🔍 Read the full analysis: A Practical Guide To Turning An AI Idea Into A Production Video on ThorstenMeyerAI.com
Get smart everyday buys 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 built and published a hand-drawn explainer video style after an initial full video proved slow and visually static. His account describes a workflow that pairs AI coding and review with human approval, measured quality checks and build gates; the results are from his own experiments, not an independent evaluation.
Thorsten Meyer says he turned an idea for a hand-drawn explainer video into a published production workflow this week, after rejecting an early full-length version as slow and boring. In an account published Nov. 6 on ThorstenMeyerAI.com, he describes using one AI system to build the engine, another to review its code, and human approval plus measurable checks before publishing. The account is a first-person report, and its performance figures come from Meyer’s own build notes and experiments.
Meyer says the idea began with a video style he liked: drawings on a whiteboard-like scene, with a camera moving across a continuous visual world. To avoid copying the original, he says he wrote down only broad style traits, used no borrowed layouts, drawings, assets or drawing code, and built the engine with Claude. The colors came from his own design system, and the handwriting font was openly licensed.
He assigned OpenAI’s Codex a separate role: reviewing the code rather than building it. Meyer says Codex identified five issues in its first review, which were fixed that evening. He says each finding received a test designed to fail on the old code, so the test could show that it detected the problem rather than merely passing after a change.
A short test video of about a minute went live that evening in an article about local votes against data centers, after Meyer watched and approved it. The next morning, he says, the first full video in the new style did not work well. He documented excessive white space, icons that did not support the narration, and drawings that stopped moving once complete. That original video remains online, unchanged, as a record of the unsuccessful version.
After measuring the footage, Meyer says the team changed how scenes were composed and animated. On the first three scenes of the rebuilt video, the share of the frame filled with content rose from 15% to 66%; time with a completely still image fell from 29 seconds to zero; and camera moves increased from three to 27. These are figures for those three scenes, as reported by Meyer, not independently verified results across a wider set of videos.
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 Video Critique to Build Gates
The account’s central point is that a working AI demo and a production-ready result are different things. Meyer says the weak first full video was improved by turning subjective feedback—too much empty space and too little motion—into specific checks that can stop a later build.
That approach matters to teams making AI-assisted media because it gives them a repeatable way to catch known quality problems before publication. Meyer says the updated system checks whether scenes are too empty and whether objects relate to the narration. Such checks cannot establish that a video is accurate, clear or engaging on their own. His account also treats a responsible person’s review as a required decision, not something automated tests replace.
The report also presents a caution about interpreting technical results. Meyer describes a separate legal-model experiment in which a model completed 12 of 15 benchmark answers after a weight edit, compared with four beforehand, while its legal score remained zero. The task-completion count alone did not show that the model’s legal performance had improved.
Products Worth Considering
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
Apple MacBook Air 15-inch with M5 chip, 16GB memory, 512GB SSD, and Liquid Retina display (Midnight)
Dell UltraSharp 32 U3226QER 4K Thunderbolt Hub Monitor
As an affiliate, we earn on qualifying purchases.
How the Video Engine Was Built
Meyer’s published account connects the video project to a broader set of practices for moving AI ideas into use. He says one system generated the code and a system from another company reviewed it. The reviewer’s role was limited to finding problems; changes still needed tests and a human decision before release.
He uses other examples to explain why passing tests do not settle whether a system is ready. In one review round, he says, all 12 tests written by the author passed, while four of six checks from a reviewer failed. He also recalls an August incident in which a guard generated about 1,300 warnings in a log that nobody read, while an alert email did not go out. These examples are Meyer’s own notes, and underline that monitoring is useful only if failures reach someone able to act.
The video itself is available through the YouTube link included in the account. Meyer says its narration uses an AI voice clone of himself. He also says the improved-look video published a few hours after his critique covered the Nasdaq’s worst day in 19 months; that description is his framing, not a separately assessed market claim in the material provided.
“A demo that runs is not a decision.”
— Thorsten Meyer
Limits of the Reported Results
The reported video metrics cover only the first three scenes of the rebuilt video. The account does not provide a broader sample, independent review of the measurements, or evidence showing how viewers responded to either version. It also does not specify how the system defines or measures whether an object illustrates the narration.
The account describes a personal workflow, not a controlled comparison of AI video production methods. It does not establish that the same process will produce similar results for other teams, topics or video formats. The source material also does not give the publication date or audience figures for each linked video beyond saying when Meyer says they went live.
Checks for Future Video Builds
Meyer says the measurements have become gates in the build process: a scene deemed too empty can be stopped before rendering, and a build can flag an object that does not illustrate anything in the narration. The next practical test is whether those checks continue to catch the same problems in later videos without blocking acceptable work.
His account does not announce a formal evaluation schedule, independent audit or further release milestone. For now, readers can compare the published video with the process he describes, while treating the results as the author’s own reported experience. The unanswered questions include how the gates perform across more scenes and whether human reviewers find the finished videos clear and useful.
Key Questions
What did Thorsten Meyer report?
He described building and publishing a hand-drawn AI video style, then revising it after an early full video had too much empty space and too little movement. His account was published on Nov. 6, 2026.
How did the video change after the first full version?
Meyer says that, across the first three scenes of the rebuilt video, frame coverage rose from 15% to 66%, still-image time fell from 29 seconds to zero, and camera moves rose from three to 27. Those figures apply to those scenes and are reported by Meyer.
Which AI systems did Meyer say he used?
He says Claude built the engine and OpenAI’s Codex reviewed the code. Codex was assigned a checking role rather than a building role, according to his account.
Were the reported improvements independently verified?
The source material presents them as figures from Meyer’s own build notes and experiments. It does not provide an independent audit or a broader evaluation of video quality.
What remains to be tested?
The account does not show how the checks perform across a larger set of videos or whether they improve viewer response. It also leaves open how reliably automated gates can judge whether visuals support the narration; Meyer’s process includes human review.
Source: ThorstenMeyerAI.com
