🔍 Read the full analysis: Making Hollow House: Our One-Night Journey Into AI Horror-Comedy on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Thorsten Meyer says Hollow House, a 97-second AI-generated horror-comedy, was made overnight on October 27-28 after three versions. The production account describes how reference images, a change in editing approach and human decisions shaped the final cut, while reporting roughly $95 in video-generation credits.
Thorsten Meyer says he made Hollow House, a 97-second AI-generated horror-comedy, during an overnight production that ran from early evening on October 27 to about 1:30 a.m. the next day. His October 28 account says the team generated three versions before settling on a faster-cut version, describing a process that shows how editing and human judgment shaped a short made with AI-generated characters, voices and sets.
The film’s premise is a ghost joining the crew of an extreme haunted attraction, where staff treat the supernatural intruder as a workplace problem. A guest uses the safe word “cotton candy,” but the scare does not stop; the shift lead, Ray, asks whether the apparent new hire signed a waiver. The account says the script, characters and jokes were Meyer’s own, while the structure drew inspiration from an AI horror-comedy short shared on X.
Meyer reports that the first cut used 20 shots and ran 2 minutes 15 seconds. Although he thought the images looked good, he judged the result boring. He compared its average of 5.6 seconds per cut with about 1.8 seconds in the short that inspired the project. The second version used 66 faster cuts and ran 2 minutes 21 seconds. The final cut used three roughly 30-second segments generated with one model given references for the full cast; it runs 1 minute 37 seconds.
Before making clips, Meyer created eight character reference images with Nano Banana 2 through the Higgsfield platform. He says the images helped maintain character and costume consistency across generated shots. For the first version, he paired still frames with Google’s Veo 3.1 fast mode, which generated dialogue and sound with the video. He also says the production used 2,220 Higgsfield credits across all three versions—about $95 at his plan price—and roughly 2,100 ElevenLabs characters for music and sound effects.
Behind the scenes · One-night production
Making Hollow House: Our One-Night Journey Into AI Horror-Comedy
A 97-second ghost story took three versions, a faster edit and a night of human decisions. Creator Thorsten Meyer recounts how the short came together—and what changed between first cut and final cut.
“The pictures were never the problem. The rhythm was.”Thorsten Meyer · Production account
01 / The edit changed the story
Three versions, one sharper rhythm
The first cut looked polished to Meyer, but its slow pacing weakened the comedy. He reworked the approach twice, ending with a compact final cut assembled from three generated segments.
Good images, slow momentum
Twenty shots and a 2:15 runtime. Meyer says the average cut lasted 5.6 seconds and the result felt boring.
More cuts, still evolving
Sixty-six cuts pushed the pace. This version ran 2:21 before another structural rethink.
Three segments, cast references
Roughly three 30-second segments were generated with one model given references for the full cast.
02 / A ghost on the night shift
A workplace problem from beyond
In Hollow House, a ghost joins the crew of an extreme haunted attraction. Staff respond to the supernatural intrusion like an awkward shift issue: a guest says “cotton candy,” but the scare carries on, and shift lead Ray asks whether the apparent new hire signed a waiver.
Build the cast
Eight character portraits establish the people and costumes.
Generate the scenes
Still frames become clips, with dialogue and sound generated along the way.
Review and revise
Meyer checks the pacing, dialogue and continuity, then changes the approach.
Land the final cut
Three referenced segments form the 97-second short.
03 / Where the work happened
AI made the material; judgment shaped the film
Meyer’s account describes a hybrid process. Generative tools produced images, voices and sets, while the creator chose references, rejected a cut, checked lines and finished the grade.
Eight reference portraits
Created at 2K resolution with Nano Banana 2 through Higgsfield; Meyer says each cost two credits and helped keep hair and costumes consistent.
Frames into video
Still images paired with Google Veo 3.1 fast mode, which generated video alongside dialogue and sound.
Listen, transcribe, retake
Whisper transcripts were checked against the script. A spelled-out shout and garbled extra syllables prompted review and a retake.
Human choices remain
Meyer selected the final structure and completed the colour grade after reviewing the generated material.
“Terribly boring and amateurish.”
Meyer on version one“Did you even sign the waiver?”
Ray, as quoted in the account“The pictures were never the problem. The rhythm was.”
Meyer on the central lesson04 / What the account can—and cannot—show
A specific production story, with open questions
The behind-the-scenes article is a first-person account by the film’s creator. Its details document one project and should be read within that scope.
Revisions and decisions
Three versions, changing cut counts, reference-image use, dialogue checks and a final colour grade, as Meyer describes them.
Independent evidence
No independent cost audit, audience-response data, full shot-by-shot comparison or complete final model setup is provided.
Project-specific limits
Meyer recounts trouble with a spelled-out “HEY!”, a head turn, a bridge pose and a floating guest. These are observations from this production.
Published with the short
The account appeared October 28, 2026, alongside “Hollow House | A Halloween Horror-Comedy Short” and production materials.
What is Hollow House?
A 97-second horror-comedy about a ghost joining an extreme haunted-attraction crew. Meyer says the characters, voices and set were AI-generated.
How long did it take?
Meyer says production ran from early evening October 27 to about 1:30 a.m. on October 28, 2026.
Why make three versions?
He rejected the first as boring despite liking its visuals, tried a faster 66-cut version, then changed the generation approach for the final cut.
What is not verified?
The account includes no independent cost check, audience measurements or complete technical breakdown of the final generation setup.
Why the First Cut Fell Flat
Meyer’s account points to a practical challenge for AI-assisted filmmaking: generating attractive images does not automatically produce a coherent or funny film. He says the first version had consistent faces, costumes and lighting, but its slow pacing and static camera weakened the comedy. The eventual change was not simply a better image prompt; it was a rethink of how quickly the story moved and how the shots were assembled.
The workflow also shows where human review remained central. Meyer says he watched and rejected the first cut, chose to provide the cast as references for the later generation, checked dialogue against a speech-recognition transcript and completed a final colour grade. These are the creator’s own descriptions, not an independent assessment of the film or its production costs. Still, the account offers readers a specific example of the choices and revisions behind a very short AI-made video.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
Jackery Explorer 300 Portable Power Station with 40W Solar Panel
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Final Cut
The production began with eight reference portraits, one for each character. Meyer says the images were made at 2K resolution and cost two credits each. He treated them as a way to keep details such as hair colour and costumes stable when generating later frames and clips. That preparation came before the video work, which started with 20 planned shots and individual start frames.
The first version exposed technical limits as well as editorial ones. According to Meyer, Veo pronounced the capitalized “HEY!” letter by letter, failed to produce a full head turn without turning the character’s body, and did not create the intended backward bridge pose. An attempted shot of a guest being lifted was initially blocked; a rewritten “floating” description passed, but the feet barely left the ground. Meyer says he used a still image with the head already reversed and added a sound effect to suggest the turn.
For dialogue checks, the creator says he ran clips through Whisper and compared the transcript with the script. That caught the spelled-out shout and prompted a retake after a line returned with extra, garbled syllables. The account describes these as production observations from this project, not a broad test of how the tools perform in other circumstances.
““Terribly boring and amateurish.””
— Thorsten Meyer, describing his reaction to version one
What the Production Account Leaves Open
The article is a first-person production account by the film’s creator. It does not provide an independent breakdown of the tools’ performance, verify the stated credit-to-dollar calculation, or identify the exact model setup used to generate the final three segments beyond saying one model had the whole cast as references. The described costs are Meyer’s estimates at his plan price, not a general price for making comparable videos.
The source also does not provide a complete shot-by-shot comparison of all three versions or explain how much generation time and discarded material contributed to the total. It says the final cut worked better, but gives no audience-response data or independent evaluation of its comedy. The short’s inspiration is described as a video shared on X, without a creator or link in the account.
The Published Short and Its Process
Meyer’s account was published on October 28, 2026, alongside the short, titled “Hollow House | A Halloween Horror-Comedy Short,” and behind-the-scenes material including character sheets, starting frames, failed attempts and a contact sheet of the final cut. The account does not announce a sequel, further production or a planned release schedule.
For now, the next step for readers is to compare the finished 1:37 film with the process Meyer describes. Any broader conclusions about AI video production would require additional examples and independently documented costs; the figures and lessons here apply to this one-night project as reported by its maker.
Key Questions
What is Hollow House?
Hollow House is a 97-second horror-comedy about a ghost joining an extreme haunted-attraction crew. Meyer says the characters, voices and set were AI-generated.
How long did it take to make?
Meyer says production began in the early evening on October 27, 2026, and continued until about 1:30 a.m. the following day. His account was published October 28.
Why were there three versions?
Meyer says he rejected the first, 20-shot version as boring, despite liking its visuals. A faster 66-cut version followed, before the final approach used three roughly 30-second segments with the cast supplied as references.
How much did the AI tools cost?
Meyer reports using 2,220 Higgsfield credits, which he valued at about $95 under his plan, plus about 2,100 ElevenLabs characters for music and sound effects. These are his project-specific estimates.
What remains unverified?
The published account does not include an independent cost audit, audience-response measurements or a full technical breakdown of the final generation setup. The production details and conclusions are Meyer’s account of his own film.
Source: ThorstenMeyerAI.com
