AI video · Case study

Chixie Rusong: a slow start that compounded, and the account nobody has settled

Start with the platform fields visible in two season pages (the AI filing number, the AI-generated label, per-season collections), then the climb timeline, the division of labor, and the boundary of the money story. We have not watched every episode.

Published · Reviewed

A Chixie Rusong still: in a Song-style command tent, an armored commander sits behind a map table flanked by officers.Watch on Hongguo (search the title inside the app; no stable work-page URL to cite) ↗
A horizontal still from the war-council scene, taken from a press republication rather than captured frame-by-frame here; the series streams free inside Hongguo's app.Poster source: Still as republished on Shanghai Observer's report page; the original content is the Hongguo drama. · Watch on Hongguo (search the title inside the app; no stable work-page URL to cite)
Platform response · first ten seasonsPeak season heat 82.06m · reported on-platform collections 11.41m+

Heat and collections are platform fields — not plays and not revenue; 12 billion is the reported cross-platform cumulative-play figure of November 6. On the same day a creator said the "ten-million return" was a media estimate and inaccurate.

Check this result
Read the context and editorial assessment

A five-person modern squad crosses into the Northern Song with their gear and no system cheat: guns jam and ammunition runs out. This hardcore premise launched quietly, then compounded season by season into 2026's most-discussed AI drama. Its six makers are ordinary people who drilled on commercial gigs for a year and built slowly to a hit standard without buying a single traffic impression. The evidence below puts the climb back on a season timeline, the division of labor back in the makers' own words, and the money story back inside its real boundaries.

Follow the projectContents7

Series progress · from public sources

Season one to eleven: the slow-compounding timeline

Each point returns to reporting or platform fields. Heat, plays, collections and reservations are four different measures and cannot be added.

  1. Season one · a quiet start

    Released ·

    Preparation began August 1 with about seven days of crew selection; two people finished the first ten episodes in a week and the season launched in roughly three. Producer Li Zhaohua later recalled a flat debut — collections around 600–700k.

    Audience and reach
    Early numbers were flat; the season-one page now shows 3.705m collections (a November 9 screenshot), accumulated over months.
    Maker’s choice
    The team aimed for a premium hit from day one, spending longer on character sheets, scene art and script adaptation than on any earlier show.
  2. Season four · top of the chart overnight

    Released

    The day after season four launched, it reached first place on Hongguo's AI new-drama chart with heat above 50m — the first verifiable turning point.

    Maker’s choice
    The rise was not a single step: heat compounded season by season, and only at season nine did the team itself feel how big it had become.
  3. Seasons nine and ten · quality and heat together

    Released

    Season ten hit 82.06m heat, the series' peak; season eleven passed 4.83m reservations before its expected November 15 launch. Season pages carry the filing number and an AI-generated label.

    Audience and reach
    Viewers began sharing Song-history trivia in the bullet comments, built annotation platforms and marked official ranks and battle maps as fact, fiction or conjecture; the audience reaction to Yue Fei's entrance surprised the makers.
    Maker’s choice
    Production quality visibly improved in seasons nine and ten; the team's Seedance video model upgraded to 2.5, whose multi-reference ability helped crowd scenes and battle spectacles.
  4. November 4–6 · breaking out to 12 billion

    Released

    On November 4 a Weibo topic about Hongguo historical dramas trended; on November 6 plays were reported at 12 billion with the top spot on Hongguo's AI chart. Earlier calibers: 4 billion (October reports) and 6 billion (November 3). These snapshots cannot be summed.

    Audience and reach
    The state-media wave followed: Outlook carried a commentary republished by CCTV, and Economic Daily and Guangming Net both praised it on the same line — lower costs do not mean lower standards.
  5. The same week · the account stalls

    Production reported by maker

    Compute spend above one million yuan is creator-confirmed (the OPC model carries no payroll, and the novel rights were selected free from Fanqie's rights library); the "ten-million-plus return" is a media estimate. Zhang Hansheng, November 6: the ten-million figure is a media estimate and inaccurate — he had never disclosed income and had not received that much.

    Maker’s choice
    The team says it has never bought traffic and will not: finish, clear review, publish across platforms.

Evidence limit Heat, collections, plays and reservations are four different platform or media measures. We have not watched every episode and no income evidence exists. Screenshots come from press pages dated November 9; platform numbers keep moving.

01

Verified material

Read the frames and the platform fields first

Two season screenshots carry the filing number, the AI label, episode counts and per-season collections; heat, plays, collections and reservations are four different measures.

Put two screens side by side. Season one, episode one: the commander Lin Zhao's character card, the synopsis line, the footer label "creator's statement: content is AI-generated", and 3.705m season collections. Season ten, episode 42: the filing banner "AI (Fanqie) micro-drama filing (2026) No. 161465", a picker reading "all 100 episodes · free", and 878k collections on that episode. Both screenshots come from Shanghai Observer's November 9 report page; platform numbers have kept moving since.

Then separate four measures: Hongguo "heat" is the platform's own composite (peak 82.06m a season), not plays; "12 billion plays" is the November 6 reported cross-platform cumulative caliber, against 4 billion in October and 6 billion on November 3 — snapshots that cannot be summed; "11.41m collections" is on-platform behavior; "4.83m reservations" is pre-launch booking for season eleven. We have not watched every episode; we checked the fields visible in screenshots and public reporting.

Hongguo's season-one episode-one vertical page: commander Lin Zhao's close-up card, the footer label "creator's statement: content is AI-generated", and 3.705m collections.Check visual source ↗
This is what the start looked like: episode one of a drama later reported at 12 billion plays is a wounded commander's close-up plus one AI label. The visible fields are what we checked; collections change over time. · Platform screenshot as republished by Shanghai Observer (November 9, 2026) · Original material
Hongguo's season-ten episode-42 vertical page: a Song-style command-tent council, the filing banner, and "all 100 episodes · free".Check visual source ↗
Beside season one, season ten's lighting and crowd work have visibly moved — the makers credit Seedance 2.5's multi-reference upgrade and a season-by-season asset library. The filing banner shows the micro-drama filing track, and the platform mandates the AI label. · Platform screenshot as republished by Shanghai Observer (November 9, 2026) · Original material

Sources for this section: Shanghai Observer (Liberation Daily): six people hand-build an AI hit — in conversation · Wenyu Value Officer's creator interview (published on 36Kr) · Lianhe Zaobao: plays pass 12 billion

02

Public account

The pick: why a sub-100k-click novel was worth making

A free rights-library pick, three selection tests, and a visual route only 7.6% of the chart takes.

Zhang Hansheng picks IP from the creators' rights library; the chosen novel sat under 100k clicks. His three tests: a time-travel premise with no system cheat, so the lead must work within historical rules; an author meticulous about Song institutions, offices and dialogue; and prose that reads upright and hot-blooded enough to move an audience. Rights via the library were free — which also contradicts outside guesses of a six-figure rights bill.

The visual route also runs against the market. DataEye counted 500 AI dramas and comic dramas on Hongguo's September board: 460 were 3D comic dramas (92%) against just 38 realistic-human dramas (7.6%). This show took the minority path — its villages, camps, battlefields and courts borrow the visual memory audiences bring to historical drama, which a fantasy 3D look cannot. The naming split sits here too: readers and many reports say "AI comic drama", Hongguo's filing says "AI micro-drama", and industry analysis files it by look as a "realistic-human drama". This case records the platform and on-screen terms; all three names point at the same show.

Sources for this section: Wenyu Value Officer's creator interview (published on 36Kr) · Zhihu column (Luogu Yinhe): consistency as an engineering problem

03

Public account

Production: the work happens before generation

A four-step pipeline, an all-character asset library, historical fact-checking, and the economics of card-pulling.

The pipeline the makers describe has four steps — read the novel, iterate the script, refine character designs, lock the tone with a sample reel — and none of them is called generation. The calendar is front-loaded too: prep from August 1, about seven days picking the crew (five or seven were considered, six chosen), the first twenty minutes done by August 9, the first ten episodes by two people in a week, the season up in roughly three. Every speaking character gets dedicated reference material — "even without a name tag the audience should know who this is" — with settings and props in one shared library.

Two layers of checking decide whether frames read as Song: official robe colors, accessories, footwear, household permits and rank etiquette were verified, and the Li Qingzhao-inspired character Li Qingtang was tuned against surviving period paintings. Only then comes the economics of card-pulling: routine shots average 3 generations, battle spectacles and crowded frames often need 10 to 20-plus; a technically clean but unmoving "60-point take" is ruled a discard and re-rolled. Project compute topped one million yuan — under OPC nearly the only cost, the makers note: no payroll, free rights, Seedance annual memberships, with the 2.5 multi-reference upgrade visibly helping crowd scenes.

Sources for this section: Wenyu Value Officer's creator interview (published on 36Kr) · China Writers Net relaying Beijing Youth Daily: AI is the tool, content is the hand · Zhihu column (Luogu Yinhe): consistency as an engineering problem

04

Public account

Organization: how six people keep a thousand episodes coming

OPC selection plus self-drive, revenue by finished minutes, comment sections as the writers' room; audience feedback reshapes later seasons.

The network holds 30-plus people across six provinces working fully online; the core six — four men, two women, including a stay-at-home mom, a former ride-hail driver, an office worker and a wholesale trader — were filtered from over a year of collaboration. The form is OPC, one person one company: no fixed salaries, revenue split by role and effective finished minutes, each production self-funding, entry gated on mature finished-work ability with trial pieces as the exit ramp. Coordination runs on WeChat, Feishu, shared drives and Tencent Meetings under a no-meeting-by-default rule; finished cuts get cross-reviewed, and a process owner signs off.

Iteration plugs straight into the audience: no script stockpile, only an outline, with each season adjusted from comment-section and fan-group feedback — the comments are the writers' room. Audience effort landed on fertile ground: Song-history bullet comments, volunteer three-tier annotation maps, and state-media approval (Outlook, Economic Daily, Guangming Net, all on one line: AI cuts costs but never standards) together turned an AI drama filed under "brain-rot" into a口碑 sample. The team answered by moving production further upstream: seasons nine and ten show visible lighting and texture gains, and the eleventh season, expected November 15, already holds 4.83m reservations.

Sources for this section: China Writers Net relaying Beijing Youth Daily: AI is the tool, content is the hand · Shanghai Observer (Liberation Daily): six people hand-build an AI hit — in conversation · Wenyu Value Officer's creator interview (published on 36Kr) · Sohu commentary: the team disputes the million-in, ten-million-out story

05

Evidence still needed

The account: what is confirmed, what was denied, what nobody knows

Compute above a million is the makers' figure; the "ten-million return" was denied by them; why they continued through a flat first season explains more than any revenue number.

Confirmed caliber: compute spend above one million yuan (creator-confirmed; with no payroll and free rights, compute is nearly the only cost), plus the platform fields above. Denied caliber: "the title earned over ten million" is a media estimate and relay; Zhang Hansheng on November 6: the ten-million return is a media estimate and inaccurate — he had never disclosed income and had not received that much. A Sohu commentary adds the team's own line that earnings have not been tallied. As of this round, no verifiable income document is public.

More informative than any return figure is the season-one decision: a flat debut and 600–700k collections should have ended the series by any ROI logic — the team kept going without buying traffic, climbing on completion and engagement inside the recommendation system until season nine confirmed the break-out. The follow-up plan is likewise testimony, not evidence: monthly output drops from twenty-plus quick titles to single digits while per-title compute rises; the new bodyguard drama launched October 23 and passed a million collections (reported caliber). Reading this as an inevitable grass-roots myth misleads — Zhang Hansheng repeats that luck was a large part, compute demands upfront spend, and learning the tools guarantees nothing.

Sources for this section: Wenyu Value Officer's creator interview (published on 36Kr) · Sohu commentary: the team disputes the million-in, ten-million-out story · 36Kr carrying Chief Business Review: one title earns ten million — why it won't repeat

A BitShovel exercise · not the maker’s tutorial

Put the front-loading to work on one scene of your own

A small BitShovel exercise from the public interviews, not the Laoshiyi workflow. The goal: settle what a scene needs before opening any generation tool.

Before you begin

  • An original or clearly licensed short story; free rights-library channels cannot be assumed everywhere, so confirm rights first.
  • Set your own generation budget: how many attempts and how much money per shot.
  1. Three questions before committing

    Ask as Zhang Hansheng did when choosing the novel: does the lead carry a system-style cheat? Do details survive scrutiny? Does the dialogue hold up? Write the answers before committing.

    Keep
    The three answers and any rejection reasons
    Check
    If any answer gives you pause, swap the story first.
  2. Pin the people down first

    Write a character card and reference sheet for everyone with a line — bit parts included — plus baseline images for each setting and prop, all into one asset library.

    Keep
    Character cards, sheets and setting/prop baselines
    Check
    With names hidden, another person can tell who is who; bit characters are as recognizable as leads.
  3. Budget the card-pulls

    Split shots into routine and spectacle, budget per the makers' account (routine within about 3 pulls, complex 10–20), and when over budget revise prompts and shot design instead of pulling again.

    Keep
    A shot budget and the actual spend log
    Check
    For each overrun: a design problem returns to the previous step; luck alone justifies one more pull.
  4. Judge it as an audience member first

    Laugh first at your own comedy, cry first at your own tragedy: show the cut to someone who has not seen the making-of; they must retell the plot, recognize everyone and name what moved them.

    Keep
    Outside feedback and a revision list
    Check
    A shot that is technically clean but moves nobody counts as a discard by the makers' own standard.

Make one final choice: If the three answers wobble, the asset library will not hold or the pull budget is out of control, shrink the subject or stop. Understanding, sustainability and clear rights come before any next scene.

The exercise design and checks are BitShovel's, informed by public interviews; they are not the team's prompts, asset files or revenue model.

Sources and limits

Where the material comes from

  • Wenyu Value Officer's creator interview (published on 36Kr)
    Creator interview · conducted November 6, published days laterChecked

    The climb line (flat season one → next-day chart-top at season four → the team's own season-nine realization → 82.06m in season ten → 12 billion reported November 6), the verbatim income disclaimer, the million-yuan compute with free rights, Seedance annual memberships, two earlier shows plus the new bodyguard title, and the no-traffic-buys statement. This case's backbone.

  • China Writers Net relaying Beijing Youth Daily: AI is the tool, content is the hand
    Interview relay · November 10, 2026Checked

    The 12-billion/11.41m/82.06m calibers; the 30-plus team with a six-person core; the three-week prep detail; no-meeting coordination and cross-checked asset libraries; front-loaded effort and OPC selection-plus-self-drive; the Qin Hongying scene and Yue Fei's entrance from the makers' side; the output cut and the warning to newcomers.

  • Shanghai Observer (Liberation Daily): six people hand-build an AI hit — in conversation
    Creator interview · November 9, 2026Checked

    The squad's five names and premise, the August 1 start and crew tests, the seven-step pipeline, the be-your-own-audience method (laugh first, cry first), the audience response to Yue Fei and Li Qingzhao, season eleven's 4.83m reservations ahead of November 15, and the origin of both screenshots.

  • Zhihu column (Luogu Yinhe): consistency as an engineering problem
    Third-party analysis · November 8, 2026Checked

    DataEye's 92/7.6 September form split on Hongguo, the four-step pipeline summary, the all-character asset library, card-pull figures, historical checking with the Li Qingtang painting reference, the two-minute fault-tolerance contrast with the Sanxingdui feature, and Hongguo's 430m MAU (QuestMobile). Analysis caliber, not the makers'; cited separately throughout.

  • Sohu commentary: the team disputes the million-in, ten-million-out story
    Commentary (self-media) · November 7, 2026Checked

    Relays Li Zhaohua on the flat season one (600–700k collections, the season-nine realization) and Zhang Hansheng's "we have not had time to calculate returns". An opinion piece; its factual claims were cross-checked against the interviews before use.

  • 36Kr carrying Chief Business Review: one title earns ten million — why it won't repeat
    Industry commentary relay · November 2026Checked

    Industry context: 221.9k new AI short dramas in H1 2026 with 1,055 breaking 100m plays (about 0.5%); the Seedance/Kling model ecosystem and platform support plans; the regulator's review-everything direction; concept stocks and where profit accrues. Commentary caliber, kept separate from the makers'.

  • Lianhe Zaobao: plays pass 12 billion
    Press report · November 7, 2026Checked

    A timestamp reference for the 12-billion caliber (November 6/7); checked via search summary only and used solely as timing corroboration.