TL;DR
Get business pricing on monitors, keyboards and dev gear
- 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, but its first full video was slow and visually weak. He measured the problems, revised the engine and turned the improvements into checks that can stop future videos from repeating them.
Thorsten Meyer says he took a new hand-drawn video style from idea to a live test in one evening, then revised it after the first full video fell short. In a post published November 6, he describes using code review by a second AI, human approval and measured quality checks to move the video engine into production on his site.
Meyer says the idea came from a video whose style he liked. To avoid copying it, he set a clean-room rule: his team recorded style traits but reused no drawings, assets or layouts. Claude built a new engine from scratch, using Meyer’s design system and an openly licensed handwriting font. OpenAI’s Codex reviewed the code separately. Meyer says it found five issues, which were fixed with tests designed to fail against the old code.
A roughly one-minute test video received Meyer’s approval and was published the same evening in an article about local votes against data centers. The next morning, however, the first full video looked too empty, its icons did not support the narration, and finished drawings stayed still. Meyer says he turned that critique into measurements for the first three scenes: frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27.
Those measurements became production gates, according to Meyer: a build stops if a scene is too empty or uses icons that do not illustrate what is being said. He also cites other experiments to argue that passing tests alone cannot establish quality. In one review, all 12 tests written by the author passed, while four of six checks by a reviewer failed. These figures come from his own build notes and experiments, not an independent evaluation.
From AI idea to production What the first video taught
Thorsten Meyer built a hand-drawn video engine in one evening. A longer video exposed quality problems, prompting measured checks before future videos can ship.
Build, review, approve
Separate roles kept a working demo from becoming automatic approval.
Claude built an engine from style traits, without reusing the reference video’s drawings, assets or layouts.
OpenAI’s Codex found five issues; fixes included tests designed to fail against the old code.
Meyer approved a roughly one-minute test video before publishing it that evening.
What the full video revealed
The short test did not expose the problems in the first full video.
Too much empty frame
The first three scenes had low frame fill, so Meyer measured it and raised the result from 15% to 66%.
Drawings stayed still
Still time fell from 29 seconds to zero, while camera moves increased from three to 27.
Icons missed the point
Some icons did not support what was being said, so that issue became a production check.
Checks have limits
Passing tests can miss problems, and these reported measurements are not an independent evaluation.
| Evidence | What it shows | What remains open |
|---|---|---|
| First three scenes | Measured visual changes | Results across other videos |
| Author’s tests | All 12 passed | Four of six reviewer checks failed |
| Production gates | Stop scenes that are too empty or use unhelpful icons | Audience understanding and rejection rates |
From Visual Critique to Production Gates
The account shows how a fast prototype can still require several distinct decisions before it is ready for regular use. Meyer’s process separated building, review and approval: Claude generated the engine, Codex reviewed it, and Meyer decided whether the video could go live. That division gives the human a clear responsibility for the result instead of treating a working demo as sufficient approval.
The measured changes also make a subjective criticism more actionable. “Too much white space” became a frame-fill target, while scenes that remain still or use unhelpful icons can now trigger a stop. The gates may catch the specific problems they encode, but Meyer’s account does not show that they can assess every dimension of video quality. His examples suggest that tests and monitoring need review by people and useful alerts to affect production decisions.
Products Worth Considering
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
EcoFlow DELTA 3 Portable Power Station (1024Wh, 1800W)
As an affiliate, we earn on qualifying purchases.
How the First Video Fell Short
The project began with an aesthetic reference, not a finished product specification. Meyer says he wrote down general characteristics of the style while avoiding the original video’s creative assets. He then used those characteristics to guide a new drawing engine. The initial one-minute test looked good enough to him for publication, but the longer video exposed problems that the short demonstration had not.
Meyer’s post places this experience alongside earlier experiments. In a weight-edit test, he says a model completed 12 of 15 legal answers, compared with four previously, while its legal score remained zero. In an August incident, a guard generated about 1,300 warnings in a log nobody read, and an alert email did not go out. The examples are personal reports; the post does not provide independent validation or full methodological details for the experiments.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Report Does Not Establish
The results are Meyer’s own measurements from his build notes. The post does not report an independent review, the number of videos tested beyond the examples described, or whether the new gates reliably improve audience understanding across different subjects. The frame-fill and camera-move figures cover only the first three scenes of one video.
It is also unclear how the system determines that an icon fails to illustrate the narration, or how often the gates reject usable scenes. Meyer’s account describes the intended checks but does not provide longer-term production results, error rates or audience feedback.
How the Checks Will Be Used
Meyer says the two visual rules are now gates that stop a build when a scene is too empty or its icons do not support the spoken content. The next step is to apply them to future videos and see whether they catch recurring problems without blocking acceptable work. The post does not give a schedule for further evaluations or report results from later videos.
His account also points to the need for monitoring to reach a person who can act on it. Meyer says the August warnings accumulated in an unread log and the alert email failed. Whether he has changed that alert path, and how the revised video checks will be monitored over time, remains unreported.
Key Questions
What did Thorsten Meyer build?
He says he built a new engine for producing hand-drawn explainer videos, using style traits as a reference without reusing the original video’s drawings, assets or layouts.
How was the engine reviewed?
Meyer says Claude built it and OpenAI’s Codex reviewed the code. He reports that five review findings were fixed, each with a test that failed against the earlier code.
What changed after the first full video?
After finding too much empty space, unhelpful icons and static drawings, Meyer measured the first three scenes. He reports frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves rose from three to 27.
Do the reported measurements prove the videos are better?
No independent evaluation is described. The figures are Meyer’s own measurements for the first three scenes, and the post does not report audience feedback or results across a larger set of videos.
Source: Thorsten Meyer AI
