TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Thorsten Meyer says he built a hand-drawn video engine in one evening, but the first full video produced the next morning was slow and visually sparse. He measured changes to the opening three scenes, then set thresholds intended to stop similar problems from passing; the account is based on his own build notes and experiments.
Thorsten Meyer says he built a new hand-drawn video engine in one evening, then revised its first full video after finding it slow and visually sparse. In an account published November 6, he describes using a separate AI reviewer, human approval and scene measurements to move the format onto his site, while stressing that the results come from his own build notes and experiments.
Meyer says the inspiration was a video with a camera moving through a hand-drawn whiteboard world. He set a clean-room rule: record general style traits, but use none of the original video’s drawings, assets or layouts. Claude built the engine from scratch, according to Meyer, using his design system’s colors and an openly licensed handwriting font.
That evening, OpenAI’s Codex reviewed the code in Meyer’s setup. He says Codex identified five findings, which were fixed with tests designed to fail on the old code. Meyer then watched a roughly one-minute test video and approved it for publication. He says it appeared that evening within an article about local votes against data centers.
The first full video in the new style was ready the next morning, but Meyer judged it slow and boring: it left too much white space, used icons that did not support the narration and held still after drawings were complete. He measured the first three scenes before and after a revision. In his notes, frame fill rose from 15% to 66%, time spent with a still image fell from 29 seconds to zero, and camera moves increased from three to 27. He says two thresholds now stop a build when the frame is too empty or an icon does not illustrate the narration.
Building an AI Video in One Evening
Thorsten Meyer built a hand-drawn video engine, then revised its first full video after finding it slow and visually sparse. His account shows how scene measurements and human review can shape a production workflow.
From idea to first video
Meyer says the build and review happened during the preceding week.
He recorded general style traits and avoided the example video’s drawings, assets and layouts.
Claude built the engine; OpenAI’s Codex reviewed the code, and Meyer says findings were paired with tests.
After approving a short test, he revised the first full video when it felt slow and sparse.
What changed on screen
Measurements cover the first three scenes, before and after revision.
Production roles
- Claude built the engine.
- OpenAI’s Codex reviewed the code.
- Two gates stop builds for sparse frames or mismatched icons.
- Meyer watched and approved the short test.
- He judged the first full video and chose what to revise.
- His account leaves publication decisions with a person.
From Visual Critique to Production Gates
The account shows how a quick AI-assisted prototype can require substantial review before it is ready for publication. Meyer describes translating subjective feedback about pacing and empty space into measurable checks, then using those checks as gates in the production process. That could make recurring defects easier to catch, although his account does not establish how well the gates perform across other videos or projects.
His other examples make the distinction between a working demo and a reliable production process. Meyer says that in a weight-edit experiment, a model completed 12 of 15 legal answers, up from four, while its legal score remained at zero. In a separate review round, he says all 12 tests written by the author passed, but four of six checks by the reviewer failed. The examples underline his stated point: passing tests alone does not prove the result is sound.
Monitoring also depends on someone receiving and acting on alerts. Meyer recalls that a guard generated about 1,300 warnings in August, but nobody read the log and an alert email did not go out. That episode gives context for why the video workflow’s human approval step matters: automation can flag issues, but the account describes a person as responsible for deciding whether a result should be published.
Products Worth Considering
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
Apple MacBook Air 15-inch with M5 chip, 16GB memory, 512GB SSD, and Liquid Retina display (Midnight)
Bose SoundLink Flex Bluetooth Speaker (2nd Gen), Black
As an affiliate, we earn on qualifying purchases.
A One-Week Build and Revision
Meyer’s article is a first-person account, not an independent evaluation of an AI video product. It was published November 6, 2026, and describes work he says took place during the preceding week. The reported sequence runs from watching an example video, to writing down abstract style traits, building an engine, reviewing a short test, and then revising a full video after seeing its shortcomings.
The described workflow assigns different roles to two AI systems: Claude builds, while OpenAI’s Codex reviews. Meyer says each finding from the initial review was paired with a test that failed on the earlier code. He also describes his own approval as a required step. The video accompanying the post is described as a roughly four-and-a-half-minute hand-drawn explainer with narration using an AI voice clone of Meyer.
“Speed was not the point.”
— Thorsten Meyer
How Well the Gates Generalize
The measurements apply to the first three scenes of this video, according to Meyer’s build notes. The account does not provide the full video, a detailed measurement method, or results from a larger set of projects. It is also unclear how the two gates are calibrated, how often they reject otherwise usable scenes, or whether they have prevented similar problems since the revision.
The code-review findings, weight-edit results and August warning count are reported by Meyer from his own notes and experiments. The post does not include an outside audit or enough detail to independently reproduce those results. It also does not specify the time needed for later production runs or compare the process with a non-AI workflow.
Testing the Rules on Future Videos
Meyer says the frame-fill and illustration checks are now gates that stop a build when a scene is too empty or an icon does not match what is being said. The next indication of whether the approach holds up will be how those rules perform on subsequent videos and whether human review catches problems the automated checks miss. The article does not give a schedule for further releases or report longer-term results.
Key Questions
What did Thorsten Meyer build?
He says he built an engine for producing hand-drawn explainer videos, with a camera moving through a continuous illustrated world.
How did he address the first video’s problems?
After describing the first full video as slow and visually sparse, Meyer measured the first three scenes and revised them. He reports frame fill rising from 15% to 66%, still-image time falling from 29 seconds to zero, and camera moves increasing from three to 27.
Which AI systems did the workflow use?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes Codex as a reviewer rather than a builder.
Was the video independently evaluated?
The source is Meyer’s personal account, based on his own build notes and experiments. It does not report an independent evaluation of the video, measurements or review process.
Source: Thorsten Meyer AI
