China's AI microdrama industry — vertical, phone-shot melodramas consumed in 60-second episodes — is on pace to top 120 billion yuan ($16.5 billion) in revenue this year, according to reporting reviewed by Rest of World. In the first quarter of 2026 alone, producers released more than 128,000 microdramas, and Rest of World found that over 95% used AI somewhere in production. A new layer of infrastructure has sprung up to feed that machine: platforms like Shenzhen-based ActID and Chengdu-based New Claw that pay ordinary people between $15 and $700 to license their faces into searchable catalogs, filterable by age, gender, and archetype ("girl-next-door," "rugged," "supermodel"), for producers to license and animate.
The Consent Problem Is Structural, Not Hypothetical
The steelman case for treating this skeptically is straightforward and shouldn't be waved away. China's Personal Information Protection Law (PIPL), in force since November 2021, classifies biometric information as sensitive personal information under Article 28 precisely because its leak or misuse can cause dignity harm or safety risk that ordinary data cannot. Article 29 requires "separate consent" — a heightened, purpose-specific opt-in — before anyone may process it, exactly because regulators anticipated that biometric data would otherwise get bundled into vague terms-of-service clickthroughs. A face licensed once, at $74 an episode on New Claw's minimum pricing, can be face-swapped into unlimited future productions, used to train downstream generative models, or resold into scenarios the original signer never previewed. Beijing lawyer Yile Deng told Rest of World the "unclear scope of authorization" is already the persistent problem in disputes reaching Chinese courts. That is not a fringe concern: the Guangzhou Internet Court alone has heard roughly 700 AI-related face-theft cases over the past three years.
The courts are not sitting this out. On March 24, 2026, the Beijing Internet Court ruled in favor of an actress whose likeness had been AI face-swapped into a 44-episode short drama without her authorization — a case the court's own English-language site confirms was decided on the theory that a recognizable AI-generated likeness violates personal rights even with slight variations, regardless of whether the underlying face-swap technique was novel. The court held the production company liable for failing to demonstrate its process was not derived from the plaintiff, and held the distributing platform liable separately for inadequate content review. That follows an earlier, first-of-its-kind Beijing Internet Court ruling in June 2024 that an AI face-swapping app's unauthorized use of models' likenesses to build face-swap templates infringed their personal information rights specifically — not merely their portrait rights — establishing that PIPL's biometric protections operate independently of China's older, narrower image-rights doctrine.
Where the Marketplace Model Actually Helps
Having stated that case fairly, the face-licensing platforms themselves are a rational, and arguably pro-consumer, market response to the underlying problem — not an evasion of it. Before ActID and New Claw existed, the alternative was not "no AI face use"; it was producers scraping faces from social media, stock photography, or unlicensed datasets with zero compensation and zero paper trail, a practice the 700 Guangzhou Internet Court cases and 85,000 unauthorized AI face/voice videos ByteDance says it has removed since early 2026 suggest was already endemic. A licensing marketplace at least creates a transaction, a price, a name, and — critically — a discoverable counterparty when something goes wrong. ActID reports roughly 800 registered users since its March 2026 launch, of whom about 300 have agreed to license images; that is a small, self-selected base opting into a market rather than being harvested from it. New Claw's operations lead, Long Lyu, put the underlying dynamic bluntly: "Whether actors choose to license or not, AI has disrupted the industry." The choice on the table is not licensed-versus-unexploited faces; it is licensed-and-compensated versus unlicensed-and-uncompensated.
The Fix Is Contractual Specificity, Not a New Statute
PIPL's separate-consent requirement already gives Chinese regulators the legal lever they need — the gap is in how narrowly platforms are required to define what "separate consent" covers, not in the statute's ambition. The 2022 CAC-MIIT-MPS Provisions on the Administration of Deep Synthesis of Internet Information Services, in force since January 2023, already require deep-synthesis providers to obtain "individual consent" before using someone's biometric data for face-generation or voice-synthesis features and to label AI-altered content. What China lacks is not authority but granularity: a requirement that face-licensing contracts specify use-cases (which genres, which platforms, whether AI-training is included), a fixed licensing term rather than an indefinite grant, and a low-friction revocation path consistent with PIPL Article 15's consent-withdrawal right. Regulators pushing platforms toward scoped, time-bound, revocable licenses — rather than either banning the marketplaces outright or leaving "separate consent" undefined — would close the real gap without killing a functioning, price-discovering alternative to face-scraping. The Beijing Internet Court's March ruling, which held platforms independently liable for inadequate review even when the production company was the direct infringer, is already nudging the market that direction by making distributors internalize the cost of thin consent. That is enforcement working as designed, not a case for rewriting the law.