Hollow House: One Night To Make An AI Horror-Comedy That Works
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Hollow House: One Night To Make An AI Horror-Comedy That Works on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

  • 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, was made between the evening of October 27 and about 1:30 a.m. on October 28, according to its creator. The first two versions fell short; the final version used three longer segments generated from references for the whole cast, while editing and color grading remained human decisions.

Hollow House, a 97-second horror-comedy about a ghost joining the crew of a haunted attraction, was made in three versions over one night, according to creator Thorsten Meyer. His October 28 account says the first two cuts did not work for him; the final version changed the production approach, using three roughly 30-second segments generated by one model with references for the full cast.

Meyer says production began in the early evening of October 27 and wrapped at about 1:30 a.m. the next day. The short’s premise puts a supernatural intruder among scare actors and guests at an extreme haunted attraction. When a guest says the attraction’s safe word, the ghost keeps attacking; the staff respond as though the problem might be a workplace or waiver issue. The characters, voices and sets are all described as AI-generated.

The creator reports spending 2,220 Higgsfield credits across the three versions, which he estimates at about $95 at his plan’s price, and roughly 2,100 ElevenLabs characters on music and sound effects. Before generating video, the team made reference images for eight characters using Google’s Nano Banana 2 image model through Higgsfield. Meyer says each image cost two credits at 2K resolution and that all eight were made in under two minutes.

The first cut used 20 shots and ran 2 minutes 15 seconds. Meyer says it looked good but felt boring: its cuts averaged 5.6 seconds, while he counted about 74 cuts in the short that inspired the project, an average of roughly 1.8 seconds. The second version had 66 fast cuts and ran 2 minutes 21 seconds. The final cut, made from three segments using one model with the entire cast as references, ran 1 minute 37 seconds. These figures and evaluations come from Meyer’s account, rather than an independent production record.

At a glance
reportWhen: Published October 28, 2026; production…
The developmentThorsten Meyer has published a behind-the-scenes account of making Hollow House, describing three AI-generated versions completed over one night and the changes that shaped the final short.
Hollow House: One Night to Make an AI Horror-Comedy That Works
Behind the scenes · October 2026

Hollow House: One Night to Make an AI Horror-Comedy That Works

A 97-second haunted-attraction comedy took three versions, a change in generation workflow, and human judgment about what made the jokes land. Here is what creator Thorsten Meyer says happened overnight.

97 secFinal runtime
3Versions made
~7.5 hrReported overnight window
$95*Estimated tool cost
The development

Three cuts, one overnight shoot

Meyer says production began in the early evening of October 27 and wrapped around 1:30 a.m. on October 28. The first two cuts missed the mark; the final version changed how scenes and cast references were generated.

Creator-reported production details

01

Good-looking, slow

The first cut used 20 shots. Meyer says its consistent faces, costumes, and color palette still felt boring.

2:15 runtime · 5.6 sec average shot
02

Faster, still not right

The second pass pushed the pace with many more cuts, but Meyer says this version also did not work for him.

2:21 runtime · 66 cuts
03

Rebuilt workflow

The final cut was assembled from three longer segments generated with one model and references for the full cast.

1:37 runtime · 3 segments
From premise to finished short

How the workflow changed

The premise puts a ghost among the staff and guests of an extreme haunted attraction. When a guest says the safe word, the ghost keeps attacking—and staff treat the crisis like a workplace or waiver problem.

STEP 01 Build the cast

Eight character reference images were made at 2K resolution with Nano Banana 2 through Higgsfield.

STEP 02 Generate scenes

The first version used shot descriptions, start frames, and Veo 3.1 fast for animation, speech, and sound.

STEP 03 Diagnose the miss

Meyer says the initial images looked strong, while the slower cut weakened the comic timing.

STEP 04 Rebuild and finish

Cast references and longer segments shaped the final generation; editing and color grading remained human choices.

What mattered most

Attractive images were only a start

The account presents a creator actively reviewing and revising AI output. Meyer says the decisive work included recognizing the pacing problem, changing the setup, checking dialogue, and grading the final image.

Pacing

Comedy depends on the edit

Meyer compared the first cut’s 5.6-second average shot length with roughly 1.8 seconds in the short that inspired the project. More cuts alone did not guarantee a successful result.

Human direction

Generation needed judgment

He selected the approach, diagnosed failed attempts, checked dialogue against the script with Whisper, and made the final color grade.

Production workarounds

Prompts met tool limits

He reports that “HEY!” was rendered as spoken letters, a head turn missed its intent, and a requested lift was blocked. Rewording the prompt passed but looked less dramatic.

Reported resources

Specific to this project

Meyer reports 2,220 Higgsfield credits across three versions, about $95 at his plan price, and roughly 2,100 ElevenLabs characters for music and sound effects.

Evidence and limits

What this account does—and does not—show

The production figures and evaluations come from Meyer’s own account, rather than an independent production record.

A single creator’s workflow

The supplied material does not include an independent credit invoice, a complete cost breakdown, or a controlled comparison of the three methods. The $95 estimate reflects Meyer’s plan price. No audience testing, viewership data, or measured response establishes that the final version worked better for viewers. The account offers a specific example, not a general price, schedule, or proof that AI can make a film without substantial direction and selection.

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

Thorsten Meyer · Creator

“Comedy lives on timing, and we had none.”

Thorsten Meyer · Creator

“A locked picture per character is the closest thing AI film has to a casting call and a costume department in one.”

Thorsten Meyer · Creator
Where to watch

See the finished short

Meyer’s account links the 1:37 film under its release title. It also describes behind-the-scenes material including character sheets, starting frames, failed attempts, and a contact sheet of the final cut.

Hollow House | A Halloween Horror-Comedy Short

View the short and consult the creator’s account for the production images and fuller explanation.

Visit the creator’s account

Why the First Cut Fell Flat

Meyer’s account points to a practical challenge for AI video: generating attractive images did not by itself produce a working comedy. He says the first cut had consistent faces, costumes and a haunted-house color palette, but its slower pacing undermined the jokes. The change was not simply a matter of generating more footage; it involved reconsidering how the cast references and scenes were organized, then shaping the material into a faster final cut.

The process also shows where human decisions remained central. Meyer identifies calling the first version boring, changing the generation setup and applying the final color grade as decisive choices. AI tools produced images, video, dialogue and sound elements, but the account describes a creator diagnosing problems and revising the approach. The short is therefore evidence of one creator’s workflow, not proof that AI tools can reliably make a successful film without substantial direction and selection.

For readers following generative video, the reported costs and production time offer a specific example, but they should not be treated as a general price or time estimate. The source does not provide independent verification, a full tool-by-tool invoice or a controlled comparison of the three methods. Its clearest takeaway is narrower: in this project, the creator says pacing and workflow mattered more than the visual quality of the initial shots.

Products Worth Considering

Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok

Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok

Jackery HomePower 3000 Portable Power Station with Two 200W Solar Panels

Jackery HomePower 3000 Portable Power Station with Two 200W Solar Panels

As an affiliate, we earn on qualifying purchases.

Three Cuts, One Overnight Shoot

The project began with a haunted-house premise that turns a safe word and workplace procedures into comedy. Meyer says the structure was inspired by an AI horror-comedy short shared on X, while the script, characters and jokes were his own. That account distinguishes the initial creative prompt from the material he says he developed for Hollow House.

For the first version, the team created 20 shot descriptions and generated a start image for each with Nano Banana 2, using the character sheets as references. They then animated those images with Google’s Veo 3.1 fast mode, which Meyer says generated speech and sound with the video. The resulting cut’s longer average shot length led him to compare its pacing with the reference short and rebuild the edit.

Meyer also describes technical limitations and workarounds. Veo rendered the capitalized “HEY!” as the letters being spoken aloud, so he says shouting needed to be written as a verbal direction rather than typed in capitals. A requested head turn did not work as intended, and one attempt to show a guest being lifted was blocked; a reworded floating prompt passed but produced a less dramatic result. Meyer says he used Whisper speech recognition to check generated dialogue against the script and identify errors for retakes.

““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 Account Does Not Establish

The production details are presented in Meyer’s own account, and the source material does not include independent confirmation of the credits used, the estimated dollar cost or the time spent. The $95 figure is tied to his plan price; it is not a universal cost for making a short with these tools. The account also does not provide a detailed breakdown of how credits were allocated among shots, retries and versions.

It is not clear from the supplied material how the second version’s 66 cuts were distributed across its 2 minutes 21 seconds, or what specific editorial changes, beyond the final version’s three-segment generation method, made the final cut work better. Meyer’s statement that only the third version worked is his creative judgment, not a measured audience response. No viewership, audience-testing or performance data is provided.

The source mentions an inspirational short shared on X but does not identify it in the supplied text. It also does not provide enough material to independently assess the relationship between that short and Hollow House. The production account describes the creator’s workflow and conclusions; broader claims about AI filmmaking would require evidence beyond this single project.

Where to Watch the Finished Short

The finished 1:37 short is linked in Meyer’s account under the title “Hollow House | A Halloween Horror-Comedy Short.” The source also says the article includes behind-the-scenes images, such as character sheets, starting frames, failed attempts and a contact sheet of the final cut. Those materials may help readers compare the described workflow with the finished work.

No follow-up production, release schedule or additional project was announced in the supplied material. The next available step for readers is to view the short and consult the creator’s account for its production images and fuller explanation. Any claims about audience reception, later revisions or new versions remain unreported.

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.

How long did it take to make?

Meyer says production ran from the early evening of October 27, 2026, until about 1:30 a.m. on October 28. That is the creator’s reported schedule; no independent production log is included.

Why were there three versions?

Meyer says the first cut looked good but felt too slow, with 20 shots averaging 5.6 seconds per cut. The second used 66 fast cuts, and the final version was generated as three segments using one model with references for the full cast.

How much did the project cost?

The creator reports using 2,220 Higgsfield credits, which he estimates at about $95 at his plan price, plus roughly 2,100 ElevenLabs characters for music and sound effects. The estimate is specific to his account and does not establish a general production cost.

Is the reported workflow independently verified?

No independent verification is provided in the source material. The timing, cost, tool use and judgments about which version worked are attributed to Meyer’s behind-the-scenes account.

Source: ThorstenMeyerAI.com

You May Also Like

The Top 10 AI Tools Every Student Organization Should Use

Discover the top 10 AI-powered tools that student organizations should use to enhance organization, productivity, and collaboration in 2024.

Ollaya – Ollama For Open-source, Jev-style Decision Models

Search interest in “Ollaya – Ollama for open-source, Jev-style decision models” is rising, but the reason for the spike is unconfirmed.

Why Reporting AI Model Misalignment Matters And How We Do It

OpenAI has released a public framework detailing how it will identify, evaluate, and disclose instances of AI model misalignment, aiming for greater transparency.

How To Weigh AI Coding’s Costs Against Its Benefits

AI coding budgets range widely, while productivity results vary. Here are the reported gains, costs and metrics teams can use to judge value.