Has *Phantom Blade* Come to Life? The Rise of AI Supporting Roles Is Truly Chilling

Having watched all the way from the era of The Legend of Sword and Fairy, I’ve actually never rejected the idea that “technology changes content.” Just like domestic wuxia games over the years have also been constantly grappling with engines, motion capture, and AI systems. The issue has never been that new technology has arrived, but whether it will first hack away the bit of human warmth that most deserves to be preserved.

On March 18, “AI actors for all roles below the second male lead” shot to the top of trending searches across major social platforms, and the discussion exploded immediately. Multiple film and TV producers and industry-verified accounts revealed that, by 2026, mainstream domestic video platforms are pushing to gradually replace second male leads, second female leads, and even lower-tier supporting roles with AI-generated characters. The core goal can be summed up in just two words: save money.

Back to this shift in direction, the pressure behind it is indeed right there in plain sight. According to anonymous producers, profits in the film and TV industry have declined over the past two years and budgets have tightened. Platforms represented by iQIYI, Youku, and Tencent Video have already begun strictly limiting new projects while also requiring producers to use AI more extensively in production and post-production. The platforms’ calculations are very clear: supporting roles don’t bring much commercial appeal, AI doesn’t need salaries, and it can also avoid risks like scandals, scheduling conflicts, and disputes.

What stings even more is that industry data shows the per-episode cost of AI production is only 1/10 that of live-action shooting. Once that number is laid out, it becomes clear to many why platforms would be tempted by the idea. The short-drama sector is moving even faster. In DataEye’s Top 100 manga-drama rankings for February, AI humanoid short dramas already accounted for nearly 60%, and viewership rose by more than 20% month-on-month. Some platforms have even introduced dedicated support for scripts of this kind.

But beyond nostalgia, I still want to say this: what feels most chilling about all this is not technological progress, but that mid- and lower-tier actors are being pushed to the edge of the cliff first. Data from the Hengdian Actors Guild shows that the frequency with which extras get work has continued to decline in recent years, with more than 60% of that linked to replacement by AI digital doubles. Quite a few actors have also said that the offers they’ve been getting lately have been cut in half compared with the past, and some have even reached the point of having no work at all.

The supporters’ perspective is not hard to understand either: cutting costs, controlling risks, and reducing acting disasters all sound pretty practical. But performance is not a collage assembled by a program. As mentioned in the article, AI actors still have a 43% distortion rate in facial micro-expressions. Experiments by the Institute of Psychology of the Chinese Academy of Sciences also show that, for subtle muscle movements tied to complex human emotions, AI’s recognition and simulation accuracy is still below 60%. Models like Seedance 2.0 have indeed given AI a basic ability to act with the eyes and have, to some extent, crossed the “uncanny valley,” but when it comes to real on-the-spot reactions, emotional flow, and the spark that comes from actors playing off one another, it is still missing that final breath of real vitality.

I rather agree with one view from within the industry: at its core, this is a tug-of-war between technological iteration and humanistic values. AI is better suited for support work, such as patching shots, fixing things in post-production, serving as a stunt double for dangerous scenes, and reducing costs in certain industrial processes, rather than clearing out human supporting actors in bulk. Because what truly makes a drama stand firm often depends on more than just the leads. Many of the characters people remember most vividly are precisely those fleshed-out supporting roles—just like the knight-errants who flash by in old wuxia stories yet still leave behind a strong sense of the jianghu.

As for where this trend is heading, I think that in the next two or three years, short dramas and low-budget projects will be the first to test AI actors on a large scale, while long-form dramas and top-tier productions are more likely to adopt a compromise model of “human-led + AI-assisted.” Legal and ethical issues are also impossible to avoid. Unauthorized “face theft” and “voice cloning” have already become increasingly common. Xiao Zhan and Luo Yunxi have both encountered AI face infringement, yet there are still no mature rules governing profit distribution and the boundaries of personality rights.

Yu Zheng also stated on the 18th that he does not deny the value of AI, but believes that live human performance and people’s emotional needs ultimately cannot be fully replaced by machines. I think that at least gets to the heart of the matter. Technology should move forward, and the industry should survive, but if in the end costs are cut while the sense of human warmth is lost, then film and television content will truly become nothing more than a polished empty shell. Like a set of moves practiced to dazzling perfection—without inner power, it is ultimately just posturing.