Summary of Key Points
This year, there has been a surge in the growth of AI-generated short dramas, but the issue of facial similarity has become a common problem within the industry: characters in these dramas increasingly resemble a “average face,” with some cases even involving the unauthorized use of celebrities' or ordinary people's images. Behind this are statistical tendencies in AI models (which tend to produce the most “safe” faces), cost pressures driving mass production (to save money and computing power by using faces that are less likely to cause errors), and the transformation of new professionals, known as “card drafters,” from creative artists to mere assembly line workers. The industry is trying to find solutions through copyright licensing (buying pre-made faces) and more refined content creation, but the contradiction between scaling up production and maintaining quality still remains unresolved.
1. AI-Generated Short Dramas and Facial Similarity: Why Do So Few Faces Appear Despite a Large Population?
When you watch AI-generated short dramas, the male protagonists all have the same “cold-faced executive” appearance, the female protagonists all have the same “oval face with big eyes,” and the supporting actors are all clones of each other. This is not coincidental; it’s the result of AI models’ preferences for producing average-looking faces and cost considerations:
- Models Naturally Prefer Average Faces: When creating faces, AI models tend to choose the most likely and “safe” options (similar to selecting the most popular internet celebrities’ faces). Since East Asian facial structures are relatively flat, this tendency leads to greater similarity between generated faces.
- Cost Efficiency Over Aesthetics: Computing power is a significant expense. The more unique a face, the more wasted footage and computational resources it consumes. To reduce costs, many companies encourage card drafters to choose the faces that AI generates most easily (i.e., average faces), and they even include the amount of computational effort required in performance evaluations. After all, producing large quantities quickly is more profitable than focusing on the quality of the faces.
- Data Bias: The training data for these models often consists of videos of internet celebrities and stars, but the labels used are generic (e.g., “beautiful woman” or “handsome man”). Vague prompts lead to AI using standardized templates, resulting in faces that blend characteristics from various sources, making it easy for viewers to mistake them for real people.
2. The Role of Card Drafters: From Creative Artists to Assembly Line Workers?
Card drafters are a new profession born out of the AI short drama industry. Their job involves breaking down scripts into segments, writing prompts for AI to generate images, and then selecting usable results from the many possibilities (similar to drawing cards in a game). However, their role is rapidly changing:
- Easy Entry, Higher Barriers: In 2025, training for just a few days was enough to start working, but now the core competencies have shifted from writing effective prompts to understanding visual and auditory language (such as composition, lighting, and emotional matching). Those who don’t master these skills are limited in their abilities, similar to factory workers, and may even be replaced by more advanced AI systems.
- Salaries Not as High as Expected: While online claims suggest that experienced card drafters can earn tens of thousands of yuan per month, this is usually true for those who understand the entire production process (e.g., AI directors). Ordinary card drafters in big cities earn around 5,000–6,000 yuan per month, with additional commissions. The role is becoming increasingly automated.
- Risks of Automation: As AI models improve rapidly, fewer people are needed to produce content. In the past, it took 8–12 people to make one episode; now, a single person can handle it all. In the future, AI might even handle the generation of content without human intervention, further narrowing the role of card drafters.
3. The Impact on Real Actors: Are AI Dramas Taking Their Jobs?
The low cost of AI-generated dramas has directly affected the real-person drama market:
- Dramatic Drop in Production: In the first quarter of 2026, the number of real-person drama productions decreased by three-quarters, and the daily wages for supporting actors were cut in half. A supporting actor from Hengdian said, “Last year, anyone could get a role; this year, those without popularity or acting skills are hardly needed.”
- Unemployment Among Mid-Level Actors: The daily wage for drama actors was below 10,000 yuan in 2023, but it rose to 30,000–50,000 yuan by the end of 2024. However, with the advent of AI, many of these actors are no longer in demand. AI has turned actor faces into standard products with nearly zero marginal cost. While traditional animations cost around 15,000 yuan per minute, AI-generated dramas cost only 800–1,200 yuan, and the cost per episode for manga-style dramas can be reduced to just a few hundred yuan.
4. The Issue of Image Theft: How to Regulate Copyright?
The similarity between AI-generated faces and those of celebrities is not accidental; many cases involve intentional copyright infringement:
- Both Ordinary People and Celebrities Are Affected: Photos of Hanfu bloggers have been used in these dramas, and celebrities like Yi Yangqianxi and Zhang Jingyi have filed complaints. However, it’s difficult for ordinary people to track where their images are being used, leaving them in a similar situation to “cyber workers.”
- Industry Standardization: Large companies and platforms are starting to regulate copyright. For example, ByteDance’s “Volcanic Ark” platform has licensed templates from Stephen Chow for creative use, while the “FacesMarket” platform connects ordinary people and models with rights holders. However, the cost of purchasing such licenses is high—costing several times more than for traditional dramas.
5. Two Paths for the Industry: Scale or Quality?
The future of AI-generated short dramas lies in two possible directions:
- Mass Production for Efficiency: Continuing to rely on volume to gain success, but facing increasing risks of facial similarity and copyright issues. Platforms like Douyin are reducing their profit shares, indicating that this approach is becoming less competitive.
- Quality-Oriented Production: Creating high-quality content, as demonstrated by Liu Ziyu from Yunnan, who spent 3,000 yuan over 10 days to produce the successful drama “Zombie Sweeper.” Her focus on detail and emotional consistency in each scene made it popular overseas. However, this approach is less efficient compared to mass production.
Liu Ziyu says, “AI is just a tool; in the end, it’s still about the story.” The question remains: if the industry continues to prioritize cost-cutting, will it also eliminate the need for creative human beings? This is the core issue that AI-generated dramas must address.
(The entire analysis is written in plain language, avoiding technical jargon and covering all key points, including the reasons for facial similarity, the current state of card drafters, the impact on real actors, copyright issues, and the industry’s future. Each section is supported by specific examples from the news to ensure it is easy for non-experts to understand.)