🔍 Read the full analysis: Hollow House: How A One-Night AI Project Became A 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
Hollow House, a 97-second AI-generated horror-comedy, was made overnight from October 27 to 28, according to its creator, Thorsten Meyer. Three versions were produced; Meyer says the first two failed to land, and the final approach used a single model with the cast supplied as references. He reports spending 2,220 Higgsfield credits, about $95 at his plan price, across the attempts.
Thorsten Meyer says he made Hollow House, a 97-second AI-generated horror-comedy, in three attempts between the evening of October 27 and about 1:30 a.m. the next day. His behind-the-scenes account describes a shift from individually generated shots to three longer segments made with one model and references for the full cast, after he judged the earlier cut “terribly boring and amateurish.”
The short’s premise puts a real ghost to work at an extreme haunted attraction, where staff respond to the supernatural intrusion as a workplace problem. When guest Derek says the attraction’s safe word, “cotton candy,” the scare does not stop. Shift lead Ray’s response, Meyer writes, is to ask whether the newcomer signed the waiver. The story also includes a failed exorcism and ends with the ghost working the entrance the following night.
Meyer says the first version used 20 shots, generated from still images and animated with Google’s Veo 3.1 fast mode. It ran 2 minutes 15 seconds, with cuts averaging 5.6 seconds. He compared that pacing with a short that inspired the project, which he counted as having about 74 cuts over 2 minutes 14 seconds, or an average of roughly 1.8 seconds per image. Meyer says the slower rhythm undermined the comedy, despite the images looking good.
For the third attempt, the creator says he generated three 30-second segments using a single model given references for the whole cast; the resulting short runs 1:37. Meyer reports that the three versions used 2,220 Higgsfield credits, which he valued at about $95 at his plan price, plus about 2,100 ElevenLabs characters for music and sound effects. Those are the creator’s figures, not an independently verified production budget.
Behind the making · October 2026
Hollow House: How a One-Night AI Project Became a Horror-Comedy
Three attempts, one overnight production window, and a decisive change in rhythm shaped Thorsten Meyer’s 97-second short about a real ghost on the haunted-attraction staff.
“The pictures were never the problem. The rhythm was.”
Thorsten Meyer · Creatorin one night
at the creator’s plan price
01 / The premise
A ghost joins the scare crew
At an extreme haunted attraction, a real ghost appears on the job. The staff respond as if the supernatural is another workplace issue: when guest Derek says the safe word, “cotton candy,” the scare does not stop. Shift lead Ray asks whether the newcomer signed the waiver.
Haunting as a shift
The ghost becomes part of the attraction’s operations, turning a paranormal intrusion into an employee problem.
Safety policy meets the supernatural
A safe word fails to end the scare. A failed exorcism adds another obstacle to the night.
Back on the clock
The ghost is working the entrance again the following night, closing the story on an absurd workplace punchline.
02 / Three attempts
From individual shots to longer segments
Meyer says the first two versions did not land. The final attempt changed the generation approach: three longer segments, one model, and references for the full cast.
20-shot first cut
Still-image start frames animated with Veo 3.1 fast mode. At 2:15, the cut averaged 5.6 seconds per shot; Meyer judged it “terribly boring and amateurish.”
66-cut revision
A faster 2:21 version tried to push the rhythm. Meyer says the first two attempts still failed to make the comedy work.
Three longer segments
The final 1:37 short used three 30-second segments generated with one model supplied references for the whole cast.
03 / Why pacing mattered
The joke lives in the cut
Meyer compared the first cut with a short that inspired the project. In his count, the reference moved much faster. The comparison helped identify pacing as a problem even when the generated images looked good.
Meyer counted about 74 cuts in the 2:14 reference short. These figures describe his comparison, not an independent analysis.
04 / The production choices
AI generated the material; people shaped the result
Eight character sheets
Meyer says he made one 2K reference image per character with Nano Banana 2 through Higgsfield. He reports two credits per image and under two minutes to create all eight.
Listen, compare, retake
Whisper transcripts were checked against the script. A shouted “HEY!” became spoken letters, and garbled syllables prompted a line retake.
Work around model limits
A requested head turn became a body turn; a lift was reworded as floating. A still image and bone-crack sound handled the head-turn effect.
“Derek says ‘cotton candy’. Nothing stops.”
05 / What the account says
Reported inputs and cost
2,220 credits
Meyer values the Higgsfield credits used across all three attempts at about $95 at his plan price.
About 2,100 characters
Reported ElevenLabs usage for music and sound effects, alongside generated speech and sound in the initial workflow.
Script, pacing, grade
Meyer identifies the writing, change in method, editing rhythm, and final colour grade as creative choices.
The reported figures come from the creator’s account. They are not an independently audited production budget. All characters, voices, and sets are described as AI-generated.
06 / The open questions
What the production account leaves open
The specific model used for the final three segments is not identified beyond the description of one model receiving references for the full cast. The grading process is also not detailed.
The overnight window describes the reported production attempts, not necessarily the full time spent writing, preparing, or finishing the film. The supplied account includes no independent audience or industry assessment.
07 / Key questions
Hollow House, at a glance
What is Hollow House?
A 97-second horror-comedy short about a real ghost joining the scare crew at an extreme haunted attraction.
When was it made?
Meyer says the three attempts ran from the evening of October 27, 2026, until about 1:30 a.m. on October 28.
How many versions were made?
Three: a 20-shot, 2:15 first cut; a 66-cut, 2:21 second version; and the final 1:37 short.
How much did generation cost?
The creator reports 2,220 Higgsfield credits, valued at about $95 at his plan price, plus roughly 2,100 ElevenLabs characters.
Were the performers real?
Meyer says every character, voice, and set was AI-generated, with human decisions shaping script, pacing, and colour grade.
What comes next?
The October 28 account points to the finished short and making-of images. It announces no sequel or further production milestone.
Why Editing Shaped the Final Short
Meyer’s account offers a specific example of how generative video production can depend on choices beyond image generation. He says the first version’s visual quality was not enough: cut length, pacing and sound affected whether the jokes worked. Reworking the production method, rather than simply refining the same shots, produced the final version he considered successful.
The process also shows how a creator can use fixed character images to reduce visual drift. Meyer says each character received a reference sheet before video generation, helping maintain details such as hair and costumes across scenes. He identifies the key decisions as human ones: recognizing that the first cut was dull, changing how the cast was supplied to the video model, and applying the final colour grade.
Products Worth Considering
HP Sprocket Photo Booth Machine Instant Color Photo Printer with 10.1 Touchscreen, LED Ring Light, 250+ Frames, 3×4 ZINK Paper, QR Code Shar
HP Sprocket Panorama Portable Color Photo and Label Printer
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Three Cuts
Before generating video, Meyer says he made one reference image for each of the eight characters with Google’s Nano Banana 2 image model through Higgsfield. He says each 2K image cost two credits and that all eight were made in under two minutes. The references then informed still frames and clips, an approach intended to keep character appearance consistent.
The initial workflow paired 20 generated start frames with Veo 3.1 fast-mode animation. Meyer says the model generated dialogue and sound alongside the images. He checked the resulting speech with Whisper, an OpenAI speech-recognition model, and compared transcripts with the script. That review caught a shouted “HEY!” rendered as the spoken letters “H-E-Y” and prompted a retake of a line that had extra, garbled syllables.
The production account also describes limits encountered during generation. Meyer says a requested head turn became a whole-body turn, an attempt to lift a character was blocked before being reworded as floating, and a planned backward bend instead produced a forward crawl. He says the head-turn effect was handled with a still image and a bone-crack sound. Meyer also credits an AI horror-comedy short shared on X as inspiration for the structure, while saying the script, characters and jokes were his own.
““terribly boring and amateurish””
— Thorsten Meyer, creator of Hollow House
What the Production Account Leaves Open
The production details and cost figures come from Meyer’s own account and are not independently audited. The supplied material does not identify the specific model used for the final three segments, beyond saying that one model received references for the whole cast. It also does not provide independent audience or industry assessments of the finished short.
Meyer describes the final colour grade as one of the decisive human contributions but does not give details of the grading process. The source material also does not specify how much time was spent on writing, preparation or post-production beyond the overnight generation window, so the reported overnight schedule should not be read as a complete measure of all work involved.
The Short and Its Making-Of
Meyer’s October 28 account links to the finished Hollow House short and says it is accompanied by behind-the-scenes images, including character sheets, start frames, failed attempts and a contact sheet of the final cut. The supplied information does not announce a sequel, a further release or another production milestone. For now, the next point of reference is the published film and the making-of material documenting the three attempts.
Key Questions
What is Hollow House?
Hollow House is a 97-second horror-comedy short about a real ghost joining the scare crew of an extreme haunted attraction. Its staff treat the ghost’s arrival as an operational problem, including asking whether it signed a waiver.
How long did it take to make?
Creator Thorsten Meyer says the work took place from the evening of October 27, 2026, until about 1:30 a.m. on October 28. His account describes three versions made during that period.
How many versions were made?
Meyer says there were three attempts: a 20-shot cut lasting 2:15, a second version with 66 fast cuts lasting 2:21, and the final 1:37 short made as three 30-second segments using one model with references for the cast.
How much did the AI generation cost?
Meyer reports using 2,220 Higgsfield credits across all three attempts, worth about $95 at his plan price. He also reports using about 2,100 ElevenLabs characters for music and sound effects. These are figures from the creator, not independently verified costs.
Were the characters and voices real performers?
Meyer says every character, voice and set in the short was AI-generated. His account describes character reference images, generated video and generated speech and sound, alongside human decisions about the script, pacing and final colour grade.
Source: ThorstenMeyerAI.com
