🔍 Read the full analysis: The First AI Video Is Usually Boring—Here’s What Matters on ThorstenMeyerAI.com
Get everyday essentials 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 his first full video from a new AI drawing engine was slow and boring, despite an earlier one-minute demo receiving his approval. He revised the engine using measured visual checks and made those checks build gates; the account is a personal report, not an independent evaluation.
Thorsten Meyer says the first full video made with his new AI drawing engine was slow and visually static, prompting him to revise the system and add checks that can stop weak scenes before rendering, as he describes in his account of taking an AI idea to production. In a personal account published on ThorstenMeyerAI.com, he describes a workflow built around AI code review, human approval and measurable quality gates—a case study in the work required to turn a quick prototype into a publishable video.
Meyer says the idea came from a hand-drawn explainer video whose style he liked. To avoid copying it, he says he wrote down a limited set of visual traits, including hand-drawn strokes and a continuous illustrated world, and used no drawings, assets or layouts from the original. Claude built the engine from scratch, using Meyer’s design system and an openly licensed handwriting font, according to his account.
He says OpenAI’s Codex reviewed the code but did not build the system. Codex identified five issues in its first review round, Meyer writes, and each fix was paired with a test that failed on the old code. He also says a human review was part of the release process: he watched a test video of about one minute and approved it for publication that evening.
The next morning’s first full video did not meet his expectations. Meyer describes excessive white space, icons that did not support the narration and illustrations that stopped moving after they were drawn. He says he measured frame occupancy and still-image duration, then revised the engine. Across the first three scenes, his figures show frame occupancy rising from 15% to 66%, stillness falling from 29 seconds to zero, and camera moves increasing from 3 to 27. Those measurements became checks in the build process, he says.
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 Gates
The account’s main point is that generating a video quickly does not establish that it is clear, engaging or accurate. Meyer’s example shows how a subjective criticism—too much empty space, insufficient motion—can be translated into specific checks that run before a finished video is produced.
That distinction matters to publishers and other teams using generative tools in production. A successful demo can conceal problems that become obvious in a longer output. Code review and automated tests can catch some technical faults, while human approval remains a separate decision about whether a piece is ready to publish. Meyer presents these steps as part of his own process, not as evidence that the method will produce the same results for other creators.
His account also warns against treating a single metric as proof of quality. A fuller frame and more camera movement may address the problems he identified, but those numbers do not by themselves establish that a video is more accurate or useful. The checks describe particular production goals, not a general measure of good video.
Products Worth Considering
Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray
Logitech MX Ergo S Advanced Wireless Trackball Mouse USB-C – Graphite
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
A Prototype, Then a Production Test
Meyer’s November 6 article describes a rapid initial build followed by a less successful full-length test. The short demo was approved and published the same evening, while the first full video, made the next morning, remained online unchanged as an example of the system’s shortcomings. He says an improved video was published a few hours after his critique.
The updated article, dated November 9, says its figures refer to 10 published Lexassay tasks and that the runs are public on lexassay.com. Meyer also describes a separate experiment in which Claude edited a model’s weights. The model answered all 10 tasks in a legal benchmark, compared with four before the edit, but its legal score remained zero, according to Meyer. The result illustrates why task completion alone may not show whether a system performs correctly.
““A demo that runs is not a decision.””
— Thorsten Meyer
What the Measurements Do Not Show
The reported video measurements are Meyer’s own figures for the first three scenes, and the article does not provide an independent audit or a full technical evaluation of the engine. It is not clear whether the revised checks improve viewer understanding, retention or factual accuracy; the measures cited concern frame occupancy, stillness and camera movement.
The account also does not give comparative results across a larger set of videos or explain how the checks handle scenes where stillness or open space may be intentional. The legal-model results are likewise reported by Meyer; the article summary does not provide enough detail to independently assess the benchmark, the model edit or the meaning of its zero legal score.
Further Runs and Human Review
Meyer says the revised video process now uses the measured checks as gates before rendering. His article also points readers to public runs for the 10 Lexassay tasks, providing material for examining that separate experiment. No independent follow-up or broader evaluation is identified in the source account.
For now, the next test is whether the gates continue to catch the problems they were designed to address without treating their metrics as a substitute for editorial judgment. Meyer’s process still calls for a person to review a finished result before publication; the account does not say that the checks can determine on their own whether a video is ready.
Key Questions
What happened to the first full AI video?
Meyer says it was slow and visually boring, with too much white space, unhelpful icons and illustrations that stopped moving. He says it remains online unchanged as an example of what did not work.
What changed in the revised video engine?
Meyer says he measured frame occupancy and stillness, then made checks for empty scenes and unsupported illustrations part of the build process. He reports higher frame occupancy, no still time and more camera moves across the first three scenes of the revised version.
Did an AI reviewer approve the video?
No. Meyer says OpenAI’s Codex reviewed the code and reported issues, while a human made the publication decision after watching the short test video.
Do the reported numbers prove the revised video is better?
No. The figures describe specific visual changes in the first three scenes. The source does not report an independent assessment or audience data showing that the revised video improved comprehension or other outcomes.
What was the result of the legal-model experiment?
Meyer says the edited model answered 10 published tasks, up from four, while its legal score stayed at zero. He presents the result as a warning that completing more tasks does not necessarily mean a model is correct.
Source: ThorstenMeyerAI.com
