🔍 Read the full analysis: From AI Idea To Working Video: Lessons From A Single Evening on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
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 sparse. After measuring the problems, he added checks that block scenes with too much empty space or visuals that do not match the narration. The account is a personal report, and its figures come from his own build notes and experiments.
Thorsten Meyer says he built a hand-drawn explainer-video engine in one evening, then redesigned it after its first full video looked empty and moved too slowly. In an account published on ThorstenMeyerAI.com and updated November 9, Meyer describes using a second AI to review code, human approval to decide whether work could go live, and measurable checks to prevent recurring video problems.
Meyer says the initial idea came from liking the style of another video, but he set a “clean room” rule for the project: record broad style traits, such as hand-drawn strokes and a continuous visual world, without using the original video’s drawings, assets, layouts or code. He says Claude built a new engine from scratch, using his own design system and an openly licensed handwriting font.
In Meyer’s workflow, OpenAI’s Codex reviews rather than builds. He says Codex identified five issues in the first review and that each fix was paired with a test that failed on the old code. Meyer argues this makes the test evidence that it can detect the problem, rather than merely confirming a change after it has been made.
A roughly one-minute test video received Meyer’s approval and went live the same evening. The first full video, made the next morning, did not satisfy him: he recorded too much white space, icons that did not support the narration and images that stopped moving once drawn. He says that version remains online unchanged as an example of the shortcomings.
For the first three scenes of a rebuilt video, Meyer reports that the filled portion of the frame rose from 15% to 66%, time spent completely still fell from 29 seconds to zero, and camera moves rose from 3 to 27. He says these measurements became build gates: scenes can be stopped if they are too empty, and the system can flag objects that do not illustrate what is being said. The improved-look video was published a few hours after his critique, according to his account.
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 Checks Shape the Workflow
The account’s central point is that a tool producing a video is not the same as a tool producing a video fit to publish. Meyer’s first full output ran, but he judged it slow and visually weak. His response was to turn specific criticisms into measurements and checks, instead of relying only on repeated aesthetic adjustments.
That approach may matter to creators using AI in production because it assigns different roles to generation, review and approval. Meyer says Claude wrote the engine, Codex searched for defects, and he made the publishing decision. The division does not establish that the system is reliable in general, but it shows how this creator says he tried to make failures visible and prevent known defects from recurring.
The numbers are limited to the first three scenes of the rebuilt video and come from Meyer’s own notes. They describe frame coverage, still time and camera moves—not audience response, factual accuracy or overall creative quality. Treating them as evidence of broader performance would go beyond what the account reports.
Popular right now
De'Longhi Magnifica Start Fully Automatic Espresso Machine with Milk Frother, Silver
Samsung 43-Inch M70H Series 4K Mini LED Smart TV (2026)
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Almond Cream
As an affiliate, we earn on qualifying purchases.
From Style Reference to Original Engine
Meyer frames the project as a personal account of moving from an idea to a production tool. The spark was a video he liked, but he says the team wrote down only general characteristics and avoided copying its visual materials. The engine’s colors came from his design system: blue for the main image and orange to mark a key point in a scene.
The article also describes a separate experiment from the same week. Meyer says a model altered through direct weight editing answered all 10 published legal tasks, up from four, while its legal score remained zero. The article’s update says the figures refer to those 10 published Lexassay tasks and that the runs are public on lexassay.com. Meyer uses the result to warn that a higher completion count does not by itself mean better performance on the intended measure.
He also points to a monitoring failure from August: a guard reportedly wrote about 1,300 warnings to a log that nobody read, while an alert email did not go out. The account does not give a comparison baseline or a broader incident dataset for that count. Meyer’s stated lesson is that monitoring has limited value if it does not reach someone who can respond.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Report Does Not Establish
The account is a first-person report, not an independent evaluation of the engine. Meyer says the figures come from his build notes, incident notes or published experiments, but the article does not provide an outside assessment of video quality, the reliability of the gates, or how the tool performs across a larger set of projects.
The reported improvements cover three scenes, and the account does not specify how the measurements were calculated or whether they predict viewer engagement. It also does not provide evidence that the engine’s checks catch every mismatch between narration and visuals. The legal-model result is a separate experiment; a score of zero remained despite the higher number of completed tasks.
It is also unclear how the workflow will change with future videos, whether the quality gates will need revision, and how the creator resolves borderline cases that require human judgment. Meyer’s account describes his own process and results; it does not show that the same setup will work for other teams or tools.
Further Videos Will Test the Gates
Meyer says the improved video is live and that the same production method was used for the video accompanying his article. The account does not announce a release schedule, outside evaluation or additional benchmark results. The immediate next test is whether the new scene gates continue to catch sparse frames and unsupported visuals in later productions.
Readers can compare the first full video, which Meyer says remains online unchanged, with the later version to see the change he describes. Any stronger judgment about consistency will require further examples and clearer reporting on how the checks are calculated and how often they stop a build.
Key Questions
What did Meyer build?
He says he built an AI-assisted engine for hand-drawn explainer videos in one evening. The first full video came the following morning, and he later rebuilt the engine after identifying visual problems.
How did the revised video metrics change?
For the first three scenes, Meyer reports that frame coverage increased from 15% to 66%, still time fell from 29 seconds to zero, and camera moves rose from three to 27. These are his reported measurements for those scenes, not an independent evaluation.
What roles did the AI systems have?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code for problems. He describes human approval as a separate step before publishing.
Does the report prove the video engine works reliably?
No. It describes one creator’s workflow and reported results. The article does not provide an independent test across multiple projects or evidence that the quality checks catch every defect.
What does the legal-model experiment show?
Meyer says a model answered 10 published legal tasks after a weight-edit experiment, compared with four before, while its legal score stayed at zero. The result illustrates why completion counts should not be treated as proof of performance on the intended measure.
Source: ThorstenMeyerAI.com
![Douglas Hofstadter: Analogy As The Core Of Cognition [Video]](https://micronomicon.com/wp-content/uploads/2026/09/douglas-hofstadter-analogy-as-the-core-of-cognition-video-featured-560x300.jpg)
