🔍 Read the full analysis: How To Turn A Boring First Draft Into An AI Horror-Comedy on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer’s account says Hollow House, a 97-second AI-generated horror-comedy, took three versions to get right during one evening and night of work. The first cut looked polished but felt boring; tighter pacing and a workflow that gave one video model references for the full cast produced the final version.
Filmmaker Thorsten Meyer says he made Hollow House, a 97-second AI horror-comedy, in three versions over one night, with the final cut emerging after he rejected an attractive but slow first edit. His October 28 account describes a production built from AI-generated characters, images, video, voices and sound, and argues that editing rhythm—not image quality—was the central problem.
Meyer says work began in the early evening of October 27 and finished at about 1:30 a.m. October 28. The final film follows a ghost who joins the staff of an extreme haunted attraction, where workers respond to the supernatural visitor as if he were a workplace and waiver problem. The short is presented as 1 minute and 37 seconds long; Meyer says every character, voice and set was AI-generated.
The production went through three versions. The first used 20 shots and ran 2 minutes 15 seconds; the second used 66 faster cuts and ran 2 minutes 21 seconds. For the final version, Meyer says he created three roughly 30-second segments using one model supplied with references for the cast. He reports spending 2,220 Higgsfield credits across the versions, which he valued at about $95 under his plan, and about 2,100 ElevenLabs characters for music and sound effects.
Before generating video, the team made eight character reference images using Nano Banana 2 through Higgsfield. Meyer says each image cost two credits at 2K resolution and the set was completed in under two minutes. For the first version, those images informed individual start frames, which were animated with Google Veo 3.1 Fast. The account says the visuals and costumes held together, but the edit felt slow and the performances sometimes differed from the script.
AI filmmaking · Case study · October 2026
How To Turn A Boring First Draft Into An AI Horror-Comedy
Thorsten Meyer’s Hollow House took three versions and one late night to find its rhythm. The lesson in his account: polished images can’t rescue a cut that takes too long to land a joke.
01 / Revision arc
Three cuts to make the comedy move
Meyer says the first version looked polished but felt “terribly boring and amateurish.” He changed the pacing and generation setup before settling on the final short.
| Version | Shots / cuts | Runtime | What changed |
|---|---|---|---|
| First cut | 20 shots | 2:15 | Polished look; slow pacing and some performances diverged from the script. |
| Second cut | 66 faster cuts | 2:21 | More frequent edits, but the overall version remained longer. |
| Final cut | 3 segments | 1:37 | One model received references for the full cast across the segments. |
Spot the drag
The 20-shot edit averaged 5.6 seconds per cut, by Meyer’s count. He judged that too slow for the intended comedy.
Try a faster rhythm
The next version used 66 cuts. More edits alone did not settle the final structure or runtime.
Rebuild the workflow
Three roughly 30-second segments used one model supplied with references for the cast.
02 / Why the first cut fell flat
Image quality wasn’t the bottleneck
For this short, Meyer’s account points to pace, continuity and revision as the practical work around AI-generated material.
Comedy needs momentum
Meyer compared the first cut’s 5.6-second average with about 1.8 seconds in the short that inspired the project. These are his counts and comparison.
Reference sheets anchor a cast
Eight character images guided frames and later clips, helping keep hairstyles, clothes and other visual traits consistent.
The team still had to choose
People reviewed the edit, checked continuity, corrected dialogue, changed the format and completed the colour grade.
03 / Production workflow
From character sheets to final cut
The process moved from visual references through generated shots and dialogue checks, with revisions shaped by what the team saw.
Design the cast
Eight 2K character references made with Nano Banana 2 through Higgsfield.
Generate shots
References informed individual start frames, animated with Veo 3.1 Fast in version one.
Check the output
Whisper speech recognition helped compare generated dialogue against the script.
Revise the format
After reviewing pace and performance, the team rebuilt the final version in three segments.
Finish the film
Selection, retakes, timing decisions and colour grading shaped the finished short.
“Terribly boring and amateurish.”Meyer on his first viewing of version one
04 / The project in numbers
A compact production, with reported costs
05 / Premise and limits
A ghost becomes a workplace problem
Hollow House follows a ghost who joins the staff of an extreme haunted attraction. A guest says “cotton candy” as the safe word, but the frightening encounter continues. Staff respond to the supernatural visitor like an operations and waiver problem; a worker in a priest costume is recruited for an exorcism, which goes badly, before the ghost later takes a position at the entrance. Meyer says the story’s structure was inspired by an AI horror-comedy shared on X; he says the script, characters and jokes were his team’s own.
- First-person report: The timing, cut counts, costs and quality judgments are Meyer’s own account.
- No controlled comparison: It does not establish that one model or method will improve other films.
- Unclear causal change: The report does not isolate which change mattered most between versions two and three.
- Generation quirks: Reported errors included spoken letters for “HEY!” and actions rendered differently than requested.
06 / Key questions
What the account tells us
What is Hollow House?
A 97-second horror-comedy short about a ghost joining an extreme haunted attraction’s staff. Meyer says its characters, voices and sets were AI-generated.
Why remake the first version?
Meyer felt the first cut looked good but moved too slowly for the comedy. He focused on rhythm and workflow rather than adding visual detail.
How many versions were made?
Three: a 20-shot, 2:15 cut; a 66-cut, 2:21 version; and the final 1:37 cut made as three segments with cast references.
How much did production cost?
Meyer estimates 2,220 Higgsfield credits, worth about $95 under his plan, plus roughly 2,100 ElevenLabs characters for music and effects.
Why the First Cut Fell Flat
Meyer’s account points to a practical distinction in AI-assisted filmmaking: generating polished images does not by itself create an effective film. In his assessment, the first cut’s main weakness was its pace. He counted an average of 5.6 seconds per cut in his 20-shot version, compared with about 1.8 seconds in the short that inspired the project. Those figures are his counts, not an independent analysis, but they explain why he decided to recut rather than simply improve the images.
The report also documents how production choices affected the result. Meyer says one model with references for the whole cast was used for the final version, while character sheets helped maintain appearance across generated material. Human decisions remained central: he judged the first edit, changed the format and completed the colour grade. The case is relevant to creators using generative tools because it shows the work still involved selection, continuity checks, timing and revision, not just prompts.
There are limits to what the example establishes. It is one creator’s account of a short film, rather than a controlled comparison of video models or a measure of how other teams will fare. The reported costs also reflect Meyer’s subscription and workflow. The useful takeaway is specific: for this short, he says a change in editing and generation approach mattered more than adding visual detail.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
De'Longhi Magnifica Evo Fully Automatic Espresso Machine with Manual Milk Frother
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Final Cut
The premise uses a familiar haunted-attraction rule as its comic setup: visitors can say a safe word to stop the scares. Meyer’s story asks what happens when a figure in the house does not follow that rule. A guest says “cotton candy,” but the frightening encounter continues; the staff then treat the ghost as an operational problem. A worker in a priest costume is recruited for an exorcism, which goes badly, before the ghost later takes a position at the entrance.
Meyer says the film’s structure was inspired by an AI horror-comedy short shared on X, while the script, characters and jokes were his team’s own. The account does not name the original short. Its production description emphasizes character reference sheets as a continuity tool: the same images were used to guide later frames and clips, reducing the chance that hairstyles, clothing or other visual traits would change between shots.
The first version exposed technical limits as well as editorial ones. Meyer reports that Veo rendered a shouted “HEY!” as the letters spoken individually, and that a planned full head turn became a whole-body turn. A requested backward bend came out as a hunched crawl. He says a lifting shot was initially blocked, then passed after the action was reworded as floating, though the character barely left the ground. The team also used Whisper speech recognition to compare generated dialogue with the script and identify errors for retakes.
“Terribly boring and amateurish.”
— Thorsten Meyer, describing his first viewing of version one
What the Production Account Leaves Open
The account is a first-person production report, and its comparisons—including the cut counts, timing averages and quality judgments—are Meyer’s own measurements and assessment. No independent review, cost breakdown beyond the stated credit estimate, or side-by-side evaluation of the three versions is included in the supplied material. It is also not clear which specific change between the second and third versions contributed most to the final cut’s effectiveness.
The source describes three versions and gives the final version’s duration, but does not provide a complete account of every edit or the exact generation settings used in each stage. It identifies an inspiration shared on X without naming it. The report therefore supports a description of Meyer’s workflow and conclusions, but not a broader claim that one model or method will reliably improve other AI films.
The Short and Its Workflow
Hollow House is presented with Meyer’s behind-the-scenes account and a linked video of the short. The account includes character sheets, start frames, examples of failed generations and a contact sheet of the final cut. No further production milestone or follow-up release is specified in the supplied material, so it is unclear whether Meyer plans to publish more process notes or another version.
For readers assessing the project, the next useful evidence would be the film itself alongside the production materials Meyer says are available. That would allow viewers to judge the pacing and performance directly, while keeping the reported cost and model limitations in the context of this particular production.
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 Meyer remake the first version?
Meyer says the first version looked good but felt boring. It had 20 shots averaging 5.6 seconds per cut, which he judged too slow for the comedy and compared with a faster-paced short that inspired the structure.
How many versions were made?
Meyer reports three versions: a 20-shot cut lasting 2:15, a 66-cut version lasting 2:21, and the final 1:37 cut made as three segments using one model with references for the cast.
How much did the AI production cost?
Meyer estimates 2,220 Higgsfield credits across all three versions, worth about $95 at his plan price, plus roughly 2,100 ElevenLabs characters for music and effects. These are his reported figures for this project, not a general production-cost estimate.
What remains unknown about the project?
The source does not provide an independent comparison of the cuts, a detailed cost audit or a breakdown showing which change mattered most. It also does not identify the short shared on X that inspired the structure.
Source: ThorstenMeyerAI.com
