AI Short Dramas Top 1,500 Daily Releases as Most Producers Struggle to Profit

The cost collapse has drawn a large wave of new teams. Live-action short dramas cost roughly 300,000 to 400,000 yuan at the low end and 1 million to 1.5 million yuan for premium titles, while an AI short drama can be made for about 50,000 yuan. Writers typically receive several thousand to 10,000 yuan, and production costs run 20,000 to 40,000 yuan.

Zhu Xiaoran, a short-drama screenwriter who left a large internet company at the end of August, said the pace has changed completely. In the live-action era, a writer might spend a month polishing a script. After AI tools spread, managers expected similar speed gains. At her busiest, she delivered four scripts in a month, and a producer could turn a script into a finished title within days. Quality has run into a ceiling: a fully human-crafted script can reach 90 or even 100 points, while an AI-assisted script usually tops out at 80. 'The boss knows this too, but he prefers four 80s over one 100,' she said. The industry is betting on volume, she said, with a hit rate of about 0.4 percent; the rest is effectively wasted output. She described a model in which a company could spend 6 million yuan on ten flops, and if the eleventh earns 7 million yuan, the losses are covered.

Exposure is the immediate test. Zhu said if a drama does not gain traction within three days of release, it is essentially dead, with no long tail. The first three days are when traffic spending is most concentrated. If it fails to take off, producers abandon promotion and move to a new title. Jin Nuo, a short-drama director with nine years of film experience, said the business depends heavily on paid traffic. Organic hits exist but are about as likely as winning a major lottery. Platforms have dedicated operations teams and traffic buyers, and outside traffic agencies also work on campaigns; a skilled buyer can turn an average script into a big hit. Even with paid traffic, AI short dramas hit bottlenecks. Jin said a short-drama platform's traffic pool is nicknamed the 'black box.' At 40 million in heat, titles are pushed to the next layer; at 50 million, further. But AI animated dramas get stuck at 40 million. The recommendation mechanism is unknown, and it is unclear whether it can be manipulated internally, she said.

Revenue is also shrinking. Under the dominant platform model, producers earn a minimum guarantee for the script plus a share of playback revenue. Jin said the domestic AI short-drama market could still make money in the first half of the year, but no longer. Many platforms have adjusted revenue-sharing rules, and AI animated dramas may only make a small profit in horizontal format, with higher requirements; self-produced titles generally lose money. Leiphone.com reported that Hongguo lowered its base revenue-sharing coefficients for AI animated dramas in April: AI simulated human dramas from 60 to 40, and 3D animated dramas from 50 to 40. It also canceled a universal guarantee for AI simulated human scripts from small and medium teams. In September, in 'Quality AI Drama Production Campaign 2.0,' Hongguo canceled the series scale coefficient that many teams relied on to profit from batch IP sequels. Zhu said only a tiny top tier gets meaningful revenue share; most projects cannot recover costs.

The AI wave has created a new role: the prompt or 'card-draw' operator. Writers finish scripts, but AI video tools cannot read them directly. A card-draw operator breaks each shot into prompt descriptions the tool can understand. Zhu gave the example of a script line, 'he looks at her,' which must be expanded with facial expression and demeanor, then fed repeatedly into the AI until a usable shot is generated. The role is essential and labor-intensive, but many bosses treat it as low-skill translation work, paying low wages while imposing heavy KPIs. Individual ability matters more, not less. Jin said AI acts as an accelerator that amplifies differences in talent. An experienced operator may generate a usable shot in two or three attempts, while someone with high rework rates needs seven or eight, more than doubling cost. Those who fall below basic output are reported as slowing progress and hurting quality. Jin also said professional directors are more necessary in the AI era. AI cannot lie: poor camera work and poor aesthetics are visible in the output. An unprofessional director will not last a week. Instead of relying on a full live-action crew, an AI director works like a large project manager, controlling script breakdown, lighting and camera instructions for card-draw operators and editors, editing rhythm and BGM style, and translating film terminology for generators without film backgrounds. The industry circulates a saying that writing prompts for Seedance 2.5 is like writing the Yongle Dadian, with camera movement, image and lighting specified second by second. Tools are stronger, but the barrier to using them is higher. Companies have also revised hiring. Some large firms initially recruited people without film backgrounds, then found training costs too high and began requiring film-related experience. Turnover remains high.

AIGC director Yue Qiaochu, one of the earliest entrants in AI film, left ByteDance at the end of 2024 and founded the Beijing AIGC studio Zero Point Clock Culture. Over two years, the studio developed three business lines: TVC advertising, AI content production and virtual idols. In April, his AI short film 'Ji Wang Xing He' was selected for a Cannes Film Festival screening; in September, another AI short film, 'Lifetime System,' was shortlisted for the Busan International Film Festival's ACMF. Despite that record, he described a structural problem: 'The closer you are to content, the farther you are from money.' The biggest profits flow to platforms. He noted that video generation uses Seedance, editing uses Jianying, series are distributed on Hongguo, and traffic buying relies on Juliang Qianchuan, a chain largely inside ByteDance's ecosystem. Only the platform is guaranteed to profit, he said. A production studio's annual revenue ceiling is often a few million yuan; after computing power, rent and labor, little profit remains, and payment collection is uncertain. Platforms continue raising content thresholds while small and medium teams bear advance-funding pressure. If a title's data falls short, payment cycles and advances can drag a company down. Labor also limits scale. Clients and producers often see things differently: a strong creative team may produce good work, but company heads will not pay if they see no profit. Yue said many people selling courses on entering the industry have already achieved financial freedom. His team is looking for leverage by becoming an influential KOL or personal IP in AI circles, which can raise quotes to 100,000 or 200,000 yuan, or even more than 1 million yuan, increasing the price per unit of time. Still, he said, doing delivery work is a certain business, and keeping the company alive comes first.

Overseas markets have become a collective choice as the domestic hit rate stays near 0.4 percent, but the trial-and-error cost is not lower. Zhao Lei, who works on Europe and U.S. scripts at a major short-drama company in South China, said European and American audiences have stronger payment habits than domestic viewers, partly because subscription platforms such as YouTube have operated there for years. Strong payment awareness does not mean easy money. Multiple industry reports cited by Leiphone.com say payback rates for overseas AI short-drama projects are extremely low, with traffic buying accounting for more than half of total costs, and North American customer acquisition cost rising to $5.28 per user. Content barriers are also higher. Zhao said three lines cannot be crossed in overseas AI short dramas: racial discrimination, protection of minors and religion. Scripts usually go through two reviews: a domestic chief editor familiar with overseas markets reviews the first ten episodes, and after the full draft is completed, an overseas chief editor checks for sensitive points. Audience preferences differ as well. European and American viewers prefer werewolves, vampires, dragons and wizards, but simply dressing domestic content in those genres does not work. Domestic narratives can be slow-burn, while overseas content must place its payoff earlier. The hardest part is finding emotional common ground across cultures. Zhao gave the example of a 75-year-old man released from prison and abandoned by a gang family who builds a gang from scratch; the title performed well, possibly reflecting a preference for individual heroism. Revenue structures are similar to China's, with companies rating titles by traffic spending, heat or revenue after launch. The overseas threshold is high and the profit margin for producers remains limited, with platforms taking the meat and producers drinking the soup, he said.

The market is running faster than its rules. An industry insider said a leading short-drama company believed in the first half of the year that it had mastered AI production and carried out large-scale layoffs, only to cut core execution staff and suffer a business setback. Zhu said bosses now agree that AI must be deeply integrated into the entire production process, but each company is doing it differently and few have thought it through. Another major short-drama boss insisted that his team develop an internal AI model; employees could not do it but were afraid to say so, and the matter ended in layoffs. Many cross-industry teams with technical backgrounds but little content experience entered blindly and mostly burned money. DataEye counted 1,216 active AI short-drama producers in the first quarter of 2026, falling to 698 in the same quarter, a 42 percent drop. Plagiarism is widespread. Zhao said copycats rush in after a hit. Even a one-to-one copy may not be sued if it does not become popular, because there is no revenue data and no evidence, and the original creator may see no need to sue. He also described small companies requiring writers to submit the first ten episodes as a test, then saying halfway that the project is canceled, only for the company to use AI to wash the script. Zhu said short dramas became popular because audiences always demand good stories. At a pace of 1,500 releases a day, 'good' is giving way to 'fast.' Demand for good stories will not disappear, but the industry is still waiting to see where the next 0.4 percent hit lands.