🔍 Read the full analysis: From AI Concept To Production: A First Video And Real Lessons on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says his team turned a hand-drawn video concept into a working production process, then rebuilt it after the first full video fell short. Frame coverage, still time and camera movement became measurable checks, while separate AI review and human approval remained part of the workflow.
Thorsten Meyer has described how he took a hand-drawn video concept from an evening prototype to a production workflow, after the first full video proved too slow and visually static. In an account published on ThorstenMeyerAI.com and updated November 9, Meyer says the team revised the engine using measured scene-level changes, added automated gates and kept human approval as a separate step before publication.
Meyer says the idea came from seeing a hand-drawn explainer whose visual style he liked. To avoid copying it, he says the team recorded a short list of general traits—such as hand-drawn strokes and a continuous visual 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.
A second AI system, OpenAI’s Codex, reviewed the code rather than building it. Meyer reports that Codex identified five issues in its initial review, which were fixed that evening. He says each fix was paired with a test designed to fail on the earlier code. A human then watched a roughly one-minute test video and approved it for publication; Meyer says that version appeared in an article about local votes against data centers.
The first full video in the new style, produced the next morning, did not meet Meyer’s expectations. He describes it as slow and boring, with excessive white space, icons that did not support the narration and finished drawings that remained still. The team measured frame coverage and the time images stayed motionless, then changed the engine. Across the first three scenes of the rebuilt video, Meyer reports that filled frame area rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. Those figures describe three scenes, not the full video or a general benchmark.
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 Feedback to Production Gates
The account matters because it shows the gap between generating a working demonstration and producing an output that is ready to publish. Meyer’s central point is that a video can render successfully and still fail on pacing, composition or whether its visuals support the words being spoken. In this case, subjective criticism led to specific measurements, and those measurements were then used to block later builds that did not meet the team’s standards.
The process also assigns different responsibilities to two AI systems and a human reviewer: one model builds, another looks for problems, and a person makes the publication decision. That division is Meyer’s description of his own workflow, not evidence that the same setup will produce reliable results in other projects. The reported scene figures are useful as an account of iteration, but they do not establish broader quality gains without more detail about evaluation methods and performance across additional videos.
Popular right now
De'Longhi Magnifica Duo Automatic Espresso Machine with Milk Frother – Silver
Hisense E6SR Series 55-Inch 4K UHD Hi-QLED Roku Smart TV
Vitamix VX1 Professional-Grade Blender, 64 oz, Black
As an affiliate, we earn on qualifying purchases.
The First Video Exposed the Gap
Meyer presents the video work as one example of a wider lesson from his AI experiments: improving a surface-level result does not necessarily improve the quality that matters. In a separate weight-editing experiment described in the same post, he says a local model answered all 10 published legal benchmark tasks, compared with four before the edit, while its legal score remained zero. The post does not provide enough detail in the supplied material to assess that benchmark independently; the result is Meyer’s reported finding.
He also cites an earlier monitoring failure from August. A guard reportedly generated about 1,300 warnings in a log that nobody read, while an alert email did not go out. Meyer uses this incident to argue that monitoring only helps if its warnings reach a person who can act. Together, the examples frame the video changes as part of a broader account of testing, oversight and the limits of relying on a single success metric.
“It was slow and boring.”
— Thorsten Meyer, describing the first full video
What the Report Does Not Establish
The account is a first-person report based on Meyer’s own build notes, incident notes and experiments. The supplied material does not include an independent assessment of the video engine, the scene measurements or the legal benchmark. It also does not explain the exact measurement method for frame coverage, how camera moves were counted, or how many scenes were tested beyond the first three in the reported comparison.
The post says the improved video went live a few hours after the critique, but the supplied excerpt does not identify a precise publication time or provide independent confirmation of the workflow. It is also unclear how often the new gates trigger, whether they have prevented later defects, and how the system performs on videos with different formats or subjects. The reported results should not be treated as a general measure of AI video quality.
How the New Checks Will Be Used
Meyer says the frame-coverage and illustration checks now act as build gates: an overly empty scene can stop rendering, and an object that does not illustrate the narration can prompt a warning. The post does not give a schedule for further releases or a planned independent evaluation. The next evidence readers can look for is whether those checks continue to catch problems in later videos, whether human approval remains part of publication, and whether Meyer publishes fuller methods and results for the reported measurements.
Key Questions
What is the development being reported?
Thorsten Meyer describes building a hand-drawn AI video engine and revising it after the first full video fell short. He says the revised workflow uses measured visual checks, code review and human approval.
What changed in the rebuilt video?
For the first three scenes, Meyer reports frame coverage rising from 15% to 66%, still time falling from 29 seconds to zero, and camera moves increasing from three to 27. These figures apply to those three scenes only.
Which AI systems were involved?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes the Codex role as review rather than development, with a human making the final decision to publish.
Does the account prove the workflow will work for other projects?
No. The material is Meyer’s account of his own project, and it does not provide independent testing or results across a wider set of videos.
What remains uncertain?
The post excerpt does not specify the measurement methods in detail, report how often the new gates stop a build, or show how the workflow performs across other video types. Independent validation is not provided.
Source: ThorstenMeyerAI.com
