This year, AI short dramas are very popular. From January to May this year, the AI short drama market has exceeded 22 billion yuan, and is expected to hit the 40 billion yuan mark for the whole year; domestic AI short drama users exceeded 600 million in the first half of the year; among the micro short dramas launched in the industry in the first quarter, AI micro short dramas accounted for more than 95%. Production surged and users surged. But when I opened my mobile phone and browsed several movies, many people were confused: in this movie, the heroine had almond eyes and petals and lips, but if she switched to another movie, she still had almond eyes and flowers and lips, and if she switched to another movie, her facial features are almost exactly the same. There are similar AI faces appearing over and over again. The faces in different dramas are similar, and the facial features of the characters in the same drama are highly similar. They are exquisite but lack memory points, which brings strong aesthetic fatigue.
Recently, many short drama platforms have introduced new regulations to rectify "high-frequency AI faces". Why is it that even though AI can generate an infinite number of faces, "copy-and-paste" characters often appear in short dramas? Is it a technical problem or a creative trade-off to make clear facial distinctions for AI characters? After the new rules of the platform are implemented, can they really change the current situation of "everyone looks the same" in AI short dramas?
Opening the short video platform, AI short dramas have become a new entertainment choice for many people in their spare time. However, as the content exploded, many viewers discovered that the protagonists in the AI short dramas they watched almost all had the same face. Ms. Li, an audience member, said that sometimes the thing that most dissuades her from binge watching is not the plot, but the copy-paste-like faces of the protagonists.
Ms. Li: I’ve been obsessed with watching AI short dramas recently. Nine of the ten female leads have the same pointy chin and big eyes, and even the arc of their smiles are exactly the same. I couldn’t stand it after watching half of it several times, so I gave up on the show. I think AI short dramas should not use the same template. It would be better to have some personal characteristics such as small moles or freckles. A recognizable face can also add to the drama.
There are thousands of short plays, but there are only a few faces of the protagonists that seem to turn over and over again. What exactly is the problem? Many AI short drama practitioners gave answers from the production side. Yang Ming, an AI short video production manager in Fujian, said that the face the audience sees often goes through a process from character setting to model generation to repeated adjustments.
Yang Ming: We usually determine the character settings based on the script first, including his age, occupation, personality, and the overall feeling. Then we design the face shape, facial features, hairstyle, etc., use AI to generate multiple versions, and select the ones that are more in line with the character's positioning. After finalizing it, you need to continue testing to ensure that the character is still the same person in different scenes and different angles.
However, such a process is time-consuming and expensive in terms of computing power. Li Tianlong, the head of a cultural media organization in Zhengzhou, Henan, who has been involved in short dramas for three years, said that this is exactly the step that most "factory-type" production teams who pursue fast output and high profits deliberately skip, and it has also directly spawned a large number of highly similar AI character faces in the industry.
Li Tianlong: Because the production cost is relatively low, their requirements for characters and whether they have appeared repeatedly on the platform or the market are basically no longer required. Instead, they only need to increase the quantity in order to make profits.
If we abandon the trouble-free route of "large quantity and manage enough" and instead focus on differentiation, how much difference will there be in output and cost? Li Tianlong gave an intuitive answer with a set of comparisons.
Li Tianlong: For ordinary dramas, which have the same molding ratio and basically the same faces, a factory-type team can produce at least twenty to thirty dramas a day. But for premium dramas, it is basically a team of seven people. It takes a month, or even a little more than a month, to produce a premium drama. The cost will be two to three times that of the previous drama, because the requirements for the agents are different and the cost is relatively high. If you want to be more precise, you might use one picture out of thirty, and this is where the cost is the most.
Li Tianlong calculated the cost account, but in Yang Ming's view, the reason was more than that. Behind the "thousands of people" of the AI short drama are multiple considerations such as technical stability and legal risks.
Yang Ming: Nowadays, models can generate many different faces, but it is still difficult to make a character remain stable in dozens of episodes and hundreds of shots, and at the same time be clearly distinguished from other characters. In addition, we will also consider risks such as copyright and portrait. We cannot directly imitate celebrities or existing film and television characters in pursuit of a particularly recognizable face. Therefore, sometimes producers will choose some more mature and stable solutions.
So, is it enough to make a different face? Yang Ming said that the real difficulty and investment lies in how to make this face withstand the test of continuous shots.
Yang Ming: The biggest difficulty is consistency. It is not difficult to design a new face. What is difficult is that the face cannot become another person in profile, in perspective, in different lights, or even in different emotions. We need to constantly adjust prompt words and test models. Sometimes it may take several hours or even longer to repeatedly determine a main character. The more characters there are, the higher the initial investment in time and computing power will be.
Recently, some short video platforms have introduced new regulations requiring improved character differentiation in AI short dramas. Hongguo Short Drama announced that it will launch "high-frequency AI faces", homogeneity and material violation management, requiring creators to self-examine and promptly modify and replace problematic dramas; Douyin Short Drama Creator Center has launched a pre-screening tool for AI drama copyright holders and directors to help creators screen for potential risks such as portrait rights and copyright before the finished film is released online. What impact will the new regulations have on the production side? How will the industry respond?
According to Yang Ming, the person in charge of short video production, the changes brought about by the new regulations are not only reflected in the finished film standards, but also throughout the entire production process.
Yang Ming: This impact is relatively direct. In the past, we paid more attention to whether the characters could be stable and whether the pictures could be generated quickly. Now, in the early stage, we have to separate the face shapes, facial features, and age of different characters. If the content that has already been prepared is not differentiated enough, it needs to be revised back. Therefore, the production cycle will definitely increase to a certain extent, and more time will be invested in early character design.
While responding to the new regulations, some production organizations have also begun to think further: Instead of finding a new "safe and stable" face for each drama, it is better to precipitate original characters and make them their own IP. Yang Ming said that they have an account of whether the initial investment is worth it.
Yang Ming: We are actually considering building our own character library. For example, the face shape, facial features, modeling, and personality characteristics of some original characters are fixed to form their own character assets. The initial investment will be relatively large, but if it can be used continuously in different works and the audience can remember this character, I think it will be cost-effective in the long run. It is not just an AI-generated character, but one of our own IP assets.
The production side is trying to break the situation, but the root cause of "thousands of people are the same" needs to be explored from the mechanism. Jiang Ming, deputy dean of the School of Journalism and Communication at Jilin University, believes that this is the result of the superposition of technical limitations and industry choices.
Jiang Ming: The model itself does have this tendency. This is the limitation of the current AI short drama production tools. If we give the AI relatively little information, it will naturally look for the most familiar character type from the training data as the output result. Therefore, in the early stages of production, the sophistication of character design directly determines the degree of the problem of being the same person. The second is time and cost, which is a very practical reason. If the project cycle and production funds are tight, some teams may choose to generate faster and lower-cost characters. This kind of operating idea will indeed have short-term positive feedback for the producer, but if there are too many such works, the audience will suffer from aesthetic fatigue and fall into the same predicament. Therefore, the key is to see whether the producers are serious about creating the characters.
Precipitating original characters into IP that can be used repeatedly is the direction given by some production institutions; and the audience's choice to "swipe away" again and again is also forcing changes in the industry. According to Zhang Yuze, a member of the Media and Digital Marketing Industry Working Committee of the Beijing Culture and Entertainment Law Society, "Thousands of people look the same" is just a superficial appearance. What it really touches is a more essential issue of content creation in the AI era.
Zhang Yuze: On the surface, the "everyone looks alike" in the AI short drama is a question of technology and production process, but looking at a deeper level, it reminds us: when machines can generate unlimited images, what humans really need to answer is no longer "can they create", but "what to create". The content competition in the future may not stop at who has stronger generative ability, but who can understand human nature, shape characters, and reach people's hearts. In this sense, AI brings not only a change in production methods, but also a reordering of content value: when images can be quickly copied, what can really remain are the stories that carry emotions, express experiences, and establish connections with audiences.
(Voice of China, China Central Radio and Television)
