🔍 Read the full analysis: Taking AI From Idea To Production Without Overcomplicating The First Video on ThorstenMeyerAI.com
Get the latest gadgets 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 a hand-drawn video engine in one evening, but its first full video was slow and visually sparse. He measured the problems, revised the engine and turned the measurements into checks that can stop future videos from being rendered when scenes miss the new standards.
Thorsten Meyer says he turned a hand-drawn video concept into a live format after revising a first full video he found slow and visually sparse. In an account published November 6 on ThorstenMeyerAI.com, he describes using a second AI to review code, human approval and measurable checks to guide the changes—an example of how his team is moving an AI-generated video idea into production.
Meyer says the idea began with a hand-drawn explainer whose style he liked. To avoid copying it, he wrote down a limited set of visual traits, including hand-drawn lines, a continuous whiteboard-like world and large colored numbers. He says the team used no drawings, assets or layouts from the reference video. Claude built a new engine, using colors from Meyer’s design system and an openly licensed handwriting font.
In Meyer’s workflow, Claude builds and OpenAI’s Codex reviews. He says Codex identified five issues in its first review; each was fixed with a test designed to fail against the old code. Meyer also describes watching a roughly one-minute test video and approving it before it went live in an article about local votes against data centers. He says that human decision is distinct from simply confirming that a demo runs.
The first full video in the new look, produced the next morning, did not meet Meyer’s expectations. He identified excessive white space, icons that did not support the narration and images that stopped moving after they were drawn. Rather than revise by feel alone, the team measured frame coverage and still time, then changed the engine so shots filled more of the frame, camera movement accompanied sentences and visual objects related to what was being said.
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 Build Rules
The account’s central point is that production readiness requires more than a working demo. Meyer describes a sequence of review, human approval and measurable checks that made subjective criticism actionable. That approach matters for teams using generative AI to produce media: a system can complete its task while still producing work that is unengaging, poorly matched to its narration or unsuitable for publication.
Turning specific shortcomings into automated gates also changes what happens after a fix. According to Meyer, a scene that is too empty can now be stopped before rendering, while an object that does not illustrate the spoken content triggers a warning. These safeguards do not establish that every video will be effective, but they can prevent the same measured problems from passing through unnoticed.
The figures are limited to Meyer’s own build notes and the first three scenes he describes. They are not an independent evaluation of the system or evidence that the workflow will produce similar results elsewhere. Their value here is as a record of how one production process was adjusted and what its creator chose to monitor.
Popular right now
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, Almond Cream
De’Longhi Classic Matte Black Espresso Machine with Milk Frother
Shark Navigator Lift-Away Upright Vacuum Cleaner, White (NV356E)
As an affiliate, we earn on qualifying purchases.
How the Video Engine Was Built
Meyer presents the video work as one part of a broader account of taking AI ideas into production. He says the initial visual concept became a working engine in one evening, but the first full video arrived the following morning and exposed problems that a short test had not. The account distinguishes that one-minute approved test from the later full video, which remained online unchanged as an example of what he considered unsuccessful.
He also describes a separate experiment involving a local model and legal questions. Meyer says Claude edited the model’s weights, after which it completed 12 of 15 answers in the team’s legal benchmark, compared with four before. However, he says its legal score remained zero. The result illustrates why task completion and task quality need separate measures; the article does not provide enough detail to assess the benchmark or the experiment independently.
Meyer cites an earlier monitoring problem from August: a guard reportedly wrote about 1,300 warnings to a log that nobody read, while an alert email did not go out. He uses the episode to argue that monitoring must reach a person to be useful. These details are the author’s own reported experiences, rather than independently verified findings.
“A demo that runs is not a decision.”
— Thorsten Meyer
Limits of the Reported Results
The article is a first-person account based on Meyer’s build notes, incident notes and experiments. It does not provide an independent review of the engine, its code, the video-quality measurements or the legal benchmark. The frame-coverage and still-time figures apply to the first three scenes of the rebuilt video, not necessarily to all scenes or later productions.
Meyer reports that the improved video went live a few hours after his critique, but the account does not specify how the automated checks perform across a larger set of videos, how often they produce false warnings or whether viewers responded differently to the revised format. The exact details of the legal experiment and its scoring method are also not included in the supplied account. Those limits mean the reported changes should be read as one team’s process and results, not a general measure of AI video production.
Whether the Gates Hold Up
Meyer says the new checks are now part of the video build: they can stop scenes that are too empty and flag visual objects that do not correspond to the narration. The next practical test is whether those rules continue to catch the problems they target in future videos without blocking acceptable work. The published account does not give a schedule for further evaluations or announce a broader release of the engine.
Readers can compare the described process with the video Meyer published alongside the article, which he says was made using the revised approach. Further results—such as checks across more scenes, viewer response or independent assessment—have not been reported in the source material.
Key Questions
What happened to the first full AI-generated video?
Meyer says the video was slow and visually sparse, with icons that did not support the narration and images that stopped moving. He says it remains online unchanged as an example of what not to do.
What changed in the revised video engine?
The team measured how much of the frame was filled and how long images stayed still. Meyer says the rebuilt engine filled more of the frame, added camera movement and required visual objects to relate to the narration.
What do the reported measurements show?
For the first three scenes, Meyer reports that filled-frame coverage rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. These are figures from his own account and apply to those scenes.
Did another AI approve the video?
No. Meyer says OpenAI’s Codex reviews code and does not build the engine. He describes personally watching the test video and deciding whether it was ready to go live.
Are the results independently verified?
The source is Meyer’s first-person report, based on his own notes and experiments. It does not include an independent evaluation of the video engine, measurements or legal benchmark.
Source: ThorstenMeyerAI.com
