The Li Jiaqi controversy has not yet subsided, and AI anchors working 24 hours a day have swept the live broadcast room

Welcome to the new members of the live broadcast room. If you have any questions, you can post them on the public screen…

The female anchor in front of her spoke with a smile on her face, introducing the selling points of the product as if reciting a text. Her speech speed and mood were as stable as her makeup. She didn't need to drink water or rest, and she didn't have a terrible headache like Li Jiaqi.

After watching for a few more minutes, you will find that her neck barely moves, and she only has a few sets of gestures and expressions. Occasionally, her mouth shape does not match, which completely exposes the identity of the AI.

More similar AI anchors are taking over the live broadcast room. Maybe you have already met them late at night. Can they, who work 24 hours a day, become the "back wave" to replace Li Jiaqi and Dong Yuhui?

Three to five minutes of feeding and batch copying will not make the workers tired.

"Are these shoes suitable for seniors?"

"Tell me which pair it is."

"125 (number linked)."

"Oh, it's the 98 yuan I said. It can be worn."

When the AI ​​anchor answered my questions in real time and chatted back and forth with me, I was shocked inside.

This anchor's mouth shape, voice, image and outfit, and even the rhythm of delivering goods are all very human-like. The content of his speech is matched with his gestures. He explains the details while holding the product, each mouthful of "treasures", and from time to time repeats "Click here to return the goods, all inclusive." , Mandarin is not that standard yet, and there is an arc of humanity between him and Erjia.

At the same time, her legs were firmly stuck to the ground like sea needles. When she was not talking, it was like power-saving mode. She nodded her head slightly and showed a confused smile. The uncanny valley theory poured into my mind consciously.

When the countdown to the new product was released, the anchor finally showed his AI true colors, and an obvious bug appeared. The background of the live broadcast did not change, but the person switched instantly, as if the program was restarted, stuck for a few seconds, and started talking again.

Her developer is Silicon Intelligence, a Tencent-based AI company founded in 2017.

Digital people are nothing new. Silicon Intelligence began to launch digital human products in 2020. At that time, a 30-minute training video was required to generate a human-like digital human.

In addition, there were AI anchors who broadcast news a few years ago, such as N Xiaohei N Xiaobai developed by Xiaobing; Chinese style virtual idol Ling Ling, who sells goods in Xiaohongshu; Huang Renxun’s digital double appeared at the NVIDIA press conference14 In seconds, he deceived the whole world.

In the era of large models, the digital human race has gained another spring. Start-ups are betting on the right opportunity, and major manufacturers are also rushing to enter the game. JD Yanxi, Tencent Smart Movie, Baidu Smart Cloud Xiling Platform, Huawei Cloud Pangu Digital People's Congress Model… I am afraid that I can't count them with two hands.

What makes people even more curious is, has technology improved and has it become easier for ordinary people to make AI anchors?

Ai Faner consulted a digital human company as a buyer. The customer service told me that the digital human can be customized according to the image of a real person. The basic version costs 8,000 yuan per year with a resolution of 1080p, and the advanced version costs 30,000 yuan per year with a resolution of 4K. The training materials are mainly divided into two types:

• A video of about 3 to 5 minutes spoken by a real person, with no words or missing teeth for the first 10 seconds. It is best to use green screen video to facilitate changing the background;
• A recording file of about 30 seconds, which is different from the video copy, is used to generate a preview video and test the lip-sync matching effect;

After the training materials are reviewed and approved, the training will take 2 working days. The platform will give feedback on the training effect. If the buyer is satisfied, the digital people can pack up and get ready to start work.

However, it should be noted that the clothes, movements, expressions, tones and beauty effects of the customized digital people are consistent with the video materials and cannot be replaced, so the real people try to stay in their best condition when appearing on camera. However, after the lip shape is trained by AI, it will automatically match according to different content.

In other words, this kind of digital person that clones the image of a real person one-to-one is more like a lip-syncing mouth substitute or a parrot's trumpet.

If you don’t want to be a digital person based on a real person, you can choose a public model on the platform for free or 1,999 yuan a year.

After determining the anchor image, there is another price to get the live broadcast running, 5,500 yuan for 30 days or 19,800 yuan for a quarter.

The digital human cannot just be an emotionless mouth broadcaster, it must interact with the audience and make everyone willing to click the link.

This is not difficult to implement. The live broadcast team can set up common QA in advance, automatically answer questions by capturing keyword patterns, and drive digital people in three ways when the broadcast starts.

• Text-driven, the output sound is AI sound, and the live broadcast text needs to be edited in advance;
• Audio driver, the output sound is a real person’s voice, and the sound file needs to be recorded in advance;
• Driven by real people, real people speak from behind to drive the digital population and interact in real time;

In this process, traces of AI are everywhere. The digital human is based on a large model and learns a large amount of human data; enter the product name and price, and the live broadcast script can be generated with ChatGPT; text-to-speech is also a natural language processing tool.

The AI ​​anchor who interacted with me had a more natural and flexible response, and was more likely to be driven by a real person. When she read out loud my online name, "I am Baozi," I dreamed back to the middle school class where I lowered my head and was called by the teacher. In the end, naturally Didn't buy anything.

Cheap does not mean cost-effective, it is better to be a middleman

When counting the benefits of AI anchors, the first one must be that they are cheap.

Tao Yadong, CEO of Qianyu Intelligence, once revealed that the cost of a traditional live broadcast room includes manpower, equipment, venues, etc., with a monthly expenditure of 150,000 to 250,000 yuan.

In contrast, AI digital people only spend a few thousand per month, and the live broadcast team has been reduced from a dozen people including anchors, assistants, field controllers, and videos to one or two people who condense the essence, makeup, venue, lighting You can save on layout and shooting equipment.

By the way, Qianyu Intelligence has a long history. It was jointly established by Silicon Intelligence and the MCN organization owned by Wei Ya and his wife, with each party holding 50% of the shares. Yes, that's right, Wei Ya is the Wei Ya who was once as famous as Li Jiaqi.

Not only in China, AI anchors are also popular in Southeast Asia. According to Huxiu’s report, the monthly salary of an Indonesian anchor is equivalent to a customized digital person who works 24/7. If you were the boss, who would you choose?

The image of AI anchors is also controllable, and they will not evade taxes, run away or fly solo. Being proficient in multiple languages ​​is a basic skill of AI anchors, which is exactly what cross-border e-commerce companies want. They don't have to work hard to find local live broadcast talents.

Their shortcomings are also obvious. Although they are not too painful, not too tiring, and not emotionally broken, the speeches are not interesting, the pictures are still rough, and the interaction with the audience is not natural enough.

Therefore, businesses tend to place AI anchors in marginal positions. After all, it is okay to let them memorize articles to introduce products and issue coupons, but products that value experience, such as beauty, clothing, and furniture, will probably not work.

AI anchors are currently more suitable to appear during garbage time to make up for the live broadcast time, or to live in local life live broadcast rooms that last 12 to 24 hours. They can bombard users with repetitive words and guide users to buy group buying coupons. Everyone comes and goes in a hurry and has no time to think. Distinguish between true and false.

If the pattern is opened up, AI anchors are also suitable as digital avatars of opinion bloggers. They sit behind a wooden table and talk, with profound expressions and professional gestures. They neither need to interact with the audience nor constantly adjust their movements. They have their own His unmistakable temperament tirelessly exudes the light of knowledge.

How platforms handling massive live broadcast rooms view AI anchors is also a matter of concern to businesses.

In August this year, Kuaishou launched the AIGC digital human product "Kuaishou Zhibo". In contrast, Douyin's attitude towards AI is somewhat vague. It released platform specifications related to AI generation in May, several of which affect the future of AI anchors.

• Publishers should prominently identify AI-generated content.
• When using virtual human images for live broadcast, real-time interaction must be driven by real people, and interaction entirely driven by AI is not allowed.
• AI-generated content does not have special traffic advantages. Douyin pays more attention to the quality of the content itself.

This is very interesting. Douyin does not restrict AI live broadcasts, but the content of AI anchors is generally not as good as real people. If it spreads everywhere, it will also harm the platform ecology. Therefore, in the face of algorithm rules, cheap does not mean cost-effective.

The effect is not good, but the risk is not small. Reports from Metropolis Express and Yibang Power both mentioned that AI anchors may be detected as "recorded broadcasts" or "non-live broadcasts" by video platforms, leading to violations or account suspensions.

Therefore, some businesses choose a compromise method, combining recording and live broadcasting, or dual-camera live broadcast, matching AI and real people to strive for less tiring work, one looping oral broadcast, and the other interacting and displaying products.

These merchants are also well aware that AI anchors only help live broadcasts save part of the cost of personnel and equipment, but live streaming is a complex system, and the human team behind it also has to fight for operations, product selection and supply capabilities, and the behind-the-scenes equipment The central control is optimistic about the situation, and it is impossible to be a hands-off shopkeeper.

According to digital people and companies, the transaction volume of AI anchors is naturally no worse than that of real people, or even better. Excellent cases are promoted on their official social media accounts and digital conferences around the world.

More merchants who are eager to try are hesitant about whether AI anchors can improve conversion rates and GMV, whether they will affect the tone of the brand, and whether they can build trust with the audience.

Those who use digital people are still groping, while those who sell digital people are already making money.

If you search casually on Xianyu, you can find a bunch of wild agents selling public models and operation services at low prices. The introduction says, "Generate unlimited videos", "Make live broadcasts easier", "Earn 20,000 a month without showing your face" " A must-have tool for making short videos."

It seems that everything is ready, only a few people are left to achieve financial freedom. Just like the original Yuanverse, this seems to be another story where pigs can fly in the wind.

Not surprisingly, the technology manufacturers and individual sellers interviewed by the public account "Shenxiang" all recommended agency services in addition to selling products. It seems that middlemen who make the difference have more prospects than digital live broadcasts.

The collapse of Li Jiaqi and the disenchantment of the head anchor

Whether you watch the live broadcast or not, this is a huge market, with nearly 500 million netizens buying things in the live broadcast room.

The personal IP of top anchors is much more famous than the brands they shout about.

Some anchors are like old friends who accompany them, such as Li Jiaqi, who once started with "all girls" and personally tried on lipstick colors.

Some anchors are comedians who are interesting and curious, such as Brother Yang who "reversely sells goods" and "tests violently".

Some anchors make you feel like you have learned something, such as Dong Yuhui who teaches English while delivering goods.

The farce between Li Jiaqi and Hua Xizi nakedly revealed that when consumers enjoy the "lowest price on the Internet", most of the benefits are taken by the top anchors. Ultra-high commissions and pit fees, as well as a price control system that locks in exclusive prices, force people to buy things in the live broadcast room, which in turn restricts brand development and makes the industry supply chain less healthy.

Even if nothing like this happens, live streaming is being disenchanted.

On the one hand, the bubble in the industry has dissipated and it has become more and more rational. Statistics from iiMedia Research in the first quarter of this year show that in cities such as Hangzhou, Shanghai, and Beijing, the average monthly salary of anchors has dropped by about 30% compared with the same period last year, and "supporting roles" such as operations and central control have also dropped by about 20%.

On the other hand, the live broadcast pattern is slowly de-heading, weakening the personal influence of big anchors, building a diversified anchor matrix, and making the live broadcast room itself a brand.

On the first day after Double Eleven in 2022, Li Jiaqi announced the launch of a new live broadcast room "All Girls", which will be launched by popular anchors and Li Jiaqi live broadcast room assistants Wangwang and Qingzi.

The former "live broadcast sister" Wei Ya, who has retreated behind the scenes, is also running her own brand Qianxun, reinventing the "second generation Wei Ya", building her own clothing and makeup brands, and developing digital human live broadcast business.

Brands no longer want to make wedding dresses for top anchors, but instead open more physical stores offline and explore brand self-broadcasting online to establish more direct connections with users. Douyin has also supported self-broadcasting by brand merchants through traffic tilting and annual discounts.

But we have to admit that the value of top anchors is still irreplaceable, and live streaming is still an industry endorsed by anchors. Li Jiaqi, who has lost millions of followers on Weibo, still has soaring sales. The number of viewers is still over 10 million, and the traffic is even stronger than usual.

When top anchors contribute 80% of performance, those affected by AI anchors are those massive, unknown "little transparents" with monthly salaries of several thousand and even lower return on investment.

This is already a big piece of cake. Sima Huapeng, founder of Silicon Intelligence, once pointed out that more than 95% of companies in the industry actually do flat broadcasts and store broadcasts, and many anchors can only read scripts and respond to questions casually.

AI streamers just have to be better and cheaper than them. Small and medium-sized brands that cannot afford to hire Li Jiaqi can often do live broadcasts on their own.

However, what small and medium-sized brands face is not just the lack of anchors. They also have to compete with the rules of the platform, the traffic promotion of new products, the artificial creation of popular products, and the prime time for online shopping. These are ladders that are more difficult to climb and more difficult to climb. for cumbersome battles.

horrible! There is no one in the 200-square-meter office in Hangzhou. It's all AI digital people broadcasting live 24 hours a day.

This short video was once widely circulated. Accompanied by weird background sounds, the camera panned left and right, and the erected monitors were full of digital people carrying goods live, using trained gestures and voices to recommend products to the screen. Thousands of lights at that end.

In the near future, perhaps more offices will be vacated. When a certain anchor attends an event, her digital clone may be selling goods in the live broadcast room simultaneously.

Digital companies are all talking about providing live broadcast opportunities for small and medium-sized brands to liberate humans from repetitive things. When AI replicates digital labor in batches, we increasingly need to reflect on how much creative work is left for humans.

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo