Chinese short-video platform Douyin's AI New Voice Project 2.0, a talent search for AI music and virtual singers, has drawn about 28,000 participants, and related content has passed 1.56 billion views, according to a 36Kr feature credited to the music outlet Music Pioneer. The piece uses three acts from the scheme to ask what makes one AI singer stand out when almost anyone can now make a song, and its answer is that the industry has, if anything, become more dependent on human judgment.
Its starting point is that AI has let single creators do work that once needed lyricists, composers, producers, vocalists and video crews, from AI songs and music videos to virtual singers with full personas. It cites AI acts such as Xania Monet and Breaking Rust reaching Billboard-affiliated charts abroad, and Chinese virtual singers including Yuri Youli and Wu Aihua gaining attention at home. With supply so plentiful, the writers argue, what is scarce is no longer a finished track but a reason to listen. Songwriting, in their telling, is shifting from a specialist skill to something anyone can reach for, yet more supply does not by itself produce better work.
The writers interviewed the people behind three accounts chosen from the Douyin scheme, on the view that their choices matter more than the tracks themselves. The first case is Rural Low Frequency, built around a fictional old man from China's northwest, Shi Mancang, and his three granddaughters, the Flower Jacket Girls. Creator Cheng Hailin studied ethnic music and played in a folk ensemble, and says he still decides himself how brass and bamboo flute should combine, where to build tension and whether dialect, singing style and regional feel fit together.
Cheng is not from the northwest, so he researched local surnames, dialect and music to build the character; the Shi surname came from material on northern Shaanxi. Instead of a platform's stock voices, he trains a locally run model on cut-up recordings of real dry vocals, imports the melody and adjusts it over and over. The feature's point is that AI can mimic a regional style on request but cannot supply the cultural understanding that makes it cohere.
The second act, Lil WuKong, recasts the Monkey King from Journey to the West as a rapper. Its creator, known as SP, came from mixing, production and hip-hop. He says he went back to the novel rather than reach for the familiar mix of traditional style and rap, seeing the character's rebellious, unruly streak as a natural match for hip-hop, and wanting to recover traits that popular adaptations had flattened. Here AI works mainly as a tool for carrying out his ideas. He has an automated workflow generate many options, chooses the one that suits the character, then reworks it under his own rules for lyrics, rhyme and style.
SP says the music videos are still mostly handmade: images are generated and turned into video, then he filters shots, adjusts frames and fixes lip-sync himself, and he knows of no tool that makes a full video in one click, even with the many agent products now on offer. He calls the video stage the most draining part of the whole process. In his view AI has raised the average standard without lifting the ceiling. A song can be done in a week or two, but chasing a result he is happy with has no end point, and some finished tracks are remade whenever a new model appears.
The third act, Ailee, is a blue-haired foreign-girl persona made by Wang Yue, who says he started out interested in technology rather than music and grew to love music because he turned out to be good at it and his releases won praise from the industry and a response from listeners. With no training in arrangement or theory, he first had AI break down reference songs and pared back his prompts; later he generated freely, waited for a melody that moved him, and worked backwards to a theme and lyrics. The digital character began as a visual to go with the songs and took shape as fans built up. The article sees his method, in which AI widens the options and a person decides what is worth keeping, as close to how real artists find their image and persona over time through their releases.
Taken together, the feature argues, these cases undercut the idea that AI music costs almost nothing. AI has removed the priciest parts of production, but listeners still judge whether a song sounds good, whether the character is distinctive and whether the whole thing makes sense, and creators remain answerable for their work. Once the novelty fades, the writers say, audiences fall back on the oldest tests, and meeting them still takes taste, expertise and repeated polishing.
The creators say costs have moved rather than disappeared. Cheng points out that lyrics, composition, arrangement, recording and mixing each used to carry a real cost, and that AI has made basic demos very cheap in time and money, yet promotion that once needed a picture and some lyrics now calls for visuals, digital humans, music videos and constant account management. Wang puts a complete release at about 10,000 yuan, even for his well-received Listening to the Tide, and argues that spending more is itself a barrier rivals cannot easily copy when a basic AI video costs a few hundred yuan.
SP says Lil WuKong now earns from music copyright, platform revenue shares and brand deals, but credits that to his past as a producer and some experience of artist management. Beyond making music he has to handle operations, promotion and distribution. Creators without that background, the article says, still find it hard to turn songs into a business, and for most of them reach and income are now the main problems rather than production.
That is where it places Douyin's programme: the more open production becomes, the more discovery, screening, verification and access to resources matter, and the platform is trying to build a way to find and nurture acts in the flood rather than just enlarge the pool. Besides an open main track across genres for individuals, music companies and rights holders, it runs themed tracks for particular trends and persona types, and ranks entries on a chart that combines online performance, quality assessment and scores from music-industry professionals, a role the writers compare to A&R.
Unlike label A&R, the platform keeps judging after release. Rural Low Frequency's song Don't Worry had no promotion and grew on its own; its creator hurried out a promo video once it took off, and fans began using Shi Mancang as their avatar and talking to the character in the comments. The feature is broadly favourable to Douyin's scheme, and the participant and view figures are the platform's own.
Sources · 1