Newshot

A junior-high-educated baker single-handedly created a sci-fi series with 10 million views per episode — Chinese AI feature-length productions are going wild!

AI summary

Domestic AI tools make “one person is an entire film crew” a reality.

Translation(model translation, not the source's own text)

Junior-High Graduate Baker Single-Handedly Creates a Sci-Fi Series With Tens of Millions of Views Per Episode — Chinese AI Feature Films Are Going Wild!

Domestic AI tools make "one person = an entire film crew" a reality

Wen Le, from FeinOH, Quantum Bit | WeChat official account QbitAI

Quantum Bit | WeChat official account QbitAI

Junior-high graduate, former baker, cross-disciplinary self-taught, single-handed... and he made a sci-fi series with tens of millions of views per episode?!

No wait, aren't these buffs stacked a bit too high...

The creator Ling Yu directed this "Codename: Wukong," which puts Wukong into a futuristic Shanghai.

The waters of the Huangpu River surge upward, and a giant beast hundreds of meters tall breaks into Lujiazui. Wukong's divine powers were sealed long ago — so what does he do?

So this monkey builds himself a mecha and drives it right into battle.

AI videos that hit tens of millions of views aren't limited to this one.

The Chinese creative team Li Li An also went viral with a "pure Korean drama" made with AI.

This "Du Li An" puts a fox and a lynx into a school setting for a light comedy; a single video exceeded 15 million views on Douyin and raked in 960,000 likes.

After digging into it, it turns out these viral works all came from the same AI film and television creation competition.

Recently, the world's first AIGC event focused on film-grade feature-length creation, the RunningHub Global AIGC Feature Film Creation Competition, officially wrapped up.

Pulling up the results, to be honest, it's a bit frightening.

More than 1,500 works were collected globally, with long series and film-grade features accounting for over 80%, and more than 20 works exceeding 60 minutes in length.

Yes, the kind of feature films that can play in a cinema for over an hour.

Among these works emerged nearly a hundred hit videos with millions of views, plus dozens of phenomenon-level features that broke ten million.

Even more hardcore: the work that took the 1 million yuan cash grand prize, "Another Stratagem in the Empty City," along with the shortlisted "Huo Yuanjia," have officially entered the project development process for theatrical films.

AI video has bid farewell to the era of ten-second highlight spectacles and gacha-style generation, and has officially charged into the era of long-form narrative and industrialized mass production.

AI film and television creation has gone straight from wild-idea technical self-indulgence to the big screen.

So let's take a look at what's novel about these works. What exactly lets them break through the length limit, break the curse of visual collapse, and even draw crowds of professional overseas film and television teams?

AI Video Charges Into the Era of Long-Form Narrative

In the past we always thought AI video creation was just testing the waters.

Roll the gacha a few times, cobble together a ten-second short to play with, post it on Moments, done!

But when you actually try to make a feature, you realize it's a completely different matter. If you imagine it as simply stacking up the number of gacha rolls for a short film, you'd be badly mistaken...

The first thing creators face is the script problem.

How do you break a tens-of-thousands-of-words literary work into industrial-grade cinematic language that AI can understand?

A film or series is often adapted from a novel of tens of thousands or even hundreds of thousands of words, and turning it into a script requires an extremely tedious structuring process.

In a traditional production pipeline, this step requires collaboration among screenwriters, directors, storyboard artists and others, taking weeks or even months of repeated meetings and fine-tuning to convert text into hundreds of executable shot scripts.

And the entire process is filled with subjective judgment, with very high communication costs.

I've seen too many teams whose scripts are a tangled mess, whose storyboard prompts spin out of control, and whose projects die right at the first step.

The works in this competition, however, cracked exactly this hard problem.

By the way, let me give you a tip: these hit features that combine narrative stickiness with film-grade audiovisuals all run on the AI film and television industrial pipeline built by RunningHub under the Alibaba Cloud umbrella.

For example, the Best Short Series award winner "This Is My World," a series made with AIGC by director Lin Shan, a veteran creative with over a decade of experience; after netizens reposted it to YouTube, a single episode also broke a million views.

Being able to organize a massive story web neatly brings us to RunningHub's RHTV Script Workshop.

This thing supports long scripts of up to 100,000 words; you just throw the novel in, and it parses it in one click into an outline, then scenes, then standardized storyboard prompts.

It's not simple text summarization — it deeply understands dramatic structure, identifies character relationships, plot turning points and emotional curves, then translates them into industrial-grade storyboard language including Scene, Shot, Action and Dialogue.

And it supports multi-user online collaboration with a shared asset library.

Screenwriters, directors, AI producers — everyone works against one set of assets, eliminating the first industrial breakpoint from literature to shots.

With the script problem solved, the next roadblock is shooting.

For example, "Codename: Wukong," which won the "Most Promising Sci-Fi IP Award," saw creator Ling Yu make core use of RHTV's infinite canvas capability.

There's a segment in the series where the audience sees Wukong piloting a mecha to fight a giant beast, but when Ling Yu was editing, he got tripped up by a small wooden puppet in a child's hand.

In the second episode, a mother and son evacuate from the Bund.

The child clutches the puppet the whole way. Each individual piece of footage looked fine, but once cut together, it turned out that as they ran, the puppet in his hand grew a size larger?

There was an even more absurd version: the two were running toward the street one second, and with a cut, they had directly turned around...

To fix this kind of continuity error, at first Ling Yu could only consult real-world maps, reference live-action photos, and hand-draw storyboards to check bit by bit — time-consuming and laborious.

But after using RHTV's 3D director's stage, the approach was completely different:

First place two character stand-ins in the scene, set up virtual camera positions, and have the characters rehearse along the evacuation route. With synchronized dual-camera preview, whether the characters' direction of travel and spatial logic are correct can be seen at a glance, preventing shot-continuity failures at the source.

Simple blocking is one thing, but what about more complex fight scenes and large-scale camera movement — how do you avoid blind gacha-style generation?

Traditional pure 2D text-to-video is very prone to "spatial loss of control" when handling spatial trajectories and fight blocking — put plainly, it just does whatever.

You tell a character to "run from left to right across an exploding street," and the AI might have him run straight through a car, or the camera might suddenly cut in from some baffling angle... the character's movement trajectory has no logic at all, basically opening a blind box.

But several works this time displayed textbook-level camera work.

In the Best Film award winner "Bronze Guardian" (creators: Mao Daren Ai Shui Da Jiao / Guangying Jihe Animation Studio), the subtitle has just read "Three months later" when a wanted poster is swept by the wind right in front of the camera.

Only after the poster drifts away does the architecture at the end of the sand sea reveal itself.

The camera enters the city, then slips into a narrow alley hung with red cloth awnings. A small robot rolls past in the foreground, and a woman in black walks out from deep in the crowd. The camera gives us her footsteps and her face, then cuts to her back as she walks into a pitch-black archway.

Both of these segments are well suited to being previsualized with 3D white models first.

The work combines an epic feel with Eastern aesthetics; its scene staging is complex and grand, its views approach ten million, and it even drew a professional production team from Netflix to gather around and study it.

This brings us to another big killer of RunningHub: the RHTV cloud set.

Put simply and bluntly, RunningHub moves Blender directly to the cloud — no installation, open it and use it.

Then, whatever you want to build, whatever camera positions you want to place, however you want to stage it, you just speak in plain language and you can command an Agent to build the 3D scene and plan camera movement trajectories for you in the cloud.

If you need to fine-tune details, you can seamlessly switch to Blender's native interface at any time for manual adjustment.

For example, in the segment above, the wanted poster first uses a thin slab as a placeholder, to test where it should fly and flip so the audience can just make out those characters, while also revealing the distant architecture as it drifts away.

At the market, the small robot first uses a short cylinder, the woman and passersby use simple stick figures, and the archway is only roughed out.

All the complex staging — multi-character blocking, complex one-take camera movement — is first rehearsed in 3D space with simple white models.

First test when the robot rolls out of the camera's view, then switch the camera to behind the woman, to see whether the audience can still follow her footsteps to find that door.

Once these white models can get the blocking straight, then let the AI "start shooting," to create the effect of paper edges fluttering wildly in the wind, red cloth awnings swaying in the alley, and the light dimming as the woman steps from the bright sun into the doorway.

The spatial relationships, character paths and camera paths determined during the white-model previz stage themselves become reusable production assets, directly serving as spatial and compositional references for downstream video generation nodes.

The entire process closes the loop within the same canvas, with no information loss from cross-software export, cutting pre-production costs directly by 50%.

This isn't gacha anymore — this is a director commanding on set.

Image Production Moves Toward Stable Mass Production

Alright, pre-production is done. Moving into mid- and post-production, an even more headache-inducing problem arrives.

A feature film often has dozens of scenes, with ever-changing lighting, angles and shot sizes — how do you ensure the protagonist doesn't "change faces" throughout?

Article title: A junior high school graduate baker single-handedly created a sci-fi series with tens of millions of views per episode, China's AI feature films are going wild!

Main text: How do you eradicate those画面崩坏 that keep popping up from time to time?

Also, AI-generated characters generally lack delicate and realistic micro-expressions. What if their demeanor is stiff?

This is a long-standing difficult problem for AI video.

Several award-winning works in this competition gave quite beautiful answers.

The Korean drama-style short series "Duli'an" mentioned at the beginning is planned for 5-8 episodes in total, 4-6 minutes per episode, with anthropomorphic foxes and lynxes as the protagonists.

The work does not make animal ears and tails into mere decorative special effects, but instead strongly binds animal features to character personality and plot emotions. The dynamics of the animal ears and tails themselves are the expression of the character's emotions.

At the same time, the series has extremely high requirements for character consistency.

In shots spanning multiple episodes, the shape, size, fur texture of the ears and tails, as well as the arcs of their swaying with emotions, must be stable and controllable, avoiding collapse problems such as sudden changes in tail size and fur texture appearing and disappearing intermittently. This requires precise local constraints and fine-tuning to achieve long-sequence stable output of live-action foundations overlaid with animalized features.

"Yi Xie Lu" (creator: Yang Yu) also achieved a high degree of stability in the Eastern supernatural art style and character states across multiple episodes and scenes.

The most exaggerated is "Lonely Whale." The creator, Yesheng Guo, single-handedly, in one month, at an extremely low cost of only 15,000 yuan, independently completed a complete 78-minute feature film——

Topping Bilibili's trending searches, with over ten million views across the internet.

This kind of cost and efficiency is simply flipping the table.

As for how they locked down consistency? It is a combination punch.

The first thing to mention is pixel-level precise local repair.

Previously, when AI generated a video, as long as one frame or one local part was broken, sorry, the whole segment was scrapped and had to be redone from scratch.

But the newly generated result might fix the hand while the face collapses again...

Now RHTV supports frame extraction second by second and local redrawing on the canvas, repairing whatever is broken wherever it is broken.

This is like the healing brush in Photoshop, saying goodbye to that desperate rework of robbing Peter to pay Paul.

At this point, you may think RunningHub has merely made a very powerful toolset.

But the real barrier goes far beyond this.

Dang Jinfeng, founder and CEO of HaiMaYun, has a judgment:

Consumer-grade AI is very difficult to make money from. What can truly land and monetize are only vertical productivity scenarios. The essence of the next stage of AI is no longer divergent probabilistic generation, but deterministic system delivery. Anyone can buy models. What truly creates the gap is compute power and Harness, that is, a "productivity engine" in which agents, context, and toolchains are deeply meshed with data.

All RunningHub tools are built on this AI productivity engine.

It is essentially a self-consistent five-layer capability stack:

At the very bottom is HaiMaYun's unique integrated rendering and inference compute layer, which relies on intelligent scheduling to raise GPU utilization by 70%, fundamentally lowering production costs;

Above that is the model layer. They do not make general large models, but focus on vertical models such as H3 RH Enhanced that have undergone scenario-based post-training, specifically solving practical problems in film and television industrialization;

At the core is the Harness system orchestration layer. It is like a neural network, scheduling the entire toolchain through agents, from LoRA character assets to cloud Blender, automatically completing complex creative tasks;

Supporting all of this operation is the data layer, which continuously absorbs the operational trajectories of professional creators on the platform, forming a flywheel loop that gets smarter the more it is used;

Finally, all capabilities converge at the application layer, namely products we see such as RHSTORY and RHTV Cloud Studio, delivered directly to creators.

This five-layer architecture is the confidence behind RunningHub's ability to support industrial-grade AI feature film creation.

The technical pipeline has brought down the production cost of feature films and greatly shortened the production cycle.

This also truly realizes "one person is a full-process cinematic crew."

But AI directors immediately face a more realistic ultimate question:

The feature film is made, where do we monetize it? Who pays for the huge compute costs?

Connecting literary IP, celebrity image licensing, and commercial distribution channels

Short videos generate traffic, and the next one has a chance to take ads. The monetization chain is very short.

But to shoot a feature film, the money for script, storyboard, and compute has to be spent first, while payment can only come after the film is finished and distribution channels are found.

The author works hard for months, and there may already be a movie lying on the hard drive, yet the first payment has not yet entered the account.

Even if this time you grit your teeth and carry through relying on passion, what about the next one?

Who do you talk to about adaptation rights for a novel you like, where does the compute for production come from, and who do you hand it to for distribution after filming?

These resources all require creators to contact them one by one and figure them out on their own.

Many independent creators are not short on creativity and talent, but are stuck in the same dilemma:

No matter how good the idea is, there are no supporting conditions to realize it; no matter how good the film is, there is no channel to reach the audience.

Lacking IP sources, lacking compute support, lacking distribution channels, many people can only create purely out of love, and it is difficult to form a sustainable creative cycle.

This time, through the competition, RunningHub integrates these originally scattered resources together, with the goal of building a complete ecosystem——

A full-stack ecosystem closed loop of IP assets + industrial pipeline + distribution studio.

First is IP. Everyone who often shoots Hollywood movies knows (doge), IP is the most valuable.

HaiMaYun has directly built the IP library for you.

They exclusively signed AIGC exclusive adaptation rights for 6 science fiction works including "Xu of Time" and "The Eye of Saishteng," and also linked with the Nebula Award and Tianwen Award to establish an IP co-creation library.

In terms of celebrity digital image cooperation, actress Irene Wan is one of RunningHub's first batch of cooperating artists.

Her licensed digital image is about to enter the RunningHub artist asset library, for creators to use within the authorized scope for AIGC film and television creation.

HaiMaYun launched a studio plan.

It plans to invest hundreds of millions in cash + unlimited compute to escort the entire process and advance 10 theatrical-level films, 100 premium IP series, and 5 film-game interactive works.

Once the film is made, it cannot just stay on the hard drive.

RunningHub has partnered with mainstream platforms such as iQIYI and Tencent, connecting production, distribution, and revenue-sharing links, promoting projects such as "Another Scheme in the Empty City" and "Huo Yuanjia" from online hits directly to the big screen.

You see, from IP sources to industrial production pipelines to commercial distribution, RunningHub has connected the entire chain.

Over the past hundred years, turning wild imagination into theatrical-level imagery has always been the "privilege" of a few.

The threshold of film and television creation has kept many creators with good stories outside the door.

Today, the digital studio belonging to everyone has officially started filming!

  • Double 11 livestream room transformed into an AI show, Luo Yonghao tries TRAE and instantly becomes the public mouthpiece for AI Coding 2026-11-11
  • Tencent Hunyuan joins the strongest open-source models! Hands-on test of Hy4: paired with WorkBuddy, multimodal Agent delivery power is maxed out 2026-11-09
  • ByteDance found the reason why DeepSeek is sometimes strong and sometimes weak 2026-10-09
  • arXiv's strictest new rule! Each person can submit at most 2 papers per month, and rejected papers do not return quota 2026-10-02

Related reading

No way, do you have to ChatGPT everything when making cars now?

AIGC is changing production models

Jia Zhangke and Li Shaohong are both making movies with AI, and even sharing experience: which feature is most practical?

Famous directors + AIGCers + Kling = recommendation/pitfall-avoidance practical posts

2025 Beijing Culture Forum AI Salon grandly held! Haidian takes the lead in breaking the "hundred-model" scale, with 105 large models accounting for one fifth of the country

Focusing on "AI reshaping digital content production"

A Junior High School Graduate Baker Single-Handedly Creates a Sci-Fi Series With Tens of Millions of Views per Episode—Chinese AI Feature-Length Films Are Going Wild!

Latest Agenda Announced! | 8.25–26, Gather in Shanghai for the First International Conference on AI-Generated Content!

1 Billion Yuan Poured Into R&D—What Exactly Is Xiaoice Up To?

DALL·E 3 Begins Testing on Bing! The “Horse Riding an Astronaut” Challenge Is Cracked; Specify 50 Objects in One Image

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by 量子位
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗