Amateurs are "losing ground" amid the rapid surge of AI-powered short dramas.
```
No announcements, no payment, not even a camera set up, and yet the face of the shameless, greedy villain on screen is uncannily similar to his own.
This situation recently happened to a Hanfu makeup blogger named Baicai (alias).
On March 31, the topic "AI short drama steals faces" trended online. The reason was Baicai discovered his previously shot photos had been used without authorization by the short drama "Peach Blossom Hairpin" and generated content through AI tech.
Even more unacceptable to Baicaiisthat his likeness in the show is a greedy, lustful character. He hopes the production team and platform will provide a positive response.
Baicai told Wallstreet News that currently the production companyhas not contacted him,the platform contacted him via email to report the issue, and he has begun collecting evidence and contacting lawyers to defend his rights. He later noticed the short drama has replaced some images synthesized from his photos, "Who knows who else has suffered unwarranted disaster."
Regarding the incident above, a relevant person from Hongguo responded that the platform conducted checks and actions promptly: The relevant AI short dramas were uploaded by third-party production companies. Hongguo has required the production company to modify or delete the infringing content. For production companies with repeated intentional infringement, the platform will punish them according to the circumstances.
However,Baicai's experience is not unique.
In the past year, thisshort dramahas not onlysurpassed traditional movie box office for the first time in terms of market scale,but fundamentally altered its production logic. Extreme cost reduction is infringing on ordinary people's portrait rights and voice actors' voice personality rights.
If the industry wants to survive healthily in the long run, these "makeshift crews" issues still have to be faced directly.
Out of ControlTheft of Voice and Face
To understand the intensity of this public outcry, one must look at the physical and technical settings of the infringement.
What Baicai encountered is the most typical type of copyright infringement in current AI short drama production: targeted LoRA (Low-Rank Adaptation) model training and ControlNet feature locking.
In the past, if short drama makers wanted to save actor costs, they could at most use crude face-changing software.
But now, infringers can use crawler technology to batch grab high-def photos of a blogger from various social platforms, especially those based on picture and text, with just ten to twenty photos from different angles they can locally train a custom fine-tuned model.
AI can not only reproduce the blogger’s facial topology, but also precisely remember their specific clothing textures and makeup styles.
The evolution of this "face theft" behavior isquiteironic.
At first,short dramas targetedpublic figures with their own traffic. Top celebrities like Xiao Zhan, Dilraba, Ju Jingyi were major targets, and many AI short dramas directly "put" celebrity faces onto trashy webnovel plots to attract attention.
However, after large entertainment law teams intervened and sent frequent legal letters, infringers quickly adjusted their strategy and turned their attention to ordinary individuals.
For short drama making teams hidden in the dark, it is a carefully calculated risk management strategy.
Ordinary internet users lack public voice, have no professional legal support, and are highly unlikely to even see the short drama that happens to use their face in the lower market.
Ordinary peoplebecamethe safest, cheapest, and inexhaustible "free repositories of material" in this commercial logic.
If "face theft" takes away visual identity, "voice theft" directly pierces the emotional core of a film. Just beforethe "AI face theft" incidentwith "Peach Blossom Hairpin," the voice acting industry faced unprecedented upheaval.
Recently, top domestic voice institutions like 729 Sound Studio, and leading voice actors Ji Guanlin, Shi Zekun, broke industry silence with intensive joint statements defending their rights.
Their anger targets a clear point: many short drama makers, without authorization, freely collect voice actors' voice data to generate AI short drama dubbing.
Legally, voice carries personal emotional expression and distinctiveness, strictly protected under the Civil Code's "voice personality rights".
But with rampant open-source voice cloning projects, as long as one gets a few minutes of clean audio from previous films or radio dramas, they can clone an AI voice model.
This infringement is devastating to voice actors.
On one hand, they suffer direct economic loss. Originally, professional dubbing cost hundreds per minute, now replaced by AI computing at fractions of cents per minute;
On the other hand, there is great emotional stress: many voice actors found their unique voices, crafted over years, being used in various trashy short dramas filled with borderline content, violence, melodrama, crying out for attention.
Whentheir voicesareforcibly cut and inserted ontoassembled cyber rolesby producers,voice actors can no longer tolerate it.
Hidden Lever of Extreme Cost Reduction
The root cause of rampant infringementlies deep in the upheaval of the short drama industry’s business model, andproducers’pursuit of "extreme cost reduction".
As "free short drama apps" like Hongguo rise, the free model disrupts original accounting logic:
Users don’t pay per episode, but exchange watching feed ads for content; the platform earns huge advertising fees, and then splits earnings with content producers based on duration, completion rate, etc.
In this model, the platform’s distribution algorithm becomesthe main driver of recommendations. Producers have shifted from gold miners fighting for traffic, to relying heavily on platforms as "contentsuppliers".
Early on, to compete with paid mini-programs for good shows, short drama platforms offered generous "minimum guarantee share" policies.
The platform promised, as long as the drama met quality standards and premiered on the platform, no matter the data, producers would get a guaranteed return. This was seen as a "safe haven"by many small studios.
The turning point erupted in March this year.
At that time,the industry revealed Hongguo began cancelling "minimum guarantee + share" for live-action projects, switching to a very strict pure revenue-sharing system. Later Douyin clarified it was an "adjustment" not "complete cancellation", but for many small and medium producers,the situation grew more difficult.
Losing guaranteed funds, producers now facegreaterbusiness uncertainty. Carefully shot dramas, if unable to trigger platform’s algorithmic recommendation thresholds at launch, will be drowned in hundreds or thousands of new dramas daily, ultimatelyearning nothing.
Huge uncertainty swung industry trends again.
Since earningscannot be determined, the only survival rule left for producers is: on the front end,some participants beginextremecost cutting.
Since March,manyproducers have stopped live-action shoots,and begunmoving to cheaper AI short dramas.
For all-AI short dramas, even extras, cameras, and venues are entirely omitted. Producers only need AI painting tools to generate scenes, and feature locking plugins to create comic strips. Production costcanbe cut to a tenth, and timeevenshortened to 48 hours.
A short drama worker in Xi'an told Wallstreet News the trend of AI replacing live actors is obvious, after platform revenue share adjustments, "Everyone feels live-action drama is too expensive, so many bosses don’t want to invest."
He believes, just as short dramas impacted long video market with "extreme cost and fast turnaround", AI will also reshape the live-action short drama market in this way.
Understanding the evolution of industry extreme cost reduction explains why "face theft" and "voice theft" have exploded recently.
In an extreme ecosystem where you might get nothing, it’s obviously uneconomical for grassroots teams to buy costly stock images, hire original artists, or pay voice actors by union rules..
In the industrial gears of "bad money drives out good", infringement is no longer accidental in legal process, but becomescarefully calculated, evenan industry-sanctioned hidden business lever.
Producers’ survival philosophy: grab first, generate fast, launch quickly—if it goes viral then it’s a big win; even if sued by the original author, the drama’s money-making phase is over, just take it offline or shut down the shell company.
Low cost of breaking the law, high cost of defending rights is also why infringement happens so often.If the penalty is just removal, but legal compliance costs are so high they could bankrupt a company, infringement becomes a risk-free proposition.
Ordinary people’s digital rights and entertainers’ professional dignity have inevitably become expendables under this giant short drama money-printing machine.
"Safe Haven" Platform Collapse
If the short drama industry wants to truly cross from a "rough" phase to "mainstream film,"clearly a new set of rules must be established within legal boundaries.
When facing systemic infringement, simple scattered individual defenses are a drop in the bucket, real regulation must come from top-level design and platform governance.
Under China’s Civil Code, both unauthorized use of others' likenesses and unauthorized cloning of voices have been clearly defined as infringements.
In 2024, Beijing Internet Court's first nationwide verdict on "AI-generated voice infringement" established at the judicial level that "unauthorized use of AI-cloned voice constitutes infringement."
However, theory is rich but practice is thin.
Ordinary people face a "three-high-one-low" dilemma with massive AI-generated content: high discovery cost, high technical threshold for evidence, high litigation time cost, and extremely low compensation.
Spending tens of thousands on legal fees and half a year’s time may only yield a few thousand yuan in compensation and a sheet of apology. This unequal game indirectly encourages rampant infringement.
The judiciary urgently needs to introduce higher "punitive damage" mechanisms in future cases to truly hurt those who profit from large-scale infringement.
In this storm of infringement,short drama platformsas hubs for traffic distribution and final profit sharing,also have gatekeeper responsibilities.
For platform-pushed, heavily distributed, or revenue-sharing short drama content, platforms should set up or connect to a "digital copyright fingerprint bank" with an all-web vision. During upload review, AI anti-counterfeiting models based on GANs should be mandatory, and once content suspected of infringement is detected, publishing should be blocked immediately.
Hongguo reps say AI short dramas as a new business are hard to identify for infringement, but platform will work to improve review capabilities. For some image rights infringement situations, defining whether authorization exists, whether personal appearance wishes are respected is harder; plus, in dynamically generated dramas, characters’ faces change with expressions and scenes, so platforms have difficulty accurately identifying and intercepting during review. They have set up reporting emails for rights holders and users to report and supervise, once confirmed, platform will handle promptly.
Technological Countermeasuresthe Path Forward
When Pandora’s box is fully opened, mere moral appeals and delayed post-hoc accountability are clearly insufficient to stop large models’ greedy consumption of high-quality human data.
Yet history proves, technological evolution is never a one-way slaughter, but an endless contest of "spear and shield".
How to prevent personal features from being stolen, so unique faces and voices become legally protected and licensable digital assets, concerns individual dignity and fosters a new, imaginative business track.
Facing all-pervasive AI, defense strategies are undergoing a major shift.
Turning back the timeline two years, when generative AI first exploded, the illustration community was the first to resist indiscriminate scraping, widely using tools like Nightshade or Glaze.
This early tech injected noise into image pixels; if AI trained on them, it would cause the generation logic to collapse.
Destructive self-rescue methods like this, meant to "burn together", showed creator anger at a time, but were controversial for possibly harming open-source ecology.
Now, as AI compliance advances, data security tech has moved from destructive pushback to constructive hidden encryption and dynamic attribution.
Currently, content safety’s new trend combines "active defense and content provenance". The industry is rapidly implementing higher level dynamic reverse engineering shields and provenance alliance standards.
Among them,the mechanismsare becomingextremelysophisticated.
For example,when a Hanfu stylist uploads a new set of photos or a dubbing agency publishes audio, the system automatically injects cryptographic or blockchain-based hidden watermarks or credentials.
These shields are not meant to "destroy AI", but create an accurate "insulation layer". Human browsing and liking is normal, but if unauthorized drama makers try to scrape facial topology or audio features to train LoRA models, reverse-engineering algorithms activate—collectors get scrambled gibberish or invalid features.
This shift from passive to proactive encryption truly equips ordinary people’s digital assets with bulletproof protection, spawning Protect AI, Reality Defender, RealAI and other nearly unicorn AI safety companies.
The maturity of safety techaims toensure legal data circulation.
Blocking is less effective than guiding—massive short drama output daily is a fact, rather than letting the gray chain thrive on the dark web, it’s better to leverage underlying encryption tech to build a legal, transparent "digital asset trading market" and open a new era for the legal monetization of "micro-IP".
What’s "micro-IP"? Traditionally, only a complete play or famous character counts as IP. But with AI-rebuilt pipelines, granularity becomes infinitely smaller.
For example,Baicai's viral Hanfu makeup set, an ordinary person's strong internet-worthy face profile, voice actor Ji Guanlin’s unique emotional crying sound—all certified by attribution tech can become independent, protected "micro-IP".
If the industry builds a compliant data licensing hub, ordinary people may package their facial features or voiceprints for official sale as "AI generative material packs".
Short drama producers needn’t risk infringement or lawsuits—just pay a small, frequent authorization fee (eg. each one-minute generated by a face model, a few yuan deducted automatically) for a legal token.
For voice actors suffering infringement, this is a chance to go from passive to proactive.
Top voice actors could partner with large, compliant AI companies, use secure data cleaning to train and publish their own "official AI voice package". With intelligent contracts, they can set strict usage boundaries for their voices.
AI reshaping content pipelines is an unstoppable trend. But an entertainment industry seeking mainstream recognition and hundreds of billions in scale cannot be built long-term by plundering ordinary people’s digital rights.
Between efficiency and dignity, protecting safety, compliance, and distributive justice is the right way for AI's commercial integration.
Risk Warning and DisclaimerThe market has risks, investment needs caution. This article does not constitute personal investment advice, nor does it consider individual users’ specific investment goals, financial situation or needs. Users should determine whether opinions, views, or conclusions herein suit their specific situation. Investing based on this is at your own risk. ```