🔍 Read the full analysis: Lessons In AI Production: One Evening And One Simple Video on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says he built a hand-drawn explainer-video engine in one evening, but its first full video was slow and visually dull. After measuring the problems, he changed the engine and turned those measurements into checks that can stop similar flaws before rendering. The account is a personal report, not an independent evaluation of the workflow.
Thorsten Meyer says he built a hand-drawn explainer-video engine in one evening, then revised it after the first full video looked slow and empty. In a report published November 6 on ThorstenMeyerAI.com, he describes measuring the problem, adding checks to the production process and publishing a better version hours after his critique.
Meyer says the project began after he watched a video whose whiteboard style he liked. To avoid copying its specific work, he says he wrote down only broad traits: hand-drawn strokes, a continuous visual world, and large numbers used as color accents. He says the engine was built from scratch, with no drawings, assets, layouts or drawing code taken from the reference. The colors came from his own design system, and the handwriting font was openly licensed.
He says Claude built the engine and OpenAI’s Codex, a separate AI system, reviewed the code rather than building it. Codex identified five findings in the first review round; Meyer says he fixed them that evening and wrote a test for each. He describes the tests as failing against the earlier code, a way to check that they could detect the issue rather than merely pass after a change.
A roughly one-minute test video went live that evening within an article about local votes against data centers, after Meyer watched and approved it. The next morning’s first full video did not meet his expectations: he says it had too much white space, icons that did not support the narration, and images that stopped moving once drawn. He left that video online unchanged as an example of the problem.
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
From Visual Critique to Production Checks
The account’s central point is that generating a working demo did not establish that the result was ready. Meyer says he moved from subjective criticism to two measurements: the share of the frame occupied by content and the time the image remained completely still. He then revised the engine so shots filled more of the frame, camera movement followed each sentence, and objects continued to move slightly after being drawn.
For the first three scenes of the rebuilt video, Meyer reports that frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. These figures describe those scenes in his own experiment, not a general benchmark. He says the measurements became gates: a scene that is too empty can be stopped before rendering, while an object with no connection to the narration prompts a warning.
For readers using AI in creative production, the practical relevance is the workflow Meyer describes: one system creates, another reviews, tests target specific failures, and a person remains responsible for approval. His account also shows a limitation: the checks and measurements are specific to his project, and the article does not independently establish how well they perform on other videos.
Products Worth Considering
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
Samsung Galaxy Tab A11 8.7-Inch Wi-Fi Tablet, 64GB, 4GB RAM, Silver
As an affiliate, we earn on qualifying purchases.
How the Video Engine Took Shape
Meyer frames the article as a personal account of taking AI ideas into production, with failures included. The video accompanying it is described as a roughly four-and-a-half-minute hand-drawn explainer about the evening build, the disappointing first full video and the measured revision. Meyer says its narration uses an AI voice clone of himself.
The article also connects the video project to a separate experiment with a local model. Meyer says he asked whether direct weight edits could improve a model’s performance on legal questions after researchers had shown that one AI could repair another by editing its weights. In his test, the model completed 12 of 15 answers, compared with four before the edit, but its legal score remained zero. The report presents completion and legal performance as different measures; the higher completion count did not amount to a better legal score.
Meyer cites another review round in which all 12 of his own tests passed while four of a reviewer’s six checks failed. He also recalls an August monitoring failure: a guard generated about 1,300 warnings in a log that nobody read, and an alert email was not sent. These examples support his stated emphasis on checks that catch failures and reach a responsible person, but they are accounts from his own notes rather than independently audited incidents.
“A demo that runs is not a decision.”
— Thorsten Meyer, in his November 6, 2026, report
Limits of the Reported Results
The figures come from Meyer’s own build notes, incident notes and experiments, according to the source. The report does not provide an independent assessment of the video engine, a full description of the measurement method, or results across a larger set of scenes. It is not clear how the gates handle borderline cases or how often they stop a render in routine use.
The legal-model experiment is also described only in brief in the supplied material. The benchmark’s full scoring method, the model and weight-edit details, and whether the result holds across repeated trials are not stated here. The account says the legal score stayed at zero, but does not provide enough information to assess the experiment independently. The reported August monitoring incident likewise does not specify which system produced the warnings or what changes followed.
The Next Test Is Continued Use
Meyer says the improved video look is now being used in published work, including a video about the Nasdaq’s worst day in 19 months and the explainer accompanying his report. The article does not announce a schedule for further releases or a broader evaluation of the engine. The next evidence readers can look for is whether the production gates continue to catch the problems he identified and whether future accounts report results beyond the first three scenes.
The source also refers to a fuller article about the legal-model experiment, but the supplied text ends before giving its details. Until more information is available, the video measurements and model results should be read as project-specific observations reported by Meyer, not as general evidence about AI production tools.
Key Questions
What is the development described in the report?
Thorsten Meyer describes building a hand-drawn explainer-video engine in one evening, then revising it after the first full video looked too empty and static. He says he added measurements and checks before rendering.
How did Meyer measure the video changes?
For the first three scenes of the rebuilt video, he reports that filled frame area rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves rose from three to 27. These are figures from his project, not an external benchmark.
What roles did the AI systems have?
Meyer says Claude built the engine, while OpenAI’s Codex reviewed the code and reported findings. He says he fixed five findings and wrote a test for each; he also says a human watched and approved the test video before it went live.
Did the separate legal-model experiment improve the model?
Meyer says the model completed 12 of 15 legal benchmark answers after a weight edit, compared with four before. Its legal score remained zero, so the higher completion count did not show an improved legal score.
Are the results independently verified?
The supplied account presents the numbers as Meyer’s own notes and experiments. It does not include an independent audit or enough methodological detail to verify the results beyond his report.
Source: ThorstenMeyerAI.com
