🔍 Read the full analysis: How To Start An AI Production Project With A Simple 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’s account of building a hand-drawn AI video engine describes a short test video followed by a weak first full video and a measured rebuild. He says the process used an independent AI code review, human approval and checks that now block scenes that fail defined criteria.
Thorsten Meyer says he built a hand-drawn explainer-video engine in one evening, but its first full video was slow and visually dull, prompting a rebuild with measured quality checks and automated gates. In an account published on ThorstenMeyerAI.com on November 6 and updated November 9, Meyer describes how a short test video, a separate AI code reviewer and his own approval fit into the process.
Meyer says the project began with a video style he admired. To avoid copying the original, he wrote down a limited set of general traits, including hand-drawn strokes and a continuous illustrated world, and says he used no original drawings, assets, layouts or drawing code. Claude built the engine; OpenAI’s Codex reviewed the code but did not build it. Meyer says five findings from the first review were fixed, each with a test that failed on the earlier code.
A roughly one-minute test video received Meyer’s approval and was published within an article about local votes against data centers. The first full video produced the next morning did not meet his expectations: he described excessive white space, icons that did not support the narration and images that stopped moving after they were drawn. He says that video remains online unchanged as an example of what went wrong.
Rather than adjust the work only by feel, Meyer says the team measured frame coverage and the time scenes remained still. Across the first three scenes of the rebuilt video, filled-frame coverage rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from 3 to 27. These figures describe those three scenes, not a broader benchmark. Meyer says checks now stop scenes judged too empty and flag objects that do not illustrate the narration.
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 Video Critique to Build Gates
The account illustrates a practical distinction for teams putting generative tools into production: a system that can complete a task may still produce work that misses the intended standard. Meyer’s first full video was technically produced, but his review found problems that completion alone would not catch. Turning specific criticisms into measurable checks gives the team a repeatable way to catch some of those problems before rendering.
The process also assigns different responsibilities to the tools and the person overseeing the result. Claude generated the engine, Codex reviewed its code, and Meyer made the publication decision. That division does not establish that an AI reviewer will catch every defect; it does show how this project used a second system as a source of findings rather than as final authority. The account’s central production lesson is that human approval remains part of release, while tests and gates can make particular standards more consistent.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
Bose QuietComfort Wireless Noise-Cancelling Headphones, Sandstone
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
How the First Video Changed
Meyer’s sequence moved from a style idea to an engine, then to a short test video and a full video that he judged unsuccessful. The short test was published the same evening; the first full version appeared the next morning. After documenting its weaknesses, he says the team rebuilt the engine and published an improved-look video a few hours after his critique. The reported frame and motion figures compare the first three scenes before and after that rebuild.
The article also cites a separate experiment involving a local model and legal tasks. Meyer says that after a model’s weights were edited, it answered all 10 published benchmark tasks, up from 4, while its legal score remained zero. He presents this as a caution that a higher completion count is not by itself evidence of improved performance. The article also recalls an earlier monitoring problem: a guard produced about 1,300 warnings in August, but, according to Meyer, nobody read the log and the alert email did not go out.
“A demo that runs is not a decision.”
— Thorsten Meyer
Limits of the Reported Measures
The reported video metrics cover only the first three scenes, and Meyer does not provide a full independent evaluation of the finished video or comparisons across a larger set of projects. The measurements show changes in frame coverage, still time and camera moves; they do not by themselves establish that viewers found the rebuilt video more engaging or that its illustrations were consistently accurate.
The account is a first-person description of Meyer’s own build notes, incident notes and experiments, not an external audit. It does not give details such as the exact thresholds used by the new gates, how often they produce false alarms, or whether the process has been tested by other teams. The legal-model example is also reported without enough information here to assess the benchmark’s scope or the reason its legal score remained zero.
Applying the Checks to New Videos
Meyer says the frame-coverage and illustration checks now act as gates in the video build, while the code-review process uses tests designed to fail on earlier faulty code. The next practical test is whether those controls continue to catch the same kinds of problems across later videos, rather than only the scenes used to revise the engine.
The published account does not announce a release schedule, broader product launch or independent evaluation. Readers can compare the unchanged first full video with the later work on Meyer’s site, but further evidence about viewer response, gate performance and use beyond this project has not been provided.
Key Questions
What is the development Meyer describes?
He describes building a hand-drawn explainer-video engine and changing its production process after the first full video looked slow and visually sparse to him.
Which AI tools were used?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes Codex as a reviewer, not a builder, and says he made the human approval decision.
What changed in the rebuilt video?
For the first three scenes, Meyer reports frame coverage rising from 15% to 66%, still time dropping from 29 seconds to zero, and camera moves increasing from 3 to 27. Those figures apply to those scenes only.
Does the account prove the new videos are better?
No independent evaluation or viewer-response data is reported. The measurements describe selected visual and motion changes, while Meyer’s judgment that the first full video was boring and the later one improved is his own assessment.
What remains unknown about the production checks?
The article does not specify the gates’ exact thresholds, their error rates or how well they perform across a larger body of videos. It also does not report an external review of the workflow.
Source: ThorstenMeyerAI.com
