After the Golden Rooster awards empty out, AI films start talking about long-form narrative

November 12, Xiamen, the day the Golden Rooster awards closed. An event called "国贸 × Golden Rooster: New Film Creation Support Program" drew nearly three hundred people. Huo Tingxiao, vice chairman of the China Film Association, opened. At the front table sat 厦门国贸资本, 上影股份 and 智象元一.
A year ago the same kind of event meant a ten-minute concept reel and applause. This year the screen held a storyboard canvas, before-and-after green screen replacement, and continuous shots cut from a long series.
What they shipped is not complicated
智象元一 took 帧赞, its professional AI film platform, to 2.0. The pitch: film and long series, deeper support for director-led creation, shot language and complicated workflows, reaching into live-action ground with green screen replacement, effects enhancement, post adjustments, difficult shots. Two canvases: ideas, storyboards.
The other item is the model. HiDream-O1-Video-1.1 adds four things, per the notes: physical laws and cause and effect, narrative logic setting shot length, character consistency across shots, native audio-visual generation. 帧赞 2.0 runs entirely on it.
On their own these read like specs. Together they point at one sore spot: a beautiful ten-second shot is easy, a scene that holds is not.
Translating Mei Tao's sentence into producer language
Founder and CEO Mei Tao said on stage that the key to the next round has shifted from how one frame looks to long narrative and world consistency.
At a launch event it sounds like a slogan. In producer terms: the actor wears navy in shot A and cannot turn grey in shot C; a wound from scene three cannot heal by scene fifteen; the emotion earns a close-up at the third line, the model cuts it into three pieces because the prompt says "5 seconds", and the scene breaks.

Narrative logic deciding shot length bundles all that into one technical promise. Whether it holds, the films will say.
The product line is not new. HiDream-I1, open sourced in April 2025, topped Artificial Analysis within 24 hours. HiDream-O1-Video-1.0 came on September 15, the brand moved from 智象未来 to 智象元一 on October 8. Version 1.1 landed two months later, quicker than most teams in China.
The same day, another group was still arguing about "professionalism"
On November 9 a Golden Rooster market panel ran under the title "After equal access: the crisis of professionalism and the reconstruction of authorship in the AI image era". Teachers from the Beijing Film Academy, Xiamen University and the Communication University of China circled one question: with tools in every hand, where is a professional's value?
Liu Xiaowei of the Beijing Film Academy put live-action, virtual production, AI generation and traditional VFX into one pipeline. Chen Liyan of Xiamen University described wiring script breakdown, shot and asset generation and virtual set scheduling into production. Zhang Jingyu of the Communication University of China went from on-set scheduling to the many AI generation rounds.
One detail kept surfacing: prompts get longer. Wang Ke, co-founder of 智象元一, turned it around. If the way people and models talk has a ceiling, a stronger model still gets stuck.
That is Mei Tao's long narrative from the other side. Can the model remember the last scene? Can the director state his intent without contorting himself?
At Tsinghua, someone let slip a more practical change
On November 6, Tsinghua's School of Journalism and Communication hosted Tang Jili, Liu Shenleilei, Zhang Long of the 虎鲸文娱 AI screenwriting workshop, and Professor Hang Min.
Tang was more interesting than the models. Forty-six years in the industry, and every film and US series he is preparing starts with a three-and-a-half to four-minute AI preview. "With three and a half to four minutes per project, I can close the investment, the distribution and the cast."
The first problem AI solves on that chain is getting a decision early. The saving comes after. Pitching an investor used to mean money for a sample. Now a sample is cheap enough to hand over as a proposal.
Zhang Long gave numbers his team measured: with live performance in the mix, over 60% of shots in some projects can be done with AI. He also shut down the obvious inference. That does not mean AI can make a theatrical feature alone.
Liu Shenleilei's "Five Elements Mountain" follows a shepherd boy who carries peaches to the Monkey King, grows up wanting to split the mountain open, loses everything to an axe-selling scam online, and lets go only after he meets an anti-fraud platform. It was cited because it proves one thing: AI can do a spectacle, not a twist.
Eighteen films are the exam paper
The event closed with 18 first-round picks: shorts, series, AIGC features, theatrical films. Gao Xiongjie, head of film and television at the Central Academy of Drama, used "Blood-Dripping Flying Arrow" to explain how historical scenes and complex action get filled in with AI.
Those 18 report back over the next year. None of their gates involve leaderboards: getting into post, clearing review, selling tickets, balancing the cost sheet.
A model company talking about long narrative at a festival is a signal. Comparing frames no longer separates anyone. What separates them is pipeline, consistency, everything between scene one and scene ninety that nobody wants to shoot.
Whether 帧赞 2.0 becomes the main tool on those 18 films, the credits will say in a year.
Related articles
Six Amateurs, 1,113 Episodes, a Thousand Yuan of Compute an Episode: How an AI Drama Got Hand-Built
The first ten seasons of Armed Into Song run to roughly a thousand episodes and have passed ten billion cumulative views, with six core creators and not one of them film-school trained. Compute runs 1,000–2,000 yuan an episode; a normal shot takes three rolls, a battle twenty-plus, and anything that looks flat is graded a sixty and thrown out.
One server, one AI streamer: inside the code that runs a nightly livestream
open-yoki is the production code behind an AI host who streams nightly on Binance Square, with the persona, voice and money features stripped out. Headless Chrome draws the frame, ffmpeg pushes RTMP, a folder of Markdown decides what she feels and what she posts.
Pinky lets an agent drive your phone, and holds the Send button until you click Allow
Pinky connects any MCP agent to a real iPhone or Android phone and enforces approval in its own gateway rather than in a prompt. Banking apps are invisible, password fields are off limits, and on Android 13 the agent gets a hidden second screen.
ByteDance's SeedRealtime takes two frames a second and decides when to speak
Doubao-Seed-Realtime-1.0 folds camera frames, screen frames and speech into one full-duplex session, at no more than two frames per second. The interesting decisions are the frame rate, the proactive speaking, and the audit reply that ships switched on.