🔍 Read the full analysis: From Brainstorm To AI Production: What I Learned In One Evening on ThorstenMeyerAI.com
Get hardware and tech 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 he turned an idea for a hand-drawn video style into a working engine in one evening, then improved its first full video after measuring problems with empty space and still images. The account describes code review by a second AI, human approval and automated checks, while stressing that passing tests or completing a task does not prove an AI system is producing a good result.
Thorsten Meyer says he turned an idea for a hand-drawn explainer-video style into a working engine in one evening, then rebuilt parts of it after the first full video looked slow and boring. In an account published November 6, 2026, on ThorstenMeyerAI.com, Meyer describes how AI code review, human approval and measurable production checks shaped the work before a revised video went live.
Meyer says the original inspiration was a video whose hand-drawn look he admired. To avoid copying it, he wrote down a short list of broad traits: hand-drawn strokes, a continuous visual world rather than slides, and large numbers used as colour accents. He says the team used no drawings, assets or layouts from the video that inspired the idea. Claude built the engine from scratch, using Meyer’s own design colours and an openly licensed handwriting font.
Meyer says OpenAI’s Codex reviewed the code that same evening but did not build it. It reported issues for Meyer to assess; five findings were fixed, he says, and each fix received a test designed to fail against the earlier code. Meyer describes that sequence as a way to check whether a test can detect a bug, rather than merely confirm a change after it has been made.
After watching a test video of about one minute, Meyer approved it for publication. He says it went live inside an article about local votes against data centers. The first full video in the new style, produced the next morning, drew a different response: Meyer found it had too much white space, icons that did not support the narration, and images that stopped moving when the drawing was complete. That version remains online as an example of the initial result, he says.
Infographic 01 · The new video look
One evening, one boring first video, one measured fix
Speed was not the point. The steps in between were: a second AI that reviews, a human who says yes, and numbers that became gates.
Share of the frame filled with content on the first three scenes after the fix. Before: 15%.
From an evening idea to a measured fix
Relative times, this week
- EveningA video I likedClean room: only a list of style traits, no drawings, assets or layouts
- Same eveningOwn engine, first reviewClaude builds; OpenAI’s Codex reviews; 5 findings fixed, each with a test that fails on the old code
- Same eveningA human says yesOne-minute test video approved and published
- Next morningFirst full video: slow and boringToo much white space, icons that did not fit, a picture that stood still
- Hours laterMeasured fix, then gatesThe first video in the improved look goes live
Source: my own build notes for the new video look, this week (engine history, review round, approval, critique, before/after measurement of the first three scenes).
Powered by Thorsten Meyer AI
How Video Quality Became a Build Check
The account offers a concrete example of a distinction that can get lost in demonstrations of AI tools: making something run is not the same as making it fit for publication. Meyer’s first full video exposed problems that were not resolved by the fact that the engine could generate a video. He says he responded by measuring how much of the frame contained content and how long the picture remained still, rather than relying only on subjective tweaking.
On the first three scenes of the rebuilt video, Meyer reports that frame coverage increased from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. Those figures apply to those three scenes, not necessarily to every video or to a general measure of quality. Meyer says the measures were then turned into gates: a scene deemed too empty is stopped before rendering, while an object that does not illustrate the narration triggers a build warning.
That approach matters to readers assessing AI-assisted production because it places human judgment and observable checks alongside generation speed. The account is a personal report about one creator’s workflow, not an independent evaluation showing that the method will work equally well for other teams or content.
Products Worth Considering
Logitech MX Keys Combo for Business Keyboard and Mouse – Graphite
Apple MacBook Pro 14-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
Ledger Nano Gen5 Crypto Wallet, Black
As an affiliate, we earn on qualifying purchases.
From Style Reference to Original Engine
The project began with a visual reference, but Meyer says the team translated that reference into general design traits rather than reusing its specific creative material. He says the engine was written from scratch and the palette came from his own design system: blue for the image and warm orange to highlight a key point in each scene. The article also describes the handwriting typeface as openly licensed.
Meyer presents the video project alongside a separate experiment from the same week involving a local model and legal questions. He says Claude edited the model’s weights, after which it completed 12 of 15 answers in a legal benchmark, compared with four beforehand. But the model’s legal score remained zero, according to Meyer. The result underscores his stated point that a task-completion count alone can give a misleading impression of performance; the source account does not provide enough benchmark detail to assess the test independently.
The article also refers to an August monitoring failure: a guard generated about 1,300 warnings in a log that nobody read, while an alert email did not go out. Meyer uses this incident to argue that monitoring only helps if it reaches a person. The account’s broader workflow therefore includes code review, tests, human sign-off, production thresholds and attention to whether alerts are actually delivered.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Report Does Not Establish
The account is written by the person who built the engine and reports figures from his own build notes, incident notes and experiments. It does not offer an independent review of the code, the videos, the benchmark design or the reported measurements. The frame-coverage and motion figures refer to the first three scenes of one rebuilt video; the article does not give a comparable evaluation across a larger set of videos.
It is also unclear from the supplied account how the checks define an overly empty scene or determine whether an object accurately illustrates narration, and whether those checks can miss problems. The legal-model result lacks details such as the benchmark’s full scoring method and the number or difficulty of questions. The source says the legal score remained at zero but does not explain the scoring scale. The article presents the workflow as Meyer’s practice, not as proof that AI review or automated gates prevent all production errors.
The Gates Face Future Videos
Meyer says the revised production rules are now part of the engine: scenes that are too empty can be stopped before rendering, and mismatches between a visual object and the narration can generate a warning. The source does not specify a future release date, a planned evaluation or whether the checks will be revised as more videos are made.
The next practical test is whether the gates continue to catch problems in later productions and whether alerts reach a human who can act on them. Meyer says the improved-look video went live a few hours after his critique of the first full version; the hand-drawn explainer accompanying the account uses the same approach. Further performance claims would need to distinguish results on individual scenes from results across a broader body of work.
Key Questions
What did Meyer build?
Meyer says he created an engine for making hand-drawn explainer videos, with a continuous visual world, camera movement and animated drawings. He says Claude built the code from scratch using his own design system.
How did the first full video perform?
Meyer described it as slow and boring, citing too much white space, icons that did not support the narration and pictures that became still after the drawing finished. He says that version remains online unchanged.
What changed in the rebuilt video?
For the first three scenes, Meyer reports that frame coverage rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. He says those measurements became checks that can stop or flag later builds.
Did a second AI approve the code?
Meyer says OpenAI’s Codex reviewed the code and reported findings, while Claude built the engine. Codex did not make the changes, and the account says Meyer fixed five findings and added tests; Meyer also watched and approved the test video himself.
Does the account prove the workflow will work for other teams?
No. It documents one creator’s experience and reports results from his own notes. It does not provide an independent audit or broad testing across other teams and video projects.
Source: ThorstenMeyerAI.com
