AI-generated films are starting to become profitable: costs are recouped within a single week, bringing in 8 million yuan in revenue sharing.
Two AI films recouped their production costs just one week after release.
On August 22, iQiyi launched *Soul Ferry: Butterfly Dream* and *Soul Ferry: The Dream of Heavenly Maiden*. The accumulated revenue sharing exceeded 3 million yuan in 3 days, and the cost was recovered in the first week. As of September 23, the total revenue sharing had exceeded 8 million yuan.
A month later, *Sanxingdui: Past and Future*, produced jointly by Bona Film Group and Emei Film Group, was scheduled for release on October 23, becoming the first domestic theater film produced with AI technology and approved for public release by the China Film Administration.
The AI film production business suddenly seems far more financially viable. However, the first batch of people who made money from this industry have already experienced its downsides.
Sun Shuche, founder of Yilian Technology, has long been engaged in AI manhua dramas, AI video production and IP commercialization. He previously told Pencil News that at the peak of AI short dramas, the quotation could reach 1,000 to 2,000 yuan per minute, with a cost of only 500 to 600 yuan, and the gross profit margin of high-quality orders once reached 60% to 70%. After a large number of teams entered the market, the quotation dropped rapidly. According to his observation, about 90% of the companies that previously only took AI short drama production orders have already exited the market.
On one hand, AI films are starting to achieve "cost recovery in one week", on the other hand, the earliest pure production teams that made profits have quickly fallen into fierce competition. After AI reduces the cost of film production, where exactly does the profit of this business end up?
01
A sum of money, circulated for more rounds
Let's first take a look at the account of *Post Journey to the West* from Mango TV.
This AI long drama has 5 episodes in the first batch, each about 40 minutes long, with a cost of about 900,000 yuan per episode, which is only about one tenth of the cost of a same-level live-action S-level drama. The project team has about 100 people. Li Dongshen, the chief director, said that the team has completed the workload that would require about 2,000 people in the traditional process.
The 900,000 yuan per episode is not all spent on models. Computing power accounts for about a quarter to one fifth, and most of the cost still goes to screenwriters, art teams, directors and post-production. AI mainly eliminates many previously high-cost links, such as set construction, photography studios, extras, location shooting, as well as part of modeling and special effects work.
A Chinese film creator previously calculated that a wedding scene that would cost about 60,000 yuan to shoot in the traditional way only cost 1,400 yuan to create a surreal version with AI. Regions including Beijing, Shanghai and Shenzhen have also begun to attract AI video teams through supporting computing power, venues and industrial policies, which can further reduce the cost borne by film producers.
However, there is no universal "super low price formula" for AI films.
This year, the AIGC online film *Strange Tales: Paper Blade Crossing the Wasteland* launched on iQiyi was produced by Dameng Innovation, and was distributed under the "minimum guarantee + revenue sharing" model. The team later broke down the cost: creative talents account for about half, comprehensive production accounts for about 30%, and computing power such as tokens accounts for 20% to 30%.
Overseas, the cost structure can even be reversed. AI video company Higgsfield produced the 95-minute feature film *Hell Grind*, with 15 people working intensively for 14 days, and the total cost was less than 500,000 US dollars, of which about 400,000 US dollars was spent on computing power. For the first 25 minutes of the film, a total of 16,181 videos were generated, and only 253 official shots were finally retained. On average, each shot included in the final film required about 64 attempts behind it.
Traditional films are often expensive for "shooting", while AI films are sometimes expensive for "trial and error". The cost of photography studios, location shooting and on-site shooting is saved, and the budget is transferred to computing power, repeated generation, screening and rework.
There is also another account to calculate: capital turnover.
The two feature films of *Soul Ferry* took less than 4 months from launch to release, and recovered the production cost on the 7th day. The same sum of funds that used to be tied up in one project for a year can now be invested in the next project much faster. The low cost of a single film is one advantage, and being able to try multiple projects within a year is another.
Bona Film Group also follows this logic.
In 2024, instead of directly investing in a 100-minute theater film, it first produced the 13-episode AI short drama *Sanxingdui: Future Revelation*, which later accumulated 160 million views across all platforms. After the theme was tested by the market, it continued to push forward the theater film *Sanxingdui: Past and Future* two years later.
In the past, people could only know whether the audience recognized the work after spending tens of millions of yuan. Now, people can first spend a small sum of money to run a test, and then decide whether to invest the next sum of money.
02
Old IPs are becoming more valuable
The 8 million yuan revenue sharing of *Soul Ferry* is quite impressive, but it itself is an IP that has been operated for more than ten years.
The first season of *Soul Ferry* was launched in 2014. The two new AI films still have the participation of the original core team, with the original actors authorizing their images and providing real-person dubbing. AI reduces the production cost, and the audience accumulated over more than ten years helps it save part of the most difficult-to-obtain resource — audience attention.
Therefore, the AI projects that are most likely to generate returns at this stage are often not completely unfamiliar new stories, but IPs that have been verified by the market, which are reproduced in a lower-cost way.
When Dameng Innovation produced *Strange Tales: Paper Blade Crossing the Wasteland*, it did not only focus on one-time revenue sharing. The team designed more than 200 characters and visual assets in total, and even supplemented the backstory for some side characters, hoping to continue producing series content, digital characters and even independent IPs in the future.
The characters, scenes and worldviews that have been invested in and created earlier can be reused repeatedly. The third *Soul Ferry* film *The Princess' Dream* is also under development, and the digital characters, scenes and props accumulated in the first two films can be directly used for it.
Some overseas players have taken this approach to a level closer to small studios.
US-based Magic Engine Studios has completed 6 feature films in less than 15 months, and the 7th one is still under development. It does not bet on one film to achieve success, but reduces the cost of a single film, continuously produces original films, sequels and series content, and then distributes the works to FAST and streaming media channels. After the first film *Pumpkin Patch* gained popularity, the company quickly developed its sequel.
What this company wants to build is very simple: a film library.
Even if one film fails, it will not drag the whole company down. If a certain character becomes popular, the team can continue to make more films about it. After AI reduces the cost of a single film, the "multi-project betting" model that used to be only accessible to mature studios has now become open to small teams.
Another US startup TrueShort even changed the length of films. It raised 12 million US dollars in financing this year, focusing on 20 to 30-minute vertical-screen AI films, which are then cut into 1 to 3-minute short episodes and updated weekly. Reports show that the production cost of each episode is about 1,000 to 3,000 US dollars, and users subscribe to the App on a weekly or annual basis.
TrueShort initially focused on the true crime genre, which also has very practical cost considerations. In the past, recreating a crime scene required actors, venues, set construction and shooting, but now using archive materials combined with AI images can save a large part of the production cost.
After the cost of films is reduced, films do not necessarily have to be 90 minutes long, nor do they have to be released in cinemas. A 20-minute film can be used to sell subscriptions, a 100-minute film can be used for platform revenue sharing, and once a character becomes popular, sequels can be continuously produced.
The changes experienced by Sun Shuche also show that simply "making films for others" is very difficult to maintain high gross profit margin for a long time. Yilian Technology has gradually shifted from taking production orders to IP operation, distribution and commercial operation, and currently has more than 2,000 web novel IPs in its portfolio. Public reports show that in the first 5 months of 2026, the company's revenue exceeded 10 million yuan, and its net profit exceeded 2 million yuan.
Making a film for others only earns the production fee once. If you own the IP, the same story can be further developed into short dramas, manhua dramas, feature films, advertisements and even works for overseas markets. As production costs continue to decline, IPs are more likely to be revalued.
03
Making profits not relying on box office
There is also a group of companies that are no longer calculating the "box office account".
Lionsgate has accumulated a large number of film and TV drama IPs. As early as 2024, it cooperated with generative video company Runway. This year, the two sides further deepened their cooperation: Lionsgate directly took a stake in Runway, and plans to use its existing IPs to continue developing AI short content and new projects.
In the past, the residual value of an old film mainly relied on reruns, copyright sales, remakes and sequels. After AI reduces the cost of re-development, some old characters that previously had too small an audience and were not cost-effective to develop can now be taken out to produce new content again.
TCL's approach is more unique.
A TV selling company has established TCL Studios itself to produce AI films. The first 5 AI short films were first screened offline, and then all launched on TCL's own free streaming platform TCLtv+.
What TCL values is not the film box office. It sells TVs itself and also controls the streaming media entry. Original films can increase user viewing time, and then realize monetization through FAST channels and advertisements. Even if a single film does not make much profit, as long as it can help the platform retain users and sell more advertisements, this business model is viable.
Moving further upstream, there is another way to make profits.
In March this year, Netflix acquired InterPositive, an AI film and television technology company founded by Ben Affleck. This company does not focus on "generating the whole film with one click", but solves specific problems in the production process, such as missing shots, background replacement, and lighting restoration. Netflix later disclosed that the cash price of this transaction was about 587 million US dollars (about 3.9 billion yuan).
In the first half of this year, Douyin alone added more than 220,000 AI dramas and manhua dramas. Films are becoming easier to produce, but the audience's total available time has not increased accordingly.
At present, AI films have developed several different profit models: *Soul Ferry* relies on mature IP and platform revenue sharing, Magic Engine builds a film library at low cost, TrueShort sells subscriptions, TCL uses films to drive traffic to its streaming platform, and Netflix spent 587 million US dollars to acquire the company that sells AI film and television tools.
The pure production link is the most vulnerable to price cuts.
AI makes films easier to produce. But in the end, what is really valuable is still IP, distribution, users and tools.
This article is from the WeChat official account "Pencil News" (ID: pencilnews), written by Song Ge, edited by Huang Xiaogui, and published with authorization from 36Kr.