🔍 Read the full analysis: From First Draft To Final Frights: Making Hollow House With AI on ThorstenMeyerAI.com
Get everyday essentials 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 first two cuts missed the intended comic pace; the final version used three longer segments generated with one model and references for the full cast. Meyer reports spending 2,220 Higgsfield credits, about $95 at his plan price, with editing and creative decisions shaping the result.
Creator Thorsten Meyer says he made Hollow House, a 97-second AI-generated horror-comedy, in three versions between the evening of October 27 and about 1:30 a.m. the next day. His account describes how the first cut looked polished but felt boring, and how changing the edit’s pace and giving one video model references for the full cast produced the version he released.
In the film, a ghost joins the staff of an extreme haunted attraction, where the crew treat the supernatural visitor like a workplace issue. A guest says the safe word, “cotton candy,” but the scare continues. When staff realise the tall figure is not one of their performers, shift lead Ray asks whether the newcomer signed a waiver. Meyer’s account says the characters, voices and sets were AI-generated.
The first version consisted of 20 shots and ran 2 minutes 15 seconds. Meyer says he created character reference images, generated still frames for each shot, then animated them with Google’s Veo 3.1 fast mode, which also generated speech and sound. He judged the result visually successful but dramatically weak: the shots averaged 5.6 seconds, while the short that inspired the project averaged under two seconds per cut, according to his count.
Meyer then made a second version of 66 fast cuts, running 2 minutes 21 seconds. The final cut, at 1:37, was assembled from three roughly 30-second segments generated using one model supplied with references for the whole cast. He reports spending 2,220 Higgsfield credits across the three versions, which he estimates at about $95 at his plan price, plus about 2,100 ElevenLabs characters for music and sound effects. Those figures are his reported production costs, not an independently verified budget.
AI Film Case Study · October 27–28, 2026
From First Draft To Final Frights
Hollow House is a 97-second horror-comedy shaped across three overnight versions. Its making offers a creator’s account of how timing, cast references and human decisions guided an AI-assisted film.
Finding the comic rhythm
The account describes a shift from a conventional shot-by-shot process to a shorter final structure generated in longer segments.
20 shots
2:15
Still frames were made for each shot, then animated. Meyer says the images looked successful, but the cut felt boring and amateurish.
Average shot: 5.6 sec66 fast cuts
2:21
The next version pushed the pace with many brief edits. Meyer kept refining the structure to find the timing the comedy needed.
A rhythm experimentThree long segments
1:37
The final version used one model with references for the full cast, assembled from three segments of roughly 30 seconds each.
Final colour grade by MeyerWhat shaped the film
Generation was one part of the workflow. Meyer’s account puts the emphasis on choices made before and after the tools produced material.
Build the cast first
Meyer created full-length reference images for each of the eight characters to help reduce changes in hair, clothing and body shape between clips. He says the sheets took under two minutes to generate.
Animate, then listen closely
The first pass animated shot stills with speech and sound. Meyer also checked dialogue transcripts against the script, catching a spoken “HEY!” rendered as the letters “H-E-Y.”
Choose what works
He identifies calling the first cut boring, giving one model whole-cast references, and applying the final colour grade as decisive human choices.
From character sheets to final cut
A simplified view of the workflow Meyer describes for making and revising Hollow House.
What the figures include
These are figures shared by the creator, not independently reviewed production records.
Across all three versions; Meyer estimates about $95 at his subscription-plan price.
ElevenLabs characters used for music and sound effects, as reported by Meyer.
The ~$95 estimate is plan-price dependent and covers the reported Higgsfield credits. The supplied account does not establish a complete production budget covering every tool, labour input or expense.
Key questions
The supplied account documents one creator’s process and leaves several points unverified.
What is Hollow House?
A 97-second horror-comedy about haunted-attraction staff dealing with a ghost who ignores the venue’s safe-word rules.
How many versions were made?
Three overnight versions: 20 shots in the first, 66 fast cuts in the second, and a 1:37 final cut.
Which tools are named?
Nano Banana 2 for reference images, Google Veo 3.1 for first-version video, Whisper for transcript checks, and ElevenLabs for music and sound effects.
What remains unverified?
The budget, audience response and comparative performance of the final workflow have no independent assessment in the supplied material.
The edit was part of the making.
Meyer’s account presents a fast AI-assisted production whose final shape came through iteration: judging the first cut, changing the pace, keeping the cast visually consistent, and selecting how to finish the result.
Editing Shaped the AI Film
Meyer’s production account puts the emphasis on decisions made around the generative tools, rather than on image generation alone. He says the first cut’s pictures were not the main problem; the timing and rhythm were. Reworking the structure, keeping character references consistent, and deciding which imperfect generated moments to use were part of getting the short to work as comedy.
The example also offers a practical view of AI-assisted production: a film can be generated quickly, but that does not make the creative process automatic. Meyer identifies three human choices as decisive: calling the first cut boring, supplying a single model with cast references, and applying the final colour grade. His account is one creator’s experience, not a controlled comparison of AI filmmaking tools or a measure of how other productions will perform.
Products Worth Considering

Bose QuietComfort Wireless Noise-Cancelling Headphones, Moonlight Grey

Sonos Roam (Gen 1) Portable Smart Speaker, White

Apple MacBook Air 15-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Three Cuts
Before generating video, Meyer says he made one full-length reference image for each of the film’s eight characters with Nano Banana 2, Google’s image model accessed through Higgsfield. He says the images took less than two minutes to generate. The purpose was to reduce visual drift—such as changes to hair, clothing or body shape—between clips. He describes those reference sheets as serving a casting and costume-design role in the workflow.
The first pass used a conventional shot list: make a still for each shot, then animate it. Meyer also checked generated dialogue with Whisper, an OpenAI speech-recognition model, comparing transcripts against the script. He says this caught a line in which “HEY!” was spoken as the letters “H-E-Y” and helped identify garbled speech that needed a retake.
The creator says the premise was inspired by an AI horror-comedy short shared on X, while the script, characters and jokes were his own. His article describes obstacles as well as successful outputs: a requested head turn became a full-body turn, a lift was initially blocked and then rendered only weakly after rewording, and a planned backward bend became a forward crawl. These are Meyer’s descriptions of results from his production, not general guarantees about the models.
““The first version looked great, and when I watched it I called it what it was: terribly boring and amateurish.””
— Thorsten Meyer, creator of Hollow House
What the Production Account Leaves Open
The account does not provide an independently reviewed production log, receipts, or a comparison of the final film against other AI workflows. The stated $95 estimate depends on Meyer’s subscription-plan pricing and covers the Higgsfield credits he reports; it should not be read as a complete cost for every tool, labour input or production expense.
Meyer’s article also does not specify the exact model used for the final three segments, beyond describing one model with the whole cast supplied as references. The source material gives no independent assessment of the film’s audience reception or of whether the final cut achieved the intended comic effect for viewers. Those points remain unreported.
The Released Short and Its Process
Meyer’s October 28 account includes a link to the released Hollow House short and describes the character sheets, opening frames, failed attempts and final-cut contact sheet. The next step for readers is to compare the finished 1:37 film with the production decisions he documents. No follow-up version, release schedule or additional project was announced in the supplied material.
Key Questions
What is Hollow House?
Hollow House is a 97-second horror-comedy short by Thorsten Meyer. Its story follows haunted-attraction workers dealing with a ghost who does not follow the venue’s safe-word rules.
How many versions did Meyer make?
Meyer says he made three versions overnight. The first had 20 shots, the second had 66 fast cuts, and the final edit ran 1 minute 37 seconds.
How much did the AI generation cost?
Meyer reports using 2,220 Higgsfield credits, which he estimates at about $95 at his plan price. He also reports using about 2,100 ElevenLabs characters for music and sound effects. These are creator-reported figures.
Which tools does Meyer name?
His account names Nano Banana 2 for character and shot reference images, Google’s Veo 3.1 for video generation in the first version, Whisper for checking dialogue transcripts, and ElevenLabs for music and sound effects.
What remains unverified?
The production costs, workflow and tool results come from Meyer’s own account. The supplied material does not include independent verification of the budget, audience response or the performance of the final workflow compared with other approaches.
Source: ThorstenMeyerAI.com
