camera scored 2.1 points. Among them, the comments of the top 3 in the number of likes are roughly as follows: "The generation process does not work, I struggled to find 20 photos, there was a bug in the generation, and the prompt had to be retransmitted, and there was not even a feedback customer service button, I suspect that it was to collect the user's mobile phone number + real photo" "There is no follow-up after passing the positive photo, and it can't be operated at all" "The contract for collecting information says that it is free, and the money is charged after the photo is transmitted." I don't know if the collected data will be used for recirculation, if I do, I will sue (anger meme)"...
AIGC into everyday life? Not by chance, but by necessity
According to Xiao Yanghua, professor of the School of Computer Science and Technology of Fudan University and director of the Shanghai Key Laboratory of Data Science, the explosion of the Miaoduck camera is an inevitable phenomenon of generative artificial intelligence (AIGC) entering people's daily lives. At present, AIGC is developing rapidly, as can be seen from the previous ChatGPT and Midjourney explosions. "Everyone has a heart for beauty, and the overnight popularity of the Duck camera is also a manifestation of everyone's continuous pursuit of a better life." Who wouldn't be happy to spend $9.9 and pass on some photos to get a personal portrait that might have cost hundreds or thousands of dollars before? ”
@差评君也表示, Miaoduck camera can be out of the circle, first, the experience threshold is low, and it can be used directly on WeChat; Second, the reputation is good, the generation effect is indeed very good, it can be said that a photo should have the look, everyone is willing to share. "I don't know how the hippocampus and Innocence Blue cope?"
"Sooner or later, the photo studio photo industry will lose 80% of its value." Xiaohongshu user @InternetDaughter commented, "Consumers don't want drills, but holes in the wall. No matter how good you make the drill, once a more efficient way to punch holes appears, it is also abandoned by consumers. ”
In view of the possible threat that Duck cameras pose to photographers, photo studios, etc., the Duck Camera team claims that they are not the photographer's enemy. "We will soon start a closed beta collaboration on invitation-based photographers. With Duck, photographers can make templates for their own style and aesthetic, so that more people can enjoy their creativity. ”
The data shows that since the official opening of the Miaoduck camera was announced on July 7, the waiting time for its generated pictures has been increasing, and it took more than ten hours to wait for the evening peak on July 17 to produce films.
In view of the problems of providing too many photos, the finished film is not realistic enough, and the production time is too long, Xiao Yanghua pointed out that the image generative artificial intelligence technology is still improving, so there will be problems such as improper detail processing and user needs far exceeding the training computing power. With more and more image samples, algorithms and models are continuously upgraded and improved, and distortion will be alleviated. "Image generation has a process of personal computer interaction, and even in AI cameras, it has a stage of photographer refinement. The more feedback more professionals give to the image generation machine, the more it can compensate for the shortcomings of the image generation technology, making the film more like a real person in more detail. ”
The reporter searched for the 4 characters of "Miaoduck Camera" on WeChat, and the page showed that there were 14 mini programs, and the developers were distributed in Shanghai, Beijing, Guangzhou, Hangzhou, Changsha and other cities. Among them, the highest-rated competitor is "Beauto beauty photos", the introduction says that you don't need 20 selfies of the Miaoduck camera, you only need to upload 1, AI beauty, one-click replacement of clothing and hairstyle.
Beauto beauty photo developer said in an interview that the team invested a lot of resources and did algorithm optimization. Therefore, users only need to upload 1 photo to get 10 ID photos and image photos. "The product is still free to use, but due to the high investment in algorithms and servers, it will soon start charging." At the same time, he believes that in the future, with the popularity of AIGC, the business of physical photo studios will definitely be affected, but it will not die. In terms of the quality of the finished film, there are more hippocampal stores, and the effect varies from person to person. "There is no problem of standardization in the application software, as long as the background technology is optimized, the film effect can be guaranteed." Of course, the ritual sense of the hippocampus in choosing clothing, making makeup, and one-on-one service is difficult for software to achieve. ”
Internet technology blogger @Godarius Ghost Guzi thinks more long-term, "9.9 yuan to unlock the photo blockbuster, which is equivalent to dividing the market share of the photography industry such as photo studios again." It is foreseeable that in the future, as the technology becomes more mature and the speed of the AI industry accelerates, all walks of life will inevitably have a trend of closure or even layoffs. How to balance the decline in jobs brought about by the development of science and technology is also a problem that needs to be considered in the future economic development. ”
User privacy leakage? How to prevent micro-development
Compared with the future, at present, people are more worried about privacy infringement issues such as their photos being leaked. On the one hand, technology developers need to collect image data of many users, and on the other hand, users use AI software with low thresholds.
Although the Miaoduck camera team promises that the photos uploaded by users will only be used for digital avatar production, will not be extracted or used for identification and other purposes, and will be automatically deleted after the avatar production is completed, some cybersecurity researchers pointed out that if the AI face changing software asks for more portrait authorization from users, this means that users will be exposed to more risks.
Xiao Yanghua said that many apps have the problem of collecting too much user privacy information. App developers are obliged to inform users of the scope of information collected and what operational options users have, and it is the responsibility of developers to protect user data from leakage.
Recently, the Cyberspace Administration of China and other seven departments announced the "Interim Measures for the Management of Generative Artificial Intelligence Services", which came into force on August 8 this year. It is specified that the provision and use of generative artificial intelligence services must not infringe on others' portrait rights, reputation rights, honor rights, privacy rights and personal information rights. Generative AI service providers shall lawfully carry out training data processing activities such as pre-training and optimized training, using data and basic models with lawful sources; Where intellectual property rights are involved, the intellectual property rights enjoyed by others in accordance with law shall not be infringed; Where personal information is involved, the individual's consent shall be obtained or other circumstances provided for by laws and administrative regulations are complied with. Providers shall perform protection obligations in accordance with law with respect to users' input information and use records, must not collect unnecessary personal information, must not illegally retain input information and use records that can identify users, and must not illegally provide others with users' input information and use records.
Shi Zhanzhong, a professor at Antai College of Economics and Management of Shanghai Jiao Tong University, believes that the idea of "inclusive prudence and classified and hierarchical supervision" proposed by China at the regulatory level is in line with the development law of technology and market, reflects China's principle of attaching equal importance to development and security, promoting innovation and governance according to law, and shows China's position of encouraging AIGC innovative development. At the same time, he suggested that the next step should be to identify the generated content such as pictures and videos in accordance with the "Provisions on the Administration of Deep Synthesis of Internet Information Services".
"From the regulator's side, there is a need to increase the supervision of AIGC." Xiao Yanghua added that of course, supervision cannot be "one tube and die", and there must be a certain degree of tolerance for new things, giving enough space for development, and guiding the healthy and orderly development of the AIGC industry under the premise of ensuring user privacy. At present, AIGC is still in the barbaric growth stage, and the low technical threshold has led to AI face change, AI voice, etc. to fake the real thing, and even online fraud cases, "but it should be affirmed that Miaoduck camera technology has a certain degree of advanced, retaining some of the user's biometrics under limited conditions, while providing a variety of templates to meet different preferences." ”
Xiao Yanghua suggested that in view of the increasing number of AIGC applications, the public should uphold a continuous learning attitude to understand the general working principle behind AIGC, including data collection, operation process, privacy risks and so on. "Before AIGC enters a perfect industrial form, users should use application software with appropriate caution, especially do not upload sensitive photos, just like ChatGPT questions, it is best not to involve trade secrets, because the platform may have collected data without the user's knowledge." You know, behind AI is still people! ”