🔍 Read the full analysis: How We Reworked Hollow House After A Boring First Version 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 his 97-second AI-generated short, “Hollow House,” went through three versions overnight after he found the first cut visually polished but boring. The final version used three roughly 30-second segments generated with one model referencing the full cast; the creator identifies pacing, rather than image quality, as the main problem.
Filmmaker Thorsten Meyer says he rebuilt his 97-second AI-generated horror-comedy “Hollow House” three times in one night after deciding the first version was “terribly boring and amateurish”, despite its polished visuals. In a behind-the-scenes account published October 28, Meyer says the successful third version relied on quicker pacing and a single video-generation model given references for the entire cast.
Meyer says work began in the early evening on October 27 and continued until about 1:30 a.m. the next day. Across three versions, he used 2,220 Higgsfield credits, which he estimates at about $95 under his plan, and roughly 2,100 ElevenLabs characters for music and sound effects. Those figures are his production estimates, not independently verified costs.
The first cut comprised 20 shots and ran 2 minutes 15 seconds. Meyer says its average shot lasted 5.6 seconds. He compared that with the short that inspired the project, which he counted as having about 74 cuts in 2:14, or roughly 1.8 seconds per cut. He concluded that the film’s images were not the main weakness: its slower rhythm left the comedy without effective timing.
For the final cut, Meyer says he generated three segments of about 30 seconds each, producing a 1:37 short, and supplied one model with references for all the characters. Before making video, he created eight character reference images. The story follows staff at an extreme haunted attraction who treat a real ghost as a workplace problem, including whether the unexpected “new guy” signed a waiver.
Why the First Cut Fell Flat
The account offers a specific example of how generative-video production can depend on editing choices as much as image generation. Meyer reports that the first version already had consistent faces, costumes and a red-and-teal haunted-house look, but the long shots and one-line exchanges weakened the horror-comedy’s timing. His own diagnosis was that the film needed faster cuts, not simply prettier pictures.
It also shows the amount of human judgment involved in this kind of work. Meyer says the decisive steps included rejecting the first cut, changing the generation approach and grading the final footage. AI tools produced the characters, voices and settings, but the source does not suggest they independently identified the pacing problem or chose the final structure.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Jackery HomePower 1000 Plus v2 Portable Power Station
Bose S1 Pro+ Portable PA Speaker
As an affiliate, we earn on qualifying purchases.
From Character Sheets to Three Cuts
Meyer says he first made one reference image for each of eight characters using Google’s Nano Banana 2 image model through Higgsfield. He describes these images as a way to keep details such as hair, clothing and costume consistent across later frames and clips. He says the eight images were made at 2K resolution and cost two credits each.
In version one, he generated a start frame for each of 20 shots with the character sheets as references, then animated them using Google Veo 3.1 Fast. Meyer says Veo generated speech and sound alongside the pictures. He also describes production problems: a shouted “HEY!” was spoken as the letters “H-E-Y,” a planned head turn became a body turn, and some requested physical actions did not render as intended. These are his reported results from this project, not a general assessment of the tools.
Meyer says he checked generated dialogue against the script using OpenAI’s Whisper speech-recognition model. That process flagged the spelled-out line and helped identify a garbled final line for a retake. The short’s premise was inspired by an AI horror-comedy shared on X, according to Meyer; he says his script, characters and jokes are original.
““terribly boring and amateurish””
— Thorsten Meyer, creator of “Hollow House”
What the Account Does Not Establish
The production details come from Meyer’s own account. The source provides no independent review, audience data or comparison test showing how viewers responded to the three versions. It is also unclear how much of the final result came from the model change itself, the shorter segments, the faster editing or other revisions made between cuts.
The account says the final version was made using one model with references to the whole cast, but does not identify that model in the supplied material. It also does not provide a complete shot-by-shot breakdown of the final cut or a detailed accounting of the credit estimate and sound-generation use. The reported process should therefore be read as one creator’s production experience, not a controlled test of AI filmmaking tools.
Where the Film Goes From Here
Meyer’s October 28 post includes a link to the finished 1:37 short and describes the accompanying behind-the-scenes material as including character sheets, start frames, failed attempts and a contact sheet of the final cut. The supplied source does not announce a sequel, further release or additional production milestone.
The next development for readers is to watch the film alongside the production materials and judge whether the faster final structure resolves the problem Meyer found in version one. No audience response or independent assessment is included in the source.
Key Questions
What is “Hollow House”?
“Hollow House” is a 97-second horror-comedy short made with AI-generated characters, voices and settings, according to creator Thorsten Meyer. Its story centers on a ghost joining the staff of an extreme haunted attraction.
Why did Meyer remake the first version?
Meyer says the first cut looked polished but felt boring and amateurish. He attributed the problem mainly to slow pacing: its 20 shots averaged 5.6 seconds each, by his count.
How many versions did the film have?
Meyer says there were three versions, made overnight from October 27 into October 28, 2026. The first ran 2:15; the final cut runs 1:37.
How much did the production cost?
Meyer reports using 2,220 Higgsfield credits, which he estimates at about $95 at his plan price, plus roughly 2,100 ElevenLabs characters for music and sound effects. These are creator-reported figures.
Does the account prove that one AI video workflow works better?
No. It documents one creator’s production process and does not include a controlled comparison or audience data. Meyer says the final approach worked for this short, but the source does not establish that it will produce better results in other projects.
Source: ThorstenMeyerAI.com
