Why Our First Take On Hollow House Was Boring—and How We Changed It
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Why Our First Take On Hollow House Was Boring—and How We Changed It on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Hollow House, a 97-second AI-generated horror-comedy, took three versions to reach its final cut. Its creator says the key problem was pacing, not image quality: the first edit averaged 5.6 seconds per shot, while the short that inspired it averaged about 1.8 seconds.

Thorsten Meyer says his team rebuilt the AI-generated horror-comedy Hollow House three times in one night after deciding the first version looked polished but felt boring. The final film runs 97 seconds, and the production account argues that its main improvement came from changing the editing rhythm and using one video model with the full cast as references—not from fixing the images.

Meyer’s account, published October 28, describes a short set in an extreme haunted attraction where a real ghost joins the scare crew and staff treat the situation like a workplace problem. In the story, a guest says the safe word, “cotton candy,” but the haunting continues; shift lead Ray asks whether the newcomer signed the waiver. The film’s characters, voices and sets were all AI-generated.

The first cut used 20 single shots and ran 2 minutes 15 seconds, with each edit averaging 5.6 seconds, according to Meyer’s count. He compared that with roughly 74 cuts across 2 minutes 14 seconds in the short that inspired the project—about 1.8 seconds per cut. Meyer said the first version’s images and character consistency were strong, but its long shots, mostly static camera and lack of music left the comedy without enough pace.

The production then moved through two further versions. The second used 66 fast cuts and ran 2 minutes 21 seconds. The final version was made as three roughly 30-second segments using one model supplied with references for the full cast; it runs 1:37. Meyer reports that all three versions together used 2,220 Higgsfield credits, which he estimates at about $95 at his plan’s price, plus about 2,100 ElevenLabs characters for music and sound effects.

At a glance
reportWhen: Published October 28, 2026; the film wa…
The developmentThorsten Meyer published a behind-the-scenes account of how the AI-generated short Hollow House was revised three times overnight after its first cut felt boring.
Why Our First Take On Hollow House Was Boring—and How We Changed It

Behind the cut · AI horror-comedy · October 28, 2026

Why Our First Take On Hollow House Was Boring—and How We Changed It

Three versions in one night. The images already looked good; the edit needed a different pulse. Here is how a haunted-attraction comedy found its rhythm.

1:37Final runtime
3Versions overnight
1.8sInspired short · avg. cut
8Character reference sheets

01 / The diagnosis

Good-looking shots can still feel flat

Hollow House follows a real ghost who joins an extreme haunted attraction’s scare crew—and is treated like a workplace issue.

The setup

A haunting, on shift

A guest says the safe word, “cotton candy,” but the haunting continues. Shift lead Ray asks whether the newcomer signed the waiver.

The first cut

Polished, but slow

The 2:15 version used 20 shots, averaging 5.6 seconds each. Meyer said the mostly static camera and absent music drained pace from the comedy.

The lesson

Editing makes the beat

For horror-comedy, timing helps build both surprise and punchlines. Attractive images alone did not create a coherent, funny short.

02 / Pace, visualized

5.6 seconds versus 1.8

Meyer compared the first edit with the short that inspired the project. The comparison and counts come from his account.

Hollow House · first cut5.6 sec
Inspiring short · reported1.8 sec

Reported comparison: 20 shots across 2:15 versus about 74 cuts across 2:14.

More cuts can sharpen a joke

The first version’s character consistency and images were strong, Meyer said. But long shots and a largely static camera left the scares and comic pauses without enough momentum.

His key revision was an editorial one: change the rhythm so each beat has room to land, then move on.

03 / Three versions

From first pass to final cut

The team rebuilt the short three times overnight. The account gives limited detail about the exact changes to version two.

VersionRuntimeEditing detailProduction note
First2:1520 shots · 5.6 sec averagePolished images, but too slow and static
Second2:2166 fast cutsA further pass toward a quicker pace
Final1:37Three segments of about 30 secOne video model with references for the full cast

04 / The workflow

References first. Then motion, sound, and judgment.

AI generated the characters, voices, and sets. People chose the story, evaluated the cuts, and decided what to fix.

1

Build the cast

Eight 2K character sheets made with Nano Banana 2 through Higgsfield; two credits each, in less than two minutes total.

2

Generate frames

Twenty still start frames used the character references, then Veo 3.1 fast animated them with dialogue and sound.

3

Check the words

Whisper helped compare generated dialogue with the script and catch a spelled-out “HEY!” and garbled syllables.

4

Rebuild the rhythm

Use cast-wide references with one model, revise the pacing, and apply a final colour grade by human judgment.

05 / Reported usage

A partial cost picture

These are Meyer’s estimates for listed AI-generation usage, not a complete production budget.

2,220 Higgsfield credits · all versions

Estimated by Meyer at about $95 using his plan’s price.

~2,100 ElevenLabs characters

Used for music and sound effects, according to the account.

Labour, subscriptions, and other expenses are not itemized. Treat the figures as a reported estimate for the listed AI-generation costs.

06 / Key questions

What the account tells us—and leaves open

What is Hollow House?

A 97-second horror-comedy about a ghost joining the staff of an extreme haunted attraction. Meyer says its characters, voices, and sets were AI-generated.

Why change the first version?

Meyer called it boring and amateurish. He attributed that response mainly to the slow editing rhythm, not image quality.

What was made by people?

Meyer says the script, characters, and jokes were his team’s own. The structure drew inspiration from an AI horror-comedy short shared on X; the supplied account does not identify it.

What remains unverified?

The counts, timing comparison, credit usage, and cost estimate come from Meyer’s first-person account and are not independently audited in the supplied material.

What happened during generation?

A shouted “HEY!” became the letters “H-E-Y”; a requested head turn became a whole-body turn. A revised “floating” prompt passed after an attempted levitation effect was blocked, though the character barely rose.

What comes next?

The short is available as a 1:37 film. The account does not announce a sequel, release schedule, audience results, or a broader production milestone.

The human decisions shaped the final cut.

Meyer points to three decisive moves: recognizing that the first edit was boring, supplying one model with references for the full cast, and applying the final colour grade. The workflow is one creator’s account; whether shorter edits and cast-wide references help other projects will take more examples to establish.

Why the First Cut Fell Flat

The account highlights a practical limit of AI-assisted filmmaking: generating attractive shots does not automatically produce a coherent or funny film. Meyer’s own comparison points to editing pace as the central issue in the first cut. The difference between an average 5.6-second shot and one around 1.8 seconds is substantial in a short built around horror-comedy beats, where pauses and cuts help shape both surprise and punchlines.

It also shows how much creative judgment remained with the human maker. Meyer identifies three consequential decisions: calling the first version boring, changing how the cast was supplied to the video model, and applying the final colour grade. The models generated the material, but the creator decided what was not working and how to revise it. That distinction matters to readers evaluating claims that AI tools can make a film with little human input.

The cost figures offer a limited view of the production burden, not a complete budget. Meyer gives the platform-credit estimate and audio-generation usage, but does not provide a full accounting of labour, subscriptions or other expenses. The reported total is therefore best read as his estimate for the listed AI-generation costs.

As an affiliate, we earn on qualifying purchases.

From Character Sheets to Final Cut

Before generating video, Meyer created one reference image for each of eight characters using Google’s Nano Banana 2 image model through the Higgsfield platform. He says the sheets were made at 2K resolution, cost two credits each and took less than two minutes in total. Those images served as visual references for later frames and clips, intended to limit changes to details such as hair, clothing and appearance between shots.

For the first version, the team generated 20 still start frames using those character references, then animated them with Google’s Veo 3.1 fast mode. Meyer says Veo generated dialogue and sound alongside the video. The production account also describes difficulties with particular instructions: a shouted “HEY!” was spoken as the letters “H-E-Y,” a requested full head turn became a turn of the whole body, and an attempted levitation effect was initially blocked. A revised description using “floating” passed, but the character barely rose.

Meyer says the script, characters and jokes were his team’s own, while the structure drew inspiration from an AI horror-comedy short shared on X. The account does not identify the original short in the supplied material. It also describes using OpenAI’s Whisper speech-recognition model to check generated dialogue against the script, catching the spelled-out shout and garbled syllables that prompted a retake.

“The first version looked great, and when I watched it I called it what it was: terribly boring and amateurish.”

— Thorsten Meyer, creator of Hollow House

What the Production Account Leaves Open

The source is a first-person production account from Meyer, not an independently verified production log. Its shot counts, timing comparisons, credit usage and cost estimate are attributed to him; the article does not include outside measurements or itemized billing records. The claimed comparison with the inspiring short is also Meyer’s count, and the source does not name that film in the material provided.

Details about the second version’s production method, the precise changes made between all three cuts and the division of work among any collaborators are limited. The account says the final cut used one model with references for the full cast, but does not specify the model version or provide a technical evaluation of how much that approach improved consistency. It remains unclear how the short performs with viewers or whether its creators plan more films using the same workflow.

The Published Film and Its Workflow

Hollow House is available as a 1:37 short, and Meyer’s October 28 account presents its character sheets, start frames, failed attempts and final-cut contact sheet as production evidence. The next step for readers is to compare the finished film with that documented workflow; the source does not announce a sequel, a release schedule for further work or a new production milestone.

Any broader conclusions about AI filmmaking will depend on more than one creator’s account. Further examples could show whether using cast-wide references and shorter edits reliably helps other projects, or whether the result depends on this particular film’s script, tools and human decisions.

Key Questions

What is Hollow House?

Hollow House is a 97-second horror-comedy short about a ghost joining the staff of an extreme haunted attraction. Meyer says its characters, voices and sets were AI-generated.

Why did the creators change the first version?

Meyer said the first cut looked good but felt boring and amateurish. He attributed that response mainly to its slower editing rhythm: 20 shots averaging 5.6 seconds each.

How many versions were made?

The team made three versions overnight. They ran 2:15, 2:21 and 1:37, respectively, according to Meyer’s account.

How much did the AI generation cost?

Meyer estimates that the three versions used 2,220 Higgsfield credits, worth about $95 at his plan price, plus about 2,100 ElevenLabs characters for music and sound effects. The source does not give a full production budget.

What remains unverified?

The production numbers and cost are reported by Meyer and are not independently audited in the supplied source. Audience response, the full production budget and plans for additional films are not stated.

Source: ThorstenMeyerAI.com

You May Also Like

What Theme Parks Get Right About Wayfinding and Anticipation

I’m excited to reveal how theme parks master wayfinding and build anticipation, transforming navigation into an immersive adventure you’ll want to experience firsthand.

From Old Look To New: Gewerkton’s Two-Day AI Agents Project

Gewerkton says AI coding agents implemented its HORIZON redesign across apps and a 40-language website on October 2–3, with human review and fixes.

The Future Of Customer Service: Avatarin’s AI-Driven 24/7 Retail Agent Explained

OpenAI reports that Avatarin has developed a 24/7 retail customer service agent powered by GPT-Realtime, marking a step toward continuous AI-driven retail support.

Design Thinking in Energy & Utilities: Empowering Consumers in Power Use

Great innovations in energy and utilities begin with understanding your needs—discover how design thinking empowers consumers and transforms power use.