🔍 Read the full analysis: Hollow House: Making An AI Horror-Comedy Under A One-Night Deadline on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says he made Hollow House, a 97-second AI-generated horror-comedy, between the evening of October 27 and about 1:30 a.m. the next day. The production took three versions and 2,220 Higgsfield credits; Meyer says the first cut looked polished but lacked the fast rhythm the comedy needed.
Filmmaker Thorsten Meyer says he created Hollow House, a 97-second AI-generated horror-comedy, in three attempts between early evening on October 27 and about 1:30 a.m. on October 28. His behind-the-scenes account describes how the production moved from a visually polished but, in his assessment, boring first cut to a faster final version, highlighting the editing and creative decisions involved in making a short film with AI tools.
The short follows a ghost who joins the staff of an extreme haunted attraction. When a guest says the safe word, “cotton candy,” the scares do not stop. The staff then try to handle the apparent supernatural threat as a workplace problem: shift lead Ray asks whether the newcomer signed a waiver, and a popcorn seller in a priest costume is recruited to attempt an exorcism. Meyer says the script, characters and jokes were his own, while the premise’s structure drew inspiration from an AI horror-comedy short shared on X.
Meyer reports that the first version used 20 single shots and ran 2 minutes 15 seconds. He says the finished short that inspired the project had about 74 cuts over 2 minutes 14 seconds, or an average of roughly 1.8 seconds per image; his first cut averaged 5.6 seconds between cuts. He judged the initial film boring despite its consistent faces, costumes and haunted-house lighting. The second version used 66 faster cuts and ran 2 minutes 21 seconds. For the third, he generated three 30-second segments using one model with the full cast as references; that version became the 1:37 release.
The workflow began with eight character reference images generated with Google’s Nano Banana 2 through the Higgsfield platform. Meyer says those images were used to keep appearances consistent in later frames and clips. He generated video with Google’s Veo 3.1 fast mode, which produced dialogue and sound alongside images. Across all three versions, he reports using 2,220 Higgsfield credits, which he estimates at about $95 at his plan’s price, and about 2,100 ElevenLabs characters for music and sound effects. These are the creator’s reported usage and cost figures, not independently audited totals.
Hollow House: Making an AI Horror-Comedy Under a One-Night Deadline
Filmmaker Thorsten Meyer says he made a 97-second short about a ghost on a haunted-attraction crew in three attempts, working from the evening of October 27 to about 1:30 a.m. the next day. His account makes the edit’s comic rhythm the central challenge.
The released short, according to Meyer.
Three versions, from October 27 into October 28.
“The pictures were never the problem. The rhythm was.”Thorsten Meyer · on the edit
A ghost becomes a workplace problem
Hollow House follows a ghost who joins the scare crew at an extreme haunted attraction. When a guest says the safe word “cotton candy,” the scares continue. Shift lead Ray asks whether the newcomer signed a waiver, and a popcorn seller dressed as a priest is recruited to try an exorcism.
Supernatural, meet HR
The staff respond to a haunting like an operational issue: paperwork, shift duties, and an improvised exorcism.
Original jokes and characters
Meyer says the script, characters, and jokes were his. He says the premise’s structure drew inspiration from an AI horror-comedy short shared on X.
Making the timing land
In Meyer’s assessment, attractive and consistent images did not make the first cut funny. The moments needed to arrive faster.
Polish could not fix a slow pace
Meyer describes a progression from a polished but, in his words, boring first version to a shorter final release. The figures below come from his account.
20 single shots
Ran 2:15. Meyer says it looked coherent but felt “terribly boring and amateurish.” Average time between cuts: 5.6 seconds.
66 faster cuts
Ran 2:21. The edit used more frequent cuts to pick up the pace, though it remained longer than the finished film.
Three segments
Meyer says he generated three 30-second segments using one model with the full cast as references. The release runs 1:37.
Cut rhythm: first version vs. reference short
Meyer reports about 74 cuts over 2:14 in the short that inspired the pacing comparison. Bar lengths use the reported average seconds between cuts.
“The first version looked great, and when I watched it I called it what it was: terribly boring and amateurish.”
Thorsten Meyer · describing version oneA workflow built around continuity and review
Meyer says he created reference images before generating video, then checked the resulting dialogue and action against his intended scenes.
Eight 2K references made with Nano Banana 2 through Higgsfield. Each reportedly cost two credits.
Veo 3.1 fast mode generated image, speech, and sound from the prepared frames and prompts.
Whisper transcripts helped Meyer compare spoken dialogue with the script and spot mismatches.
He revised prompts, retook a garbled line, and used a still pose plus a bone-crack sound for a missed head turn.
A shout became letters
Meyer says a shouted “Hey” was spoken as the letters “H-E-Y,” so he checked the result against the script.
The whole body turned
For a requested head turn, he says the model also turned the character’s body. He adapted the shot using a reversed-head still pose.
Action needed another try
A lifting scene did not work on the initial attempt, and Meyer says he retook a garbled final line.
What the production account counts
These numbers are Meyer’s reported usage and estimates, not independently audited totals or a full production budget.
2,220 credits
Reported Higgsfield credit use across all three versions, estimated by Meyer at about $95 at his plan price.
~2,100 characters
Meyer reports using about 2,100 ElevenLabs characters for music and sound effects.
One night
His account places the work from early evening on October 27 until about 1:30 a.m. on October 28, 2026.
Fast generation still left the key choices with the creator: what to keep, how quickly to cut, and how to adapt when a shot missed its mark.
The practical takeaway from Meyer’s accountWhat this account does—and does not—establish
The published account is a first-person production report dated October 28, 2026. It includes behind-the-scenes images such as character sheets, generated frames, failed attempts, and a contact sheet of the final cut.
The overnight timeline, three attempts, credit usage, estimated cost, and descriptions of the tools and editing decisions are based on Meyer’s account.
The supplied material does not include an independent production log, a full per-tool credit breakdown, or a comparable conventional production budget.
There is no shot-by-shot comparison showing how much of the final pacing came from the third generation, the edit, or other changes between versions.
The examples document one creator’s workflow with specific tools. They do not show how broadly the results apply to other creators or models.
Quick questions
What is Hollow House?
A 1:37 horror-comedy short about a ghost joining an extreme haunted attraction’s scare crew. The staff treat the supernatural visitor as a workplace problem, even after the safe word fails to stop the scare.
How long did it take to make?
Meyer says work ran from early evening on October 27 to about 1:30 a.m. on October 28, 2026, across three attempts. The account does not provide an independently verified production log.
Which tools does Meyer say he used?
Nano Banana 2 through Higgsfield for character references, Veo 3.1 fast mode for video, Whisper for dialogue checks, and ElevenLabs for music and sound effects.
What is the clearest takeaway?
In Meyer’s experience, generating polished images was not enough. The edit’s timing, the selection of shots, and adapting to failed outputs shaped the final short.
Why Editing Set the Pace
Meyer’s account points to a practical distinction in AI-assisted filmmaking: generating attractive images did not, by itself, make the short work as comedy. He says the first cut had coherent visuals but lingered too long between moments, while the later versions increased the pace. That makes shot duration and comic timing, rather than image generation alone, the central production challenge described in his account.
The production also shows how the filmmaker handled inconsistencies and tool limits. Meyer says he used character reference sheets to reduce visual drift, checked dialogue with OpenAI’s Whisper speech-recognition model, and reworked prompts when video outputs missed the intended action. He describes a shout being spoken as the letters “H-E-Y,” a head-turning effect that the model did not produce as requested, and a lifting scene that was blocked on an initial attempt. The examples are one creator’s experience with a specific workflow, not a general evaluation of how every model performs.
The reported cost offers a limited measure of the project’s production inputs, but it does not establish that AI filmmaking is cheaper than a conventional production. The figure covers the cited Higgsfield credits at Meyer’s plan price, while the account does not provide a comparable budget for hiring people, licensing, equipment or post-production. The clearest takeaway is narrower: rapid generation still left important decisions—what to keep, how quickly to cut, and how to adapt failed shots—with the creator.
Products Worth Considering
Logitech MX Keys S Wireless Keyboard, Graphite
Jackery HomePower 1000 Plus v2 Portable Power Station
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
From References to Three Cuts
Before generating video, Meyer created one reference image for each of the short’s eight characters. He says each image was made at 2K resolution with Nano Banana 2 and cost two credits. The references were intended to preserve details such as hair, clothing and appearance from shot to shot, a recurring challenge when separately generated images and clips must appear to show the same cast.
For the first version, Meyer generated 20 start frames and animated them with Veo 3.1 fast mode. He says the model combined video with speech and sound, but the output still needed review. Comparing Whisper transcripts with the script helped identify spoken lines that did not match, including the spelled-out “H-E-Y” and a garbled final line that he retook. For the planned head turn, the model repeatedly turned the character’s body as well; Meyer says he instead used a still pose with the head already reversed and added a bone-crack sound.
The published account is dated October 28, 2026 and includes behind-the-scenes images, including character sheets, start frames, failed attempts and a contact sheet of the final cut. Meyer describes the project as completed over one night and the final short as 1:37. The timeline, workflow and costs in this article are based on his account.
““The first version looked great, and when I watched it I called it what it was: terribly boring and amateurish.””
— Thorsten Meyer, describing his first version
What the Account Does Not Verify
The available account is a first-person production report, and the timeline, credit use and estimated cost are not independently verified in the supplied material. Meyer gives the price as roughly $95 based on his plan, but does not detail the value of any subscription, the precise per-tool allocation of all credits or other production costs.
It is also unclear how much of the final film’s pacing came from the third generation, the edit itself or other changes between versions. Meyer says the final approach used three 30-second segments with the cast supplied as references, but the account does not provide a shot-by-shot comparison of the second and third cuts or detailed settings for the final workflow. The published examples document his process; they do not establish how broadly the results apply to other creators or tools.
The Short and Its Production Record
Meyer’s account links to the completed 1:37 short and presents the behind-the-scenes images used to explain its production. No further release, screening or follow-up project is specified in the source material. The next point readers can verify is the finished film itself, alongside the creator’s examples of reference sheets, generated frames, failed attempts and final-cut contact sheet.
Key Questions
What is Hollow House?
Hollow House is a 1:37 horror-comedy short about a ghost joining an extreme haunted attraction’s scare crew. The staff treat the supernatural visitor as a workplace problem, even after a guest’s safe word fails to stop the scare.
How long did it take to make?
Thorsten Meyer says work ran from early evening on October 27 to about 1:30 a.m. on October 28, 2026, and that the project took three attempts. The published account does not provide an independently verified production log.
What tools does Meyer say he used?
Meyer says he made character references with Nano Banana 2 through Higgsfield and generated video with Google’s Veo 3.1 fast mode. He also reports using Whisper to check dialogue against the script and ElevenLabs for music and sound effects.
What did the project cost?
Meyer reports 2,220 Higgsfield credits across the three versions, which he estimates at about $95 at his plan price, plus about 2,100 ElevenLabs characters. The source does not give a full production budget or an independent cost comparison.
Why were there three versions?
Meyer says the first cut had 20 shots averaging 5.6 seconds each and felt too slow for the comedy. The second used 66 cuts, and the final version used three 30-second segments generated with the full cast as references; Meyer says that third version worked best.
Source: ThorstenMeyerAI.com
