Cover Image | Produced by Pencil News

Two AI-generated films managed to recoup their production costs within just a week of release.

On August 22, iQIYI launched Soul Ferry: Butterfly Dream and Soul Ferry: Dream of the Celestial Maiden. Within three days, the cumulative revenue sharing surpassed 3 million yuan, with costs recouped in the first week. By September 23, cumulative revenue sharing had exceeded 8 million yuan.

A month later, Bona Film Group, in collaboration with Emei Film Group, released The Future of Sanxingdui, scheduled for October 23. This film became the first domestically produced theatrical film using AI technology to receive a public screening license from the National Film Administration.

The business of AI filmmaking appears suddenly lucrative. However, the first group to profit from this venture has already begun to experience its downsides.

Sun Shuche, founder of Yilian Technology, has long been involved in AI animated dramas, AI video production, and IP commercialization. He previously told Pencil News that during the peak of AI short dramas, one minute of content could be quoted at 1,000 to 2,000 yuan, with costs only around 500 to 600 yuan. Good orders once achieved gross margins of 60% to 70%. However, after many teams entered the market, quotes quickly dropped. According to his observations, about 90% of companies relying solely on AI short drama production orders have since exited the market.

On one hand, AI movies are achieving 'one-week cost recovery,' while on the other, the earliest profitable pure production teams are being rapidly outcompeted. After AI makes movies cheaper, where does the money in this business remain?

- 01 - Maximizing Returns on Investment

Let's consider iQIYI's Journey to the West: The Next Generation as an example.

The first batch of this AI long-form drama consists of 5 episodes, each approximately 40 minutes long, with a per-episode cost of about 900,000 yuan—roughly one-tenth of the cost of a similar live-action S-tier drama. According to chief director Li Dongshen, the project team of about 100 people completed the workload equivalent to about 2,000 people in traditional processes.

Not all of the 900,000 yuan was spent on AI models. Computing power accounted for about one-fourth to one-fifth, with most costs still going to screenwriters, art design, directors, and post-production. AI primarily eliminated many labor-intensive steps, such as set construction, soundstages, extras, location shooting, and some modeling and special effects.

A Chinese filmmaker previously calculated that a wedding scene that would cost about 60,000 yuan to shoot traditionally could be produced in a surreal version using AI for just 1,400 yuan. Cities like Beijing, Shanghai, and Shenzhen are also attracting AI video teams through computing power, venue, and industrial policies, further reducing costs borne by producers.

However, there is no uniform 'bargain formula' for AI movies.

The AIGC web film Wonder Tales: Paper Blade Survives the Wasteland, produced by Dameng Innovation and released on iQIYI this year, adopted a 'guaranteed + revenue sharing' distribution model. The team later broke down costs, finding that creative talent accounted for about half, comprehensive production about 30%, and computing power (e.g., Tokens) 20% to 30%.

Abroad, the cost structure can even reverse. AI video company Higgsfield produced the approximately 95-minute film Hell Grind with 15 people concentrating for 14 days, at a total cost of less than 500,000 USD, with about 400,000 USD spent on computing power. The film's first 25 minutes involved generating 16,181 video clips, with only 253 final shots retained—an average of about 64 attempts per shot in the final film.

Traditional films are often expensive to 'shoot,' while AI films can be expensive to 'try.' Money saved on soundstages, locations, and on-site shooting shifts to computing power, repeated generation, screening, and rework.

Another important consideration is capital turnover.

The two Soul Ferry feature films went from initiation to release in less than four months, recouping production costs by the seventh day. The same funds, which might have been tied up in a project for a year previously, now have the opportunity to be invested in the next project sooner. Low per-film costs are one factor; the ability to try multiple times a year is another.

Bona Film Group uses the same logic.

In 2024, instead of directly betting on a 100-minute theatrical film, it first produced the 13-episode AI short drama Sanxingdui: The Apocalypse, which accumulated 160 million views online. After the theme was market-tested, it proceeded with the theatrical film The Future of Sanxingdui two years later.

Previously, tens of millions might be spent before knowing if audiences would respond; now, a small-scale trial can inform the next investment decision.

- 02 - Established IPs Gain More Value

The 8 million yuan in revenue sharing for Soul Ferry is impressive, but the franchise itself is an IP with over a decade of operation.

The first season of Soul Ferry debuted in 2014, and these two AI films still involved the original core team, with original cast members licensing their likenesses and providing voiceovers. AI reduced production costs, while the decade-plus audience accumulation helped save on acquiring something difficult to buy—attention.

Thus, the AI projects most likely to yield returns at this stage are often not entirely new stories but established IPs remade more cheaply.

When Dameng Innovation produced Wonder Tales: Paper Blade Survives the Wasteland, it didn't focus solely on one-time revenue sharing. The team designed over 200 characters and visual assets, with some supporting characters receiving dedicated backstories, hoping to continue with serialized content, digital characters, or even standalone IPs.

Characters, scenes, and worlds created earlier can be reused. The third installment of Soul Ferry, The Princess's Dream, is already in progress, with digital characters, scenes, and props from the first two films directly carrying over.

Abroad, some take this approach even further, resembling mini-studios.

U.S.-based Magic Engine Studios has completed six feature films in under 15 months, with a seventh in development. Instead of betting on a single film's success, it keeps per-film costs low, continuously producing original films, sequels, and serialized content, then distributing them across FAST and streaming channels. After the success of its first film, Pumpkin Patch, the company quickly developed a sequel.

This company aims to build something simple: a film library.

If one film underperforms, it won't drag down the company; if a character succeeds, they keep filming. After AI reduces per-film costs, the 'multi-project betting' strategy once feasible mainly for mature studios becomes accessible to smaller teams.

Another U.S. startup, TrueShort, even alters film length. It raised 12 million USD this year, specializing in 20- to 30-minute vertical AI films, cut into 1- to 3-minute episodes updated weekly. Reports indicate per-episode production costs range from 1,000 to 3,000 USD, with users subscribing to the app weekly or annually.

TrueShort initially focused on true crime, driven by practical cost considerations. Recreating crime scenes traditionally required actors, locations, set design, and shooting, whereas using archival footage with AI-generated visuals can significantly reduce production costs.

With cheaper films, movies no longer need to be 90 minutes or require theatrical release. Twenty minutes can sell subscriptions, 100 minutes can generate platform revenue sharing, and successful characters can spawn sequels.

The changes Sun Shuche experienced also illustrate that simply 'making films' for others is hard to sustain with high margins. Yilian Technology gradually shifted from production orders to IP, distribution, and commercial operations, now holding over 2,000 online literature IPs. Public reports show that in the first five months of 2026, the company's revenue exceeded 10 million yuan, with net profits surpassing 2 million yuan.

Producing a film for others earns one-time production fees; owning IP allows the same story to be adapted into short dramas, animated dramas, feature films, advertising, or even international distribution. As production costs continue to decline, IP becomes easier to reprice.

- 03 - Earning Money Beyond Box Office

Some companies no longer calculate based on 'box office returns.'

Lionsgate, with its extensive library of film and TV IPs, partnered with generative video company Runway as early as 2024. This year, the two companies deepened their ties, with Lionsgate directly investing in Runway and planning to develop AI short content and new projects using existing IPs.

Previously, the residual value of an old film mainly came from rebroadcasts, copyright sales, remakes, and sequels. After AI reduces redevelopment costs, some older characters previously deemed too niche or budget-prohibitive get another chance for content creation.

TCL's approach is more circuitous.

A TV manufacturer, TCL established TCL Studios to produce AI films. The first batch of five AI short films screened offline before being released on TCL's free streaming platform, TCLtv+.

TCL doesn't prioritize box office revenue. By selling TVs and controlling streaming access, original content increases viewing time, monetized through FAST channels and advertising. Even if a film doesn't earn much individually, retaining users and selling more ads justifies the investment.

Further upstream lies another profit method.

In March, Netflix acquired InterPositive, an AI film and TV technology company founded by Ben Affleck. This company doesn't focus on 'one-click film generation' but solves specific production issues like missed shots, background replacement, and lighting restoration. Netflix later disclosed that the cash price for this transaction was approximately 587 million USD (about 3.9 billion yuan).

In the first half of this year, Douyin alone added over 220,000 AI dramas and animated dramas. While films become easier to produce, audience time doesn't increase proportionally.

Now, AI films have developed several distinct profit models: Soul Ferry relies on established IP and platform revenue sharing, Magic Engine builds a low-cost film library, TrueShort sells subscriptions, TCL uses films to drive streaming traffic, and Netflix spent 587 million USD to acquire an AI film and TV tool company.

The easiest to undercut, however, is pure production.

AI makes films easier to produce. But ultimately, what remains truly valuable are IPs, distribution, users, and tools.

This article does not constitute any investment advice.