🔍 Read the full analysis: AI Production For Beginners: Start Small And Learn As You Go 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 describes building a hand-drawn video engine, then revising it after the first full video proved slow and visually dull. His account says the revised process uses independent code review, human approval and measurable checks, while a separate legal-model experiment illustrates why completion rates alone can mislead.
Thorsten Meyer has published a first-person account of building an AI-assisted, hand-drawn video engine and changing it after its first full video proved unsatisfactory. The report describes a workflow combining separate AI code review, human approval and measurable production checks, offering a practical example of the controls he says are needed to move an AI idea beyond a demo.
Meyer says the initial style concept came from a video he admired. To avoid copying it, he says he wrote down a limited set of general traits, including hand-drawn strokes and a continuous illustrated world, and used no drawings, assets, layouts or drawing code from the original. He says Claude built the engine, using his own design system and an openly licensed handwriting font.
In his setup, OpenAI’s Codex reviewed the code but did not build it. Meyer says Codex identified five issues in the first review and that each fix was paired with a test that failed against the earlier code. He also describes watching a roughly one-minute test video himself and approving it before it went online. He stresses that a working demo alone was not the basis for publication.
The first full video made the next morning was, in Meyer’s description, slow and boring. He recorded concerns about empty space, icons that did not support the narration and drawings that stopped moving when complete. He says the team then measured frame coverage and still-image time, and added checks to the build. On the first three scenes, the filled-frame measure rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from 3 to 27. Meyer says those figures are now used as production gates.
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
Turning Video Critique Into Checks
The account matters as a practical description of the gap between producing AI output and deciding whether it is ready to publish. Meyer’s first full video ran, but he found it visually weak. His response was to turn specific criticisms into measurements and checks that can stop a later build, rather than relying only on repeated subjective adjustments.
That approach does not make the output automatically good. Frame coverage and camera movement describe particular visual features; they do not establish whether a video is accurate, useful or engaging. Meyer also says that a person responsible for the result must review it. The report presents this as one creator’s process, not independent evidence that the same workflow will produce similar results for others.
A second experiment in the article reinforces the limits of relying on a single metric. Meyer says a model edited for a legal task answered all 10 published benchmark tasks, up from four, while its legal score remained zero. The figures are from his own experiment, and the article’s point is that a higher completion count did not, by itself, show an improvement in legal performance.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
From One Test Video to Gates
The account is a personal report rather than a peer-reviewed study or an independent evaluation. Meyer says its figures come from his build notes, incident notes and published experiments. In an update dated November 9, 2026, he says the legal figures refer to 10 published Lexassay tasks and that the runs are public on lexassay.com.
He describes an initial one-minute test video going online the evening the engine was built, within an article about local votes against data centers. The first complete video followed the next morning and remains online unchanged as an example of what he considered a poor result. A rebuilt video, using the revised visual checks, was later published in an article about the Nasdaq’s worst day in 19 months. Meyer says it was released a few hours after his critique of the first full version.
The article also recounts an earlier monitoring failure from August. Meyer says a guard generated about 1,300 warnings in a log that nobody read, while an alert email did not go out. He uses that incident to argue that monitoring is useful only when its signals reach someone able to act on them.
“A demo that runs is not a decision.”
— Thorsten Meyer
Limits of the Reported Measures
The account does not provide an independent review of the video engine, its measurements or the legal-model experiment. The reported visual figures cover the first three scenes; the article does not establish that the same changes apply across every scene or project. The measurements also do not directly assess factual accuracy, accessibility, audience response or whether the visuals communicate the narration well.
The legal experiment’s task results and score are reported by Meyer, but the supplied account does not explain the scoring method, the exact model or the details of the weight edit. The basis for comparing the results beyond the stated 10-task set is not fully described here. The article also does not give audience data for the videos or explain how often the new gates reject a build.
How the Production Gates Develop
Meyer says the frame-coverage and still-time checks now act as gates in the video build, with scenes flagged or stopped when they fail the stated criteria. The account does not announce a future release date or a formal product launch. The next evidence to watch for is whether the checks remain in use across more videos and whether Meyer publishes further results, including how the gates perform beyond the first three measured scenes.
For readers considering a similar process, the report’s immediate lesson is limited but concrete: test a small version, record specific shortcomings, check fixes against earlier behavior, and have a responsible person review the finished work. Meyer’s examples do not establish a universal standard; they show how one creator says he is making his own production decisions more visible and repeatable.
Key Questions
What did Thorsten Meyer report?
He described building an AI-assisted hand-drawn video engine, revising it after an unsatisfactory first full video, and adding code review, human approval and visual checks.
What changed in the revised video measurements?
For the first three scenes, Meyer reports that filled-frame coverage rose from 15% to 66%, still time dropped from 29 seconds to zero, and camera moves rose from three to 27. These are his reported measurements, not an independent evaluation.
Which AI tools did Meyer say he used?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes Codex as a reviewer, not the system that built the engine.
Did the legal-model experiment show that the model became better at law?
Not on the reported legal score: Meyer says the model answered all 10 published tasks, compared with four before, while its legal score remained zero. The account does not provide enough detail to independently assess the experiment.
What remains unknown about the video workflow?
The report does not provide independent validation, audience-response data or results across a larger set of scenes. It also does not show whether the visual checks reliably predict overall video quality.
Source: ThorstenMeyerAI.com
