640 Viewers, No Blockbusters: The First Box Office Report Card for AI Movies
In 2026, the first feature films made almost entirely with AI hit theaters in China and South Korea. Every one of them flopped at the box office, some with only a few hundred viewers. We lined up the numbers for all of them and compared them with the hits AI micro-series are scoring on phones.

What's going on
2026 was the first year that feature films made almost entirely with generative AI reached theaters. These are programs that generate images, motion and voices from text descriptions. In South Korea, "Man in Hanbok" and "I'm PoPo" opened in May, followed by "Steelay" in October. In China, Bona Film's "Sanxingdui: Future Past" opened on October 23, the first AI film to get an official theatrical release license. On November 25, MBC C&I's "Raphael" opens in South Korea.
The question this article asks: what do the box office numbers show? In short, none of these films came anywhere close to earning back its money. At the same time, AI micro-series on phones are reaching millions of people. We test the theory that the technology didn't fail, the theater did as a venue. Our answer: that's only partly true.
The numbers
For this table, we compiled figures from several sources: Maoyan (China's ticketing platform), KOBIS (Korea's box office reporting system, as cited by Tech42, JoongAng Daily, Edaily and Cine21), and figures from the producers. The break-even point is the number of viewers at which a film recoups its costs.
| Film | Release | Budget | Screenings | Admissions | Gross | Break-even |
|---|---|---|---|---|---|---|
| Man in Hanbok (KR) | May 21, 2026 | not disclosed | not disclosed | 640 | approx. 5.63M won | not disclosed |
| I'm PoPo (KR) | May 21, 2026 | not disclosed, approx. 2 months of work | not disclosed | 781 | approx. 8.13M won | not disclosed |
| Steelay (KR) | Oct 21, 2026 | under 1B won (approx. $743,000) | not disclosed | approx. 5,200–5,300 | 43.7M won (as of Oct 25) | approx. 300,000 admissions |
| Sanxingdui: Future Past (CN) | Oct 23, 2026 | over 10M yuan (approx. $1.5M) | 8,469 on opening day | 57,000 (as of Nov 2) | 1.80M yuan (as of Nov 12) | not disclosed |
| 중간계 / Run to the West (KR, hybrid) | Oct 15, 2025 | not disclosed | not disclosed | approx. 28,000 | not disclosed | approx. 200,000 admissions |
| Raphael (KR) | Nov 25, 2026 | not disclosed | – | – | – | – |
We couldn't find screening counts for "Man in Hanbok," "I'm PoPo" or "Raphael." For "Sanxingdui," Sina reports 8,469 screenings on opening day, a 2.8 percent share of all screenings. Other reports, citing Maoyan, mention more than 10,000 scheduled screenings. According to Maoyan, an average of about 1.1 people sat in each screening, dropping to just 0.3 on October 26 and 27. On November 11, the film took in only 3,900 yuan, and on day 21 it was playing 29 screenings in 17 theaters in Beijing.
Running the numbers: so far, "Sanxingdui" has grossed less than a fifth of its costs. In China, taxes and fees and the theaters' share still come out of box office revenue. Under the usual splits, we estimate the producers are left with a few hundred thousand yuan at most. For "Steelay," viewers paid an average of around 10,000 won per ticket. The film is more than 98 percent short of its break-even point of 300,000 viewers. Tickets for "중간계" cost 8,000 won, and the film reached about 14 percent of its break-even point.
How the films were made
The production details show that "cheap" is relative. According to Bona, almost 100 people worked on "Sanxingdui" for more than two years. They generated more than 1.2 million source images and revised the entire film more than a hundred times. Most of the technology came from Vidu, a video model from Beijing-based Shengshu. Bona CEO Jiang Defu puts the cost in the tens of millions of yuan. According to Sina, the project would have cost 800 to 900 million yuan as a live-action film.
According to Edaily, "Steelay" was made in six months by a team of just over ten people. They generated about 13,000 clips of roughly 15 seconds each, using tools including Seedance, Kling, Higgsfield, Veo, and Suno for the music. According to JoongAng Daily, 500 to 600 million won, more than half the budget, went on AI tools alone. "Raphael" runs 138 minutes, was made over 14 months without a single real shooting day with actors, and was backed by the culture ministry's state AI funding program. No budget has been released. In its review, Korea Times praises the clean visuals but finds the story stale.
Critics in China and Korea largely agree on why the films failed. With "Sanxingdui," they say, the images came first and the story was built around them afterward. The result feels like "several group projects glued together." Zhang Peng of Nanjing Normal University told the Global Times:
"Technology can't make up for a weak story."
In Korea, viewers told Edaily it felt like watching "one long tech demo" and criticized the wooden emotions. Writing about "Steelay," the Daily Ann said you need a reason to spend your money on this film rather than another one.
The flip side: success on phones and Netflix
The same tools really are producing hits on small screens. According to FTNN, Bai Feikong, a 40-year-old former 3D designer from Sichuan, single-handedly created a 96-episode series about a reincarnated praying mantis using ByteDance's Jimeng tool. It racked up more than 560 million views. It cost him 4,000 yuan and earned him about 60,000 yuan. According to Sohu, a Harry Potter fan film by the creator Foyege gets more than a million views per episode on YouTube and TikTok, at a cost of about 1,000 yuan per episode. China's online media industry association says that of the roughly 128,000 micro-dramas released in China in the first quarter of 2026, more than 95 percent were AI-generated. Micro-dramas are phone series with episodes of one to two minutes.
The clearest case is Kang Yoon-sung's "중간계" (Run to the West). In October 2025, it drew only about 28,000 viewers in theaters. It landed on Netflix on June 5, 2026, and according to Wikitree it hit No. 1 on Korea's movie chart the next day and stayed there for at least five days. But it isn't a pure AI film. It stars well-known actors such as Byun Yo-han and Kim Kang-woo, and only 16 to 18 creatures and some effects were generated with AI. That makes it only partly useful as evidence for whether the venue is what matters.
Short-form isn't a gold mine either. According to DataEye, more than 221,900 AI series launched on Douyin in the first half of 2026. Only 1,055 of them, or 0.47 percent, topped 100 million views. Revenue per 10,000 views has reportedly dropped from 30 to 100 yuan to 5 to 10 yuan. And during the Spring Festival, according to the industry association, only about one in fifty short dramas was shot with real actors, yet those got 25 times as many views as the AI series.
Build on a model and you depend on its provider
There's a second problem that tends to get overlooked. "Critterz," an animated film with a budget of under $30 million, was made with OpenAI tools. In March 2026, OpenAI discontinued Sora, and the app went offline on April 26. As a result, the Cannes premiere didn't happen. According to The Next Web, sales agent AGC showed buyers only early footage at the film market. According to Futurism, the producers are now aiming for the first quarter of 2027. They haven't said what tool will replace Sora, and no distributor has been announced.
"Steelay" director Seol Woo-in also told JoongAng Daily that uncertainty over updates to foreign AI tools was the "biggest challenge." Kim Kwang-jib of Studio Meta-K calculated for Variety that a 100-minute film made with the most expensive tools already runs up more than 500 million won in compute costs today. Providers are currently selling their services below cost. If prices rose tenfold, an AI film would cost more than a regular Korean feature film. Lee Sang-wook of MBC C&I also warned that scenes made with year-old technology look "very old." That's exactly the criticism leveled at "Sanxingdui" because of its long development time.
The bottom line
Our take: the theory that "the technology didn't fail, the theater did" has a kernel of truth but doesn't go far enough. In its favor: a movie ticket costs money and time, and you have to get to the theater. On a phone, all it takes is a swipe, and short, eye-catching visuals work there without much of a plot. But the theater market wasn't the problem. In South Korea, cinema attendance actually rose 34 percent in the first half of 2026, according to KOFIC. So the AI films failed in a growing market. China's market is weaker: by November 15, it had taken in around 320 billion yuan, compared with 518 billion yuan for all of 2025. The animated film "Niu Lai" shows that theaters still leave room for surprises. Reportedly made for around $200 and mocked as "worse than AI," it grossed about $10 million, according to Dexerto.
A more honest summary goes like this: the technology is now good enough for pretty pictures, but not yet for 90 minutes that hold people's attention. At the same time, studios depend on providers that can change prices or shut down services. Whether Netflix and other streaming services will turn out to be the better venue is an open question. "중간계" points that way, but as a hybrid film with stars, it isn't clean evidence. Bona is pressing on regardless: its next AI film, "Unplanned: Operation Guanghan," has been in production since October and is slated for 2027. The first real test for a major AI production with a theater chain behind it is "Raphael," opening November 25. Judging by this year, there's little reason to expect it to turn things around.
Sources
- Fortune: China's first AI film Sanxingdui box office fortune.com
- Sina Tech: 首部AI电影进院线:博纳影业的长片实验 finance.sina.com.cn
- Sina Tech: 千万级投入、180万票房,首部AI院线电影何以遇冷? finance.sina.com.cn
- Huxiu: 首部AI院线电影票房180万元 m.huxiu.com
- Sina: 全AI电影《三星堆》票房惨败 k.sina.com.cn
- Sina: 全AI电影《三星堆》为何没人看 k.sina.cn
- Global Times: Sanxingdui reception globaltimes.cn
- Dexerto: China's first fully AI movie flops dexerto.com
- Tech42: AI가 찍은 영화, 640명만 봤다 tech42.co.kr
- Edaily: 외면 받은 AI 영화 edaily.co.kr
- Digital Inclusion News: AI 영화 화제성, 극장 표로는 이어지지 않았다 dginclusion.com
- Korea JoongAng Daily: AI can make movies cheaper, getting audiences to watch is another matter koreajoongangdaily.com
- Edaily English: Director Seol Woo-in interview (Steelay) en.edaily.co.kr
- Cine21: Steelay Detailinfo cine21.com
- Nate/Steelay: 제작비 10억으로 줄였지만 관객은 고작 4000명 news.nate.com
- Dailian: Steelay-Kritik dailian.co.kr
- Wikitree: 중간계 Netflix Platz 1 wikitree.co.kr
- Wikitree: 중간계 Netflix-Erfolg wikitree.co.kr
- Kyunghyang: 중간계 Kinostart khan.co.kr
- Korea Times: Review Raphael koreatimes.co.kr
- Korea Times: AI movies make leap from short clips to big screen koreatimes.co.kr
- Newspim: 라파엘 11월 25일 롯데시네마 개봉 newspim.com
- Variety/Yahoo: AI Could Make Filmmaking Costlier Than the 35mm Era yahoo.com
- The Next Web: Sora shutdown leaves Critterz at Cannes without its model thenextweb.com
- Futurism: OpenAI's attempt at an AI Pixar-style movie in shambles futurism.com
- FTNN: 白非空 96集 AI短劇 ftnn.com.tw
- Sohu: Foyege Harry-Potter-KI-Kurzfilm sohu.com
- 36Kr: DataEye-Halbjahresdaten zu KI-Dramen eu.36kr.com
- Xportsnews: KI-Kurzform vs. Realdreh xportsnews.com
- TechNode: China AI microdrama boom technode.com
- Asiae: Koreas Kinoumsatz im ersten Halbjahr 2026 view.asiae.co.kr
- Sina: 2026年度总票房破320亿 k.sina.cn
- Eastmoney: 博纳影业第二部AI电影项目开机 finance.eastmoney.com
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.