🔍 Read the full analysis: From AI Inspiration To Production: A Practical First Video on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says he built and published a hand-drawn video engine after an initial version fell short. His account describes a workflow that combines separate AI coding and review roles, human approval, measured quality checks and build gates; the published results are the author’s own experiments, not an independent evaluation.
Thorsten Meyer says he turned an idea for a hand-drawn explainer video into a working production tool and published a revised video within hours of identifying problems in its first full output. In an account updated November 9, 2026, Meyer describes a process built around separate AI coding and review roles, human approval and quality checks that can block weak scenes before rendering.
Meyer says the idea came from a video style he admired, but he set a clean-room rule: his team wrote down a short list of general traits, including hand-drawn strokes and a continuous illustrated world, and did not reuse the original video’s drawings, assets or layouts. Claude built the engine from scratch, using Meyer’s design system and an openly licensed handwriting font.
OpenAI’s Codex then reviewed the code, according to Meyer. He says Codex had a checking role rather than a building role, and that five findings from its first review were fixed that evening. Each fix was paired with a test that Meyer says failed on the earlier code, a way to check that a test could detect the problem it was meant to catch.
A roughly one-minute test video received Meyer’s approval and appeared in an article about local votes against data centers. But the first full video in the new style, made the next morning, struck him as slow and boring. He identified empty space, icons that did not support the narration and drawings that stopped moving once completed. Meyer says that version remains online unchanged as an example of the shortcomings.
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
Turning Video Critique Into Build Checks
The account’s central point is that a video engine can produce a finished file without producing an effective video. Meyer reports tracking frame coverage, still time and camera movement after reviewing the weak first full version. On the first three scenes of the rebuilt video, he says frame coverage rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from 3 to 27.
Those figures describe Meyer’s measurements of those scenes, not a general benchmark for AI video tools. His practical change was to make some of the observations into production gates: a check can stop a scene judged too empty before it is rendered, while a build can flag an object that does not illustrate the narration. That shifts the process from repeated subjective tweaking toward checks that can catch recurring issues.
The workflow also separates different kinds of responsibility. An AI generates code, another tool reviews it, and a human decides whether the video is ready to publish. Meyer’s account suggests these steps can help a small production workflow make problems more visible, but it does not establish that the same results will hold across other projects or teams.
Products Worth Considering
DJI Osmo Mobile 8 3-Axis Phone Gimbal Stabilizer
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
Bose Lifestyle Ultra Wireless Home Speaker, White Smoke
As an affiliate, we earn on qualifying purchases.
From One-Minute Test to Published Video
Meyer frames the project as a personal account of taking AI ideas into production, with failures included. The sequence began with a style observation, moved to an evening prototype and a short human-reviewed test, then exposed shortcomings in the first complete video the following morning. A revised version went live a few hours after Meyer’s critique, he says.
His updated article also refers to a separate experiment involving a locally run model and legal tasks. Meyer says an AI-assisted weight edit increased the model’s completion count from 4 to all 10 published tasks, while its legal score remained at zero. The result illustrates his warning that completing tasks is not the same as answering them correctly; it is an author-reported experiment, not independent validation of the video workflow.
““It was slow and boring.””
— Thorsten Meyer, describing the first full video
Limits of the Published Results
The reported video measurements cover the first three scenes of the rebuilt video, and Meyer does not provide an independent assessment of whether viewers found the revised version clearer or more engaging. The article’s figures come from his own build notes, incident notes or experiments; the claims have not been independently verified in the supplied material.
It is also unclear how reliably the scene checks judge whether an illustration supports a spoken line, or how often the gates block scenes that a human editor would accept. Meyer’s account mentions an earlier monitoring problem in which a guard generated about 1,300 warnings in a log nobody read and an alert email did not go out. The supplied excerpt does not explain what later monitoring changes were made or provide longer-term results for the video engine.
Whether the Gates Hold Up
The account describes the checks as part of the current build process, but does not give a schedule for further releases or independent testing. The next evidence to watch for is whether the gates keep catching empty or poorly matched scenes across more videos, and whether a human receives and acts on any alerts. Meyer’s updated post says runs for the 10 published Lexassay tasks are public on lexassay.com; those runs concern the separate legal-model experiment, not a validation of the video system.
Key Questions
What did Meyer publish?
He published a personal account of building and refining an AI-assisted engine for hand-drawn explainer videos, including a revised video and the checks he says were added to the workflow.
Which AI tools did he use?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes Codex as a reviewer, not a second coding agent.
What changed after the first full video?
After identifying empty space, static drawings and icons that did not support the narration, Meyer says the team measured scene coverage, still time and camera movement, then added checks intended to stop some weak scenes before rendering.
Are the reported improvements independently verified?
Not in the supplied source material. The video figures are Meyer’s own measurements, and the reported comparison covers the first three scenes of the rebuilt video.
Source: ThorstenMeyerAI.com
