🔍 Read the full analysis: One AI Idea, One Evening, One First Video: What I Learned on ThorstenMeyerAI.com
Get hardware and tech 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 he built a hand-drawn video engine in one evening, then found its first full video slow and visually dull. After measuring frame coverage, still time and camera movement, he turned the results into checks for later videos. The account is a personal report based on his own build notes and experiments, not an independent evaluation.
Thorsten Meyer says he built a hand-drawn video engine in one evening, but its first full video the next morning was slow and visually dull. In a personal account published November 6 and updated November 9, he describes using a second AI to review code, his own approval before publication, and measured checks to revise the system. The account matters as a concrete example of the work between an AI-generated demo and a production video; its results are reported by Meyer from his own notes, not independently verified.
Meyer says the idea came from a hand-drawn explainer video he liked. To avoid copying it, he says he wrote down a limited set of style traits—such as hand-drawn lines and a continuous visual world—without reusing its drawings, assets or layouts. He says Claude built the engine from scratch using his own design system and an openly licensed handwriting font.
In his workflow, OpenAI’s Codex reviewed the code but did not build it. Meyer says Codex found five issues in the first review and that each was fixed with a test designed to fail on the old code. He also describes watching and approving an approximately one-minute test video before it went live in an article about local votes against data centers. He presents that human review as a decision point, not a formality.
The first full video in the new style did not meet his expectations. Meyer says it had too much empty space, icons that did not support the narration, and drawings that stopped moving once completed. He measured the first three scenes of a rebuilt video: frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. He says those measures became build gates intended to catch scenes that are too empty or visuals that do not match the spoken content.
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 Metrics to Build Gates
The account highlights a practical distinction for people using AI to make media: producing an output is not the same as producing a satisfactory one. Meyer says the first full video was finished but failed his own creative review. Measuring specific shortcomings gave him a basis for changing the engine rather than relying only on repeated subjective tweaks.
Turning those measures into checks may also make later production more consistent. According to Meyer, the system now stops a scene when it is too empty and flags visuals that do not illustrate what is being said. Those controls address some identifiable production problems, but they do not establish that a video is accurate, useful or legally safe. A human still has to judge the result, and the account does not provide an independent assessment of the engine or its output.
Meyer connects the video work to a separate experiment in which a model’s completion count rose while its legal score remained zero. That example reinforces his central caution: a better-looking metric can conceal a failure on the outcome that matters. The lesson applies to AI workflows beyond video, though the examples and figures here are his own.
Products Worth Considering
Ledger Flex Crypto Wallet
Logitech MX Master 3S Bluetooth Wireless Mouse, Graphite
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen
As an affiliate, we earn on qualifying purchases.
How the First Video Changed
Meyer’s account is framed as a personal report on taking an AI idea into production, with failures included. He says the short test video was approved and published the same evening the engine was built. The next morning, he produced a full video that he describes as slow and boring; he says it remains online unchanged as an example of what did not work.
His revised approach focused on frame coverage, stillness and camera movement. The figures in the November 9 update refer to the first three scenes of the rebuilt video, not the entire video or a wider sample. Meyer says the improved-look video, about the Nasdaq’s worst day in 19 months, went live a few hours after his critique. The update also says runs for 10 published Lexassay tasks are public on lexassay.com.
In the separate legal-model experiment, Meyer says a model answered all 10 published tasks after a weight edit, compared with four before, while its legal score stayed at zero. The source does not provide the benchmark’s scoring method in this account. The example illustrates why completion counts alone cannot establish that a model’s answers are correct.
“It was slow and boring.”
— Thorsten Meyer, describing the first full video
What the Report Does Not Establish
The measurements and production history are Meyer’s own account. The source does not describe an independent review of the videos, code, or measurements, nor does it state how frame fill and still time were calculated. The reported changes cover the first three scenes; results across the full video or across additional projects are not given.
It is also unclear how reliably the new gates identify poor visual choices. A check that flags empty scenes or mismatched objects may catch particular issues, but the account does not report false alarms, missed problems, or whether the system has been tested on a larger set of videos. The legal benchmark’s scoring details and the reason its score remained zero are also not explained here. Meyer’s description of the original video’s style traits does not by itself establish whether the result is sufficiently distinct from the work that inspired it.
More Runs, Human Review
Meyer says the 10 published Lexassay task runs are publicly available on lexassay.com following the November 9 update. That makes the task results available for inspection, although the source does not announce a planned independent audit or give a timetable for further testing.
For the video engine, Meyer’s account describes checks that now run before rendering, alongside human approval of a test video. He does not specify a release schedule or say how many future videos will use the system. The next useful evidence would be results across more scenes and projects, including whether the gates catch problems consistently and whether a human reviewer considers the finished videos successful.
Key Questions
What did Thorsten Meyer build?
He says he built an AI-assisted drawing engine for hand-drawn explainer videos. Claude built the engine, while OpenAI’s Codex reviewed the code in his described workflow.
What was wrong with the first full video?
Meyer says it had too much white space, icons that did not support the narration, and drawings that stopped moving after they were completed. He described it as slow and boring.
What changed in the rebuilt video?
For the first three scenes, Meyer reports that frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from three to 27. He says these measures became build checks.
Does the report prove the engine makes better videos?
No. It reports Meyer’s own measurements and judgment and does not include an independent review or results from a broad set of videos. He says human approval remains part of the process.
What does the legal-model example show?
Meyer says a model answered 10 published tasks after a weight edit, up from four, while its legal score remained zero. The example shows why a higher completion count alone does not establish correctness.
Source: ThorstenMeyerAI.com
