TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Thorsten Meyer says he built a hand-drawn video engine in one evening, then found the first full video too slow and visually empty. After revising it, he reports improvements across the first three scenes and says two quality checks now stop overly empty scenes and irrelevant icons. The account is a first-person report based on his own build and incident notes.
Meyer says the idea came after he watched a video using a hand-drawn explainer style, with a camera moving across a whiteboard-like world. To avoid copying it, he says he created a “clean room” brief containing only style traits, without borrowing drawings, assets or layouts. Claude built the engine, according to Meyer, while OpenAI’s Codex reviewed the code. He says five review findings were fixed that evening, each with a test that failed against the earlier code.
Meyer says he watched and approved a roughly one-minute test video later that evening, which was published in an article about local votes against data centers. The next morning, the first full video in the new style looked “slow and boring” to him: it had too much white space, icons that did not match the narration, and scenes that stopped moving once a drawing was complete.
After revising the video, Meyer measured the first three scenes. His notes report that the share of the frame filled with content rose from 15% to 66%, still imagery fell from 29 seconds to zero, and camera moves increased from three to 27. He says two checks now stop a build when scenes are too empty or icons do not illustrate what is being said. These are figures from Meyer’s own project notes, not an independently assessed comparison.
My AI Idea Reached Production With One Boring First Video
Thorsten Meyer reports that a working prototype still produced a slow, visually sparse video. He revised it and added checks before future builds can pass.
From idea to production
Meyer says he built the hand-drawn video engine in one evening, then revised the first full video and published it on his site within hours.
Claude built the engine from a brief describing style traits without copying drawings, assets or layouts.
OpenAI’s Codex reviewed the code; Meyer says five findings were fixed and tested that evening.
After approving a roughly one-minute test, Meyer judged the first full video slow and visually sparse, then revised it.
What changed on screen
His project notes compare the first three scenes before and after revision.
- 15% of the frame filled
- 29 seconds of still imagery
- Three camera moves
- Icons did not match narration
- 66% of the frame filled
- No still imagery
- 27 camera moves
- Two checks flag empty scenes and irrelevant icons
Three decisions in the workflow
The account separates code review, human approval and evaluation of the finished video.
Inspect the code
Codex reported issues; Meyer says it did not build the system.
A person decides
Meyer says he watched and approved the test video before publication.
Check what viewers see
The first full video ran, but its pacing and visuals still fell short for him.
How Video Quality Became Measurable
The account shows how a functioning prototype can still fall short when assessed as a finished product. Meyer’s first full video ran, but he judged its pacing and visuals inadequate. He then translated that critique into measurements and checks that can flag similar issues in later builds.
For readers building AI-assisted tools, the described process makes three separate decisions visible: code review, human approval and output evaluation. Meyer says the reviewer AI does not build the system, and that his own approval was required before publication. His example suggests that passing code checks alone does not establish that a generated video communicates well.
The measurements also have limits. They cover only the first three scenes and track visual properties such as frame coverage and camera movement. They do not, by themselves, show whether viewers understood or liked the video. Meyer does not report audience testing or broader performance results for the new format.
Products Worth Considering
DJI Mic 2 Wireless Microphone System (2 Transmitters, Receiver, Charging Case)
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
DJI Mic 3 Wireless Microphone Kit (2 Transmitters, 1 Receiver, Charging Case)
As an affiliate, we earn on qualifying purchases.
A One-Evening Prototype
Meyer presents the project as a personal account of his production workflow, rather than as a controlled evaluation of AI development. He says the engine was built in one evening and the first full video was revised and published hours after its initial release. The video accompanying his account is described as a roughly four-and-a-half-minute hand-drawn explainer, narrated using an AI voice clone of Meyer.
His notes also refer to earlier experiments intended to support his argument for verification. In a weight-edit experiment, he says a model completed 12 of 15 legal answers, compared with four previously, while its legal score remained at zero. He also says one review round had 12 of the author’s tests passing even though four of six reviewer checks failed. In an August incident, he says a guard generated about 1,300 warnings in a log that nobody read, while an alert email did not go out. These examples are his reported project history; the article does not provide independent records for them.
““The next morning, the first full video in the new look came out. And honestly, it was slow and boring.””
— Thorsten Meyer
What the Measurements Leave Open
The account does not identify the video that inspired Meyer, provide audience feedback on either version, or explain how the two quality gates detect empty scenes and mismatched icons. It is also unclear whether the reported changes improved comprehension or viewer retention; the measurements cited concern the first three scenes’ appearance and motion.
The underlying build notes, review findings and incident records are not included in the source material, so the reported figures and process details cannot be independently checked here. Meyer describes this as a personal account and says its numbers come from his own notes and experiments.
Further Checks on Published Videos
Meyer says the two checks are now gates in the build process, so future videos can be stopped when scenes are too empty or their icons do not match the narration. His account does not give a schedule for further releases or say whether he plans to test the videos with viewers. The next evidence of how well the approach works would come from later videos and any reported audience or quality results.
Key Questions
What did Thorsten Meyer publish?
He published a first-person account of building and revising an AI-assisted, hand-drawn video style for his site. The account was dated Nov. 6, 2026.
How did the first full video change after revision?
Meyer reports that, in the first three scenes, frame coverage rose from 15% to 66%, still imagery dropped from 29 seconds to zero, and camera moves increased from three to 27.
Which AI tools did he say he used?
Meyer says Claude built the video engine and OpenAI’s Codex reviewed the code. He says Codex’s role was to inspect and report issues, not to build.
Were the reported video improvements independently verified?
The source describes them as measurements from Meyer’s own build notes. It provides no independent review or audience testing of the results.
Source: Thorsten Meyer AI
