Cyber ​​beauties are born in large numbers on Xiaohongshu

Recently, users may come across some beautiful pictures on Xiaohongshu from time to time. Some of the authors of these pictures may not be real people, but AI painting software. Users can enter prompt words in the software to generate the image they want on demand. Behind this kind of image generation method, the boundary between reality and virtuality has also become blurred.

If you recently saw a group of beautiful women with sweet smiles, impeccable figures, appearance, makeup and hair on Xiaohongshu, but their palms were hidden, and clicked on them, then you may have encountered real beauties drawn by AI.

Many people were attracted by the real-life beauty of these "virtual bloggers" and liked, collected and commented on their posts, and voluntarily contributed their attention. Some virtual bloggers attracted nearly 10,000 fans in less than 20 days.

The butterfly effect continues.

Douyin and Bilibili quickly created a batch of videos explaining how to use software to draw real beauties. Some bloggers only took one morning to draw dozens of realistic pictures and posted them on Xiaohongshu, which quickly attracted a wave of traffic.

More importantly, these "virtual people" IPs, which have not yet taken shape, have already begun to be commercialized. In terms of dressing, advertising, and even bringing goods, virtual bloggers have perfectly demonstrated the charm of the attention economy. Wherever the user's eyes are, the brand will follow.

Everyone is smart. Although they know that these are virtual people drawn by AI, they are not stingy with their attention. Even though they cannot interact and connect with people like the virtual idols that appeared earlier.

The scale of AI's invasion of reality is deepening, and people's definitions and standards of truth and falsehood are also changing.

1. Mistaking the fake for the real thing

The styles can be different, such as the queen in the three-dimensional world, the campus goddess, the girl next door, the Taiwanese school girl... As long as you enter the prompt words, the AI ​​painting software - stable diffusion can generate them on demand. In addition to being like a real person, they must also be beautiful, which are the two common characteristics of all AI virtual beauty bloggers.

"I've never seen this face model before. If you hadn't told me, I would have thought it was a real person." A fan commented on an AI blogger on Xiaohongshu.

"Xiaoduo who loves traveling" is an AI virtual blogger who only works on Xiaohongshu. She is a Korean fashion hottie who looks like a girl group member. At around 2:30 pm on February 21, Xiaoduo updated a series of 13 notes, almost every one of which received hundreds of thousands of likes and collections. In less than 20 days, she has attracted 26,000 followers.

After more than 20 days on the market, Xiaoduo seems to have gone through the four seasons.

Another AI virtual blogger tends to go for the sexy route. Her homepage outfits are more straightforward, with cosplay professional suits or tights that show off her figure.

In addition to looking as beautiful as real people, these AIs also have their own personalities.

Some hope to become their fans’ ever-changing girlfriends; others give themselves the persona of a “rich and generous” academic master, posting daily videos of themselves studying hard in the library and taking various exams; others directly introduce themselves as a virtual idol who is good at music, acting, fashion, beauty and pop culture.

After seeing the photo by chance, a Xiaohongshu user with the ID of photographer couldn't believe that AI could draw such a detailed picture, with backlight, aperture and depth of field, and even the makeup on the face. He suspected that this was a photo converted into AI.

The fake beauty and appearance attracted a lot of traffic and clicks. Some users even left their personal dating information in the comment area without knowing it, asking about the relationship status of the AI ​​blogger in an attempt to attract attention. Many girls also asked for clothing links after praising the beauty of AI.

The emergence of AI virtual humans has made people in the painting industry even more frightened.

A user who majored in painting left a comment in the comment area, "Thinking about the time when I spent a lot of money to practice the human body and colors. Looking at the current AI, I fell into anxiety. But what can I do? Can the master teach me?"

Such AI real beauties have infinitely satisfied people's imagination. Beauties are no longer rare species. Just open the software and type a few commands to get the pictures you imagine. The input prompt has become the core competitiveness.

AI painting has broken through to the level of painting real people, which is really novel enough for the public. Rather than saying that the public likes AI real beauties, it is better to say that they are curious about artificial intelligence.

In the past, hyperspace, cyberpunk, metaverse, digital collections, ChatGPT, these technological elements and inventions were all human explorations of the unknown, but now it is the turn of AI painting.

Interestingly, as fans lamented, the boundary between the real and virtual worlds is becoming increasingly blurred. After seeing these AI real-life photos, one user exclaimed, "Real people are becoming more and more like AI, and AI is becoming more and more like real people."

2. 5-second generation, perfect to distorted

If there is a flaw in the AI ​​real-life blogger, it is probably that the photos are almost perfect but distorted, "even the hair has a soft filter-like silky feeling." These photos are beautiful, but apart from the beauty, there is no blank space or imagination.

They don't work hard to create the atmosphere of daily life like real-life bloggers do in real life.

AI virtual bloggers don't care too much about adjusting the light and shadow of the photo, leaving blank space in the composition, capturing and pondering the expressions and movements of the characters. For them, the effects of light and shadow and characters are just to create a sense of reality in the three-dimensional world, so more often than not, the final photos often give people a sense of disharmony that is too hard.

There is no need for any unnecessary skills, as long as the beauty is high-definition and can greatly satisfy the public's imagination of the beauty's figure and face. Such beauties can be mass-produced in less than a minute as long as the operator has a little mastery of the technology.

Many AI virtual bloggers told Xinmei Daybreak that it is very simple to make an AI real-life picture, and the time it takes to produce it is measured in seconds. Even if you are a beginner, as long as you have basic knowledge of English and software operation, and can program, you can open the open source software - stable diffusion, and get started after ten minutes of familiarity. You can make the picture you want in an hour, and become an art director in a day.

However, it is also easy to have similar faces. One user said, "This is the third identical face I have seen on Xiaohongshu from different IPs." An AI blogger explained that in this case, you only need to train the face model during operation, and then fix the face model and change the description of the background and clothes.

The AI ​​blogger said that his virtual model's face is a mixture of multiple Lora models (a category of models used to generate a specific object or style), and the weight of each mixed model is fixed at around 0.4, so in the end there is still a bit of randomness. Another AI virtual blogger said, "The face is automatically generated, and I adjusted some prompts to make the face look more unique."

From what we know so far, most of these AI virtual bloggers, which seem to be cast in the same mold, are not completely trained by operators. They are mostly a mixture of different Lora models, or on this basis, some prompt generation is adjusted, and there are also elements of randomness and uniqueness.

With low technical threshold and simple operation process, the generated photos will naturally have technical flaws. For example, the current AI can draw flying hair, but can't draw hands. A blogger said that it is necessary to draw either three hands or six fingers, and sometimes you have to draw many pictures to get a few satisfactory ones.

Some people joked that it is recommended to keep the feature that AI cannot draw hands, otherwise it will be really difficult to distinguish between virtual and real. Because when ordinary people see a group of beautiful women's photos, to distinguish whether it is AI or real people, they just need to see whether the photos show hands. If not, it is almost a "fake person".

3. Not a virtual idol yet

As long as they continue to optimize, some bloggers can still draw hands that are closer to real people. Currently, the AI ​​virtual beauties on Xiaohongshu have shown some small commercial potential.

AI blogger Xiaoru has an AI real-life blogger account on Xiaohongshu, which has gained 10,000 followers in four days. He said that he has received many advertising cooperation offers and some want to buy pictures. Unfortunately, his account has been restricted recently.

Enwen uses stable diffusion to generate images. While retaining the two-dimensional features of the human face, it also pays more attention to the texture design of the clothes. Therefore, the comment section is full of comments recommending the clothes, and fans have given enthusiastic feedback, hoping that some merchants can make these clothes.

On the social platform, a user who follows AIGC's exploration believes that if a clothing brand inputs sample pictures of clothing at this time, can it achieve the "wearing clothes" effect and realize alternative sales?

Xinmei Daybreak noticed on Xiaohongshu that some people are already exploring in this direction.

Xixi is an AIGC entrepreneur. Seeing that everyone was enthusiastic about using AI models in women's clothing advertisements, he searched on Xiaohongshu and found a spring women's clothing with slightly complex patterns and structure. He decided to do a body test on the AI ​​model, and the results were pretty good.

If you use the top pictures generated by Xixi to search for pictures on Taobao, you can also find the same style, which matches the Taobao link. However, Xixi said that the pattern formed by the algorithm is still impossible to match completely, and the details are often calculated incorrectly, but for those who do not have such high requirements, it is almost ready to try commercialization.

This wave of AI virtual bloggers that appeared on Xiaohongshu has a strong viewing value, but lacks the sense of interaction with the real world. In the future, if this problem can be solved at low cost, it may become a "nightmare" for ordinary Internet celebrities.

Xixi said that the video effects currently produced by AI real-life models are stylized and flickering, not very stable, and require a lot of manpower.

Before this, some "virtual people" that were spawned by capital had already made further progress on the road to commercialization.

Asia's first virtual person, imma, is a Japanese Harajuku girl. Since her debut in 2018, she has made friends with celebrities such as Takashi Murakami and KAWS, and has shot commercials with Dou Jingtong and Ayase Haruka. She also designs fashion, curates exhibitions, and works as a magazine reporter.

Today, imma has 33,000 followers on Xiaohongshu. In addition to staying at home to apply facial masks and visit vintage shops, she travels to Japan, New York and Paris, attends the LVMH Prize with Gu Ailing, and stands with the well-known Japanese comedian Uncle Croquette to take a photo with a funny expression.

A number of virtual digital people have also been born in China, such as AIAYI, Axi, Ling, and Reddi. Among them, AIAYI was developed by Ranmai Technology, a hyper-realistic digital human design incubator. It reached its peak as soon as it debuted. AIAYI has successively reached cooperation with Guerlain and LV, and was invited to check in at Disney and Sorayama's offline activities.

The pictures are from Xiaohongshu, AIAYI, Axi, Ling, and Reddi.

Ah Xi is a lively, smart and playful girl. She can blink and frown while chewing food. She already has 240,000 followers on Douyin. Three years ago, she posted her first video on Douyin. In April last year, Ah Xi was invited to become the first manager of the Huadianshijian Metaverse Flower Shop; in June, she filmed two advertising videos driving the OMODA 5.

Reddi and Ling share their daily lives on Xiaohongshu. Reddi is the owner of a Metaverse digital jewelry fashion brand, while Ling has an oriental face and is an ancient style virtual idol who likes Peking Opera, calligraphy and Tai Chi. In 2021, Ling not only reached a cooperation with KEEP, but also served as the virtual spokesperson for the domestic beauty brand Huaxizi. During the Dragon Boat Festival that year, Ling started her "first show with goods".

Unlike these virtual idols who were born with a silver spoon in their mouth, this wave of AI virtual beauties, although they can quickly attract a wave of traffic, have failed to create characters that leave enough room for imagination, apart from the static viewing value that is close to reality.

The first obstacle to being personified is technical iteration. At least for now, they cannot generate dynamic videos, let alone speak to interact with fans and establish connections. And they still have a long way to go to become virtual idols with vitality.

(Xiaoru, Enwen, and Xixi are pseudonyms in this article)

Author: Li Huan

Source public account: Xinmei daybreak (ID: new-daybreak), understand people and discover changes.