BitShovel
中

AI video · Case study

The Accident: how two students built a serial drama after an earlier attempt stalled

An earlier series that did not take off, a trailer that attracted attention, and the difficulties of keeping a serial story on screen.

Published · Reviewed

The Accident title over a school corridor, from an interview with the creators.Watch original film ↗
Watch episode one, then read how its makers handled characters, blocking and unexpected generated footage.Poster source: Image from Xiaoyunque's creator interview, published by Sina News · Poster source · Watch original film
Reported audience response · interview dateFour episodes, roughly 30–40m views and over 2m likes

Figures come from an interview produced by Xiaoyunque, which supports the project, not from platform analytics inspected by us. Views and likes are not revenue.

Check this result
Read the context and editorial assessment

This AI campus drama offers characters and a story to watch, plus a useful creation trail: the two makers say their earlier collaboration failed to gain traction, while a new trailer drew attention. Their interview explains writing, generation and visible errors, but not production cost or receipts.

Follow the projectContents6
01

Public account

Watch the episode, then read the reach claim

Episode one is a creator post; cumulative figures come from a supporting platform's interview.

whata-ha posted episode one on August 14, 2026 and credited Gouyuansuanna as co-creator. Episode two followed on the same account. The original episodes are the first-hand record for judging performance, pacing and story.

A September 4 interview published by Xiaoyunque said four episodes had gathered roughly 30–40 million views and more than two million likes. The original posts and interview show attention, while the interview does not provide analytics that can be independently recalculated.

Sources for this section: whata-ha: The Accident EP01 · whata-ha: The Accident EP02 · Xiaoyunque interview with both The Accident creators

02

Public account

Why the creators changed to an American high-school story

The makers say their previous China-set series did not take off; a different trailer led to the show.

In the Xiaoyunque interview, the makers say their first, China-set collaboration did not take off. A later US-style trailer by whata-ha drew attention, so they developed it into a serial story. Without comparable analytics, the new premise alone cannot be named as the sole cause of its reach.

The baseball captain and school-paper editor came from the makers' viewing habits and taste. They describe an ensemble town whose characters have lives and growth beyond a popular look.

Sources for this section: Xiaoyunque interview with both The Accident creators

03

Public account

Who wrote the script and who made the shots

The interview describes collaboration, references and prompt thinking, but not the complete production files.

According to the two makers, they discuss episode arcs, character moments and cliffhangers together. whata-ha writes the script, Gouyuansuanna generates shots, and the footage goes back for editing. They say the script is mainly human-written, Doubao assists some translation and a model produces the English voices.

The interview says they keep base character images on a canvas and reconfirm new looks for each episode. Gouyuansuanna describes prompting for a character's state of mind, rather than issuing only a facial-action command. The article shares an approach and excerpts, not enough to reconstruct a full original prompt or settings for a shot.

The interview says most video shots used Seedance 2.5. This is the makers' account in a vendor-published interview; we have not inspected their project files or treated vendor pricing and claims as this show's actual cost.

Sources for this section: Xiaoyunque interview with both The Accident creators

04

Public account

Drifting positions, continuity errors and a useful accident

The makers describe specific problems and fixes; generation-by-generation comparisons are not public.

Gouyuansuanna says characters sometimes sat in the wrong place or jumped from the back of a classroom to the front; background students were also hard to keep consistent. One fix was to cut to a close-up so the unstable background left the frame. After attention grew, some shots took seven or eight generations, while others took one to three. These are the maker's recollections, not our generation log.

In one first-episode shot, a character moved toward another faster than scripted. The makers kept the deviation, and the interview says it became a widely shared moment. The useful decision is whether an unexpected result serves the character and story, not treating every loss of control as creative success.

Sources for this section: Xiaoyunque interview with both The Accident creators · whata-ha: The Accident EP01

05

Evidence still needed

What attention did and did not pay for

The interview reports platform credit support; spending, sponsorship and revenue are not disclosed.

The Xiaoyunque interview says the two joined its featured-IP support programme and received credits up front. That can cover some generation use, but the article does not disclose credit volume, total cost, sponsorship contracts, platform revenue sharing or money received. Views and likes cannot stand in for those figures.

A comparable trial would begin with a short scene that tests character and story, record generation spending and rejected shots, then inspect actual audience response and platform terms. Premise validation, production efficiency and earnings require separate evidence.

Sources for this section: Xiaoyunque interview with both The Accident creators · whata-ha: The Accident EP01

A BitShovel exercise · not the maker’s tutorial

Test an original scene that could become a series

BitShovel designed this trial from The Accident makers' account of changing premise, sharing production roles and handling blocking errors. Test one short teaser or scene for a character relationship before committing to a series. This is not their shot log or original prompt, and teaser response alone cannot prove a premise will succeed.

Before you begin

  • Use an original story you may make, one manageable setting and a basic description of two characters. Do not reuse the film's characters, dialogue or campus imagery.
  • Have still, short-video and editing tools. Set duration, a two-retry limit and a budget cap, and assign writing, shot making and final story judgment.

  1. Write the relationship and one defining action

    In three sentences state what each person wants, why their relationship has tension and which question remains at the end. Choose an action that reveals character rather than starting with a trending premise or attractive face, then board setup, action and reaction.

    Keep
    A one-page original scene card, three-shot script and one sentence the viewer should understand.
    Check
    Without a character introduction, do the three shots convey the relationship and unresolved question? Revise the script before motion if they do not.

  2. Lock character identity, blocking and state of mind

    Prepare authorised references for each person and state clothing, position and current state of mind. Draw a simple left-right blocking map; for each shot note who sits where, movement direction and whether background people are needed.

    Keep
    Two character cards, one blocking map and three storyboard stills.
    Check
    Are both people recognisable across frames? Has anyone jumped position, or changed clothes or eyeline without cause? Fix the stills before motion.

  3. Generate short shots and handle actual continuity faults

    Ask for one visible action per shot, saving each input, candidate, tool, attempt count and rejection reason. If background people or seating drift, try a closer frame, but check that it still explains the relationship. Keep an unexpected movement only when it helps the story.

    Keep
    Three shot candidates, an error log and a keep-or-redo reason for each.
    Check
    Check people, blocking, background and emotion separately. After two failed retries, reduce background complexity or reframe; an attractive accident is not automatically a good story beat.

  4. Cut a teaser and ask what viewers want to know

    Edit a short teaser from usable shots. Ask two people new to the story who they recognised, what relationship they saw and what they want to know next; note where they became confused. If published on your own account, separately retain URL, date and platform-defined view and comment measures.

    Keep
    A first teaser, two comprehension notes and, if published, an original-platform response log.
    Check
    Do viewers ask about the story or only praise the imagery? Views and likes are not income, and one teaser's reach cannot establish that a premise change caused growth.

  5. Decide on episode two from evidence

    Summarise character errors, per-shot attempts, time, spend and viewer questions. Write one new story development for episode two rather than repeating attractive faces. Check the destination platform's actual terms and record credits separately from cash receipts.

    Keep
    An episode-two outline, a continue, revise or stop decision and a separate cost log.
    Check
    If people and blocking remain unstable, viewers cannot explain the relationship or spend exceeds the cap, simplify or stop. Only real agreements and receipts support an earnings claim.

Make one final choice: Revise story and boards if the teaser fails to establish characters. Simplify the scene after two failed motion rounds. Draft another episode only when specific viewer questions and affordable production support it. Use your own records, not this case's interview reach, for the decision.

This is a BitShovel reader exercise, not The Accident makers' published shot-by-shot method. The interview discusses blocking faults and a retained surprise, but not complete rejected clips, settings, costs or receipts; these practice steps are not part of their project record.

Sources and limits

Where the material comes from

  • whata-ha: The Accident EP01
    Original episode from a co-creatorPublished Checked

    Episode one, posted Aug 14, 2026, and its co-creator credit; engagement changes over time.

  • whata-ha: The Accident EP02
    Original episode from a co-creatorPublished Checked

    A continuation from the same creator account.

  • Xiaoyunque interview with both The Accident creators
    Creator interview produced by a supporting vendorPublished Checked

    Earlier attempt, premise, roles, concrete errors and support; audience figures and tool details are attributed claims.