🔍 Read the full analysis: The Making Of An AI Horror-Comedy: Hollow House In One Night 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’s account of making Hollow House describes three versions produced overnight, with the final 97-second AI horror-comedy built from three segments generated using one model and the full cast as references. Meyer says the main challenge was not image quality but pacing; the project used 2,220 Higgsfield credits, which he valued at about $95 under his plan.
Thorsten Meyer says he made Hollow House, a 97-second AI-generated horror-comedy, in three attempts over one night, replacing an attractive but slow first cut with a faster version made from three 30-second segments. In a behind-the-scenes account published October 28, Meyer says the short used 2,220 Higgsfield credits—about $95 at his plan price—and that the key creative decisions, including rejecting the first edit and changing the production method, were made by a person.
The film’s premise puts a real ghost inside an extreme haunted attraction, where staff treat the supernatural threat as a workplace problem. When a guest says the safe word, “cotton candy,” the scaring does not stop. The workers then try to handle the apparent ghost through ordinary shift procedures, including asking whether he signed a waiver. Meyer says the short’s characters, voices and sets are AI-generated.
Meyer describes the first attempt as 20 shots running 2 minutes 15 seconds, generated from still images and animated with Google’s Veo 3.1 fast mode. He says it looked good but felt boring. By his count, it averaged 5.6 seconds per cut, while the short that inspired the project averaged about 1.8 seconds. The second version used 66 fast cuts and ran 2 minutes 21 seconds; the third, which he says worked, was cut to 1:37.
For the final attempt, Meyer says he generated three roughly 30-second segments with one model supplied with references for the full cast. The workflow began with eight character reference images made using Google’s Nano Banana 2 through Higgsfield. Meyer says those images helped keep appearances and costumes consistent. He also reports using about 2,100 ElevenLabs characters for music and sound effects. The figures and production account come from Meyer; the article does not provide independent cost verification.
The Making of an AI Horror-Comedy: Hollow House in One Night
Thorsten Meyer says three versions of the short emerged overnight. The finished 97-second cut turns a real ghost into a workplace problem—and puts human editing decisions at the center of an AI-assisted workflow.
Three cuts, one faster landing
Meyer’s account describes a shift from a shot-by-shot build to three larger segments with the full cast supplied as references. His central measure was pace: the first version averaged 5.6 seconds per cut, against about 1.8 seconds in the short that inspired the project.
Attractive, but slow
Still-image start frames animated with Veo 3.1 fast mode. Meyer says the faces, costumes and lighting held together, but the comedy dragged.
More cuts, still long
A faster cutting approach increased the cut count, but the version remained longer than the final short.
Built for momentum
Three roughly 30-second segments were generated with one model and references for the full cast.
The version that workedThe shorter comparison comes from Meyer’s account of the source short; the source material does not independently verify the timing.
A real ghost joins the scare crew
At an extreme haunted attraction, staff treat a supernatural threat like another shift issue. When a guest says the safe word “cotton candy,” the scaring does not stop. Workers try ordinary procedures—including asking whether the ghost signed a waiver.
Haunted attraction
The story places the supernatural inside a workplace built to frighten paying guests.
Policy meets paranormal
Staff respond to a real ghost through familiar shift routines and paperwork logic.
Human judgment
AI generated characters, voices and sets, according to Meyer; pacing, selection and revision shaped the finished short.
A production workflow with human checkpoints
Meyer says he created character references before making video, then revised the workflow after reviewing each attempt. The account includes both useful continuity and failed generations.
Build the cast
Eight character reference images made with Nano Banana 2 through Higgsfield. Meyer says all eight took under two minutes.
Generate shots
The first method used 20 written shots, start frames and Veo 3.1 fast mode for video, speech and sound.
Review failures
Reported issues included letter-by-letter “HEY,” garbled speech and a failed 180-degree head turn.
Rebuild the pace
He checked dialogue against the script, changed a visual setup, and made the final three-segment version.
“Terribly boring and amateurish.”Thorsten Meyer · on his reaction to version one
“The decisive moves were human.”Thorsten Meyer · on key production decisions
Useful figures, with limits
The totals below come from Meyer’s own production account. They are project-specific figures, not independently verified costs or a general estimate for making an AI short.
2,220 credits
Reported Higgsfield usage, valued by Meyer at about $95 under his plan price. The account does not break down credits by version.
~2,100 characters
Reported ElevenLabs usage for music and sound effects. No separate monetary cost is given.
8 references
Character images aimed to keep hair, costumes and identifying details consistent across generated material.
A maker’s account, not an independent audit
The behind-the-scenes article offers concrete production detail, while leaving several questions unanswered.
No receipts, full expense breakdown, unused-credit calculation or allocation across the three versions is provided.
The exact model and workflow for every part of the final three segments are not specified.
The account mentions a final color grade, editing and generated sound materials, but does not detail how much work was manual.
No audience figures or independent assessment of the finished film’s reception are included.
Why the First Cut Failed
Meyer’s account puts the emphasis on editing and performance rather than image generation alone. The first cut had consistent faces, costumes and lighting, he says, but its longer shots and lack of score weakened the comedy. His comparison between 5.6-second average cuts in the first version and about 1.8 seconds in the source short shows how the filmmaker used pacing as a practical production measure.
The process also illustrates where generative tools required correction or workarounds. Meyer says Veo spelled out a shouted “HEY” letter by letter, failed to produce a planned 180-degree head turn and returned garbled speech in one clip. He used speech recognition to check dialogue against the script, and changed the visual setup for the head turn after repeated attempts failed. The final outcome therefore reflects both model output and human selection, editing and grading, rather than a fully automatic process.
For readers tracking AI video production, the account offers a specific example of the trade-offs involved: reference images can help maintain continuity, but they do not solve timing, action or dialogue problems. Meyer’s reported credit total also gives a project-level cost figure, though it should not be treated as a general price for making a short film with AI tools.
Products Worth Considering
Bose S1 Pro+ Portable PA Speaker
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Final Cut
Meyer says production began before any video clips were made. He created one reference image for each of eight characters, including the shift lead, a guest, scare actors and the ghost. According to his account, each image cost two credits at 2K resolution and all eight were generated in under two minutes. He used the images as visual references for later frames and clips, aiming to limit changes in hair, clothing and other identifying details.
The first method used a written list of 20 shots. Meyer generated a start frame for each with Nano Banana 2 and then animated them in Veo 3.1 fast mode, which he says generated speech and sound with the video. After reviewing the result, he judged the pacing too slow and rebuilt the edit. The final version’s three-segment approach differed from the first shot-by-shot structure and included the full cast as references, according to his description.
Meyer says the premise was inspired by an AI horror-comedy short shared on X, while describing Hollow House’s script, characters and jokes as his own. The article does not identify the other short or provide its creator’s account, so the relationship is reported here as Meyer’s description of the inspiration.
““Terribly boring and amateurish.””
— Thorsten Meyer, describing his reaction to the first version
What the Production Account Leaves Open
The production details, timings and costs are Meyer’s own account; no receipts, model logs or independent review are included in the source material. The roughly $95 figure is based on his plan price, and the source does not explain whether that amount accounts for unused credits or other production expenses. It also does not give a breakdown of how the 2,220 credits were divided among the three versions.
The account names the image and video tools and describes several limitations, but does not specify the exact model or workflow used for every part of the final three segments. It is also unclear how much of the sound mix, editing and color grading was done manually, beyond Meyer’s reference to a final color grade and ElevenLabs-generated material. The article gives no audience figures or independent assessment of the completed film’s reception.
The Short and Its Workflow
Meyer’s account links to the completed 1:37 Hollow House short and presents behind-the-scenes images, including character sheets, start frames, failed attempts and a contact sheet of the final cut. Those materials offer readers a way to compare the described process with the finished video.
No additional release, follow-up film or production milestone is announced in the source. Further details about costs, the exact final-generation workflow and audience response would depend on information Meyer or the project’s publisher may share later.
Key Questions
What is Hollow House?
Hollow House is a 97-second horror-comedy short about a real ghost joining the scare crew at an extreme haunted attraction. Meyer says its characters, voices and sets were AI-generated.
How many versions did Meyer make?
Meyer describes three versions: a 20-shot, 2:15 first cut; a 66-cut, 2:21 second version; and the final 1:37 short made from three segments.
What did the production cost?
Meyer reports using 2,220 Higgsfield credits, which he values at about $95 at his plan price, plus roughly 2,100 ElevenLabs characters for music and sound effects. These are figures from his account, not independently verified totals.
Why did the first version not work?
Meyer says the first version looked good but moved too slowly for the comedy. He counted an average of 5.6 seconds per cut, compared with about 1.8 seconds in the short that inspired him.
What is not yet confirmed independently?
The source is Meyer’s production account. It does not include independent verification of the credit expenditure, a complete cost breakdown, audience data or detailed records of each generation and editing step.
Source: ThorstenMeyerAI.com
