🔍 Read the full analysis: From AI Idea To First Video: A One-Evening Build 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 says he built a hand-drawn video engine in one evening and published a short test video after reviewing it himself. The first full video the next morning was slow and visually sparse; he measured the problems, changed the engine and added checks intended to stop similar flaws before future renders.
Thorsten Meyer says he built a hand-drawn video engine in one evening, but the first full video it produced the next morning was slow and visually sparse. In an account published November 6 on ThorstenMeyerAI.com, he describes revising the engine using measurements and adding checks before later videos are rendered, offering a firsthand example of the review and approval steps behind a quick AI-assisted production build.
Meyer said the project began after he watched a hand-drawn explainer with a camera moving across a whiteboard-style scene. He wanted to develop a related visual approach without copying the original. His stated rules were to use only a short list of general style traits, create the drawing engine from scratch, and use no drawings, assets or layouts from the video that inspired the idea. He said the engine’s colors came from his own design system and that its handwriting font was openly licensed.
Meyer said Claude built the engine, while OpenAI’s Codex reviewed the code without making changes. Codex reported five findings in its first review round, he wrote, and each was fixed with a test that failed against the old code. Meyer said he then watched a test video of about one minute and approved it for publication. That version appeared the same evening in an article about local votes against data centers.
The next morning’s full video exposed problems the short test had not resolved. Meyer described too much empty space, icons that did not support the narration and finished drawings that remained motionless. He measured frame coverage and still-image time, then changed the engine so shots filled more of the frame, camera movement followed each sentence, and objects continued moving slightly after they were drawn. On the first three scenes of the rebuilt video, he reported that 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 are his measurements of those scenes, not a broader performance study.
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 Critique to Build Checks
The account’s practical point is that a working demonstration and a publishable video are different milestones. Meyer’s first short test received his approval, but the following morning’s full video revealed shortcomings that required a separate critique. His process added human review and measurable production checks rather than treating successful rendering as evidence of quality.
Turning the measurements into gates could help prevent some repeat problems: Meyer says a scene that is too empty now stops before rendering, while an object that does not illustrate something in the narration triggers a warning. Those checks address specific visual issues in his workflow. They do not, by themselves, establish that a video is accurate, clear or suitable for publication; the account still places a human approval decision in the process.
The reported gains also need to be read narrowly. They describe the first three scenes of one rebuilt video, and the source does not provide the measurement method, a larger sample or results across later projects. The account is therefore most useful as a case study in how one publisher responded to a disappointing output, not as evidence that AI-generated video quality improves by a predictable amount.
Products Worth Considering
DJI Mic 3 Wireless Microphone Kit (2 Transmitters, 1 Receiver, Charging Case)
Samsung Galaxy Tab A7 Lite 8.7-Inch 32GB Wi-Fi Tablet, Gray
Sonos Era 100 SL Compact Microphone-Free Speaker, White
As an affiliate, we earn on qualifying purchases.
A One-Evening Prototype
Meyer frames the project as a personal account of moving an idea into production, with both the initial failure and the subsequent changes included. The sequence began with a style reference, a written list of visual traits and a new engine. It continued with a code review by a separate AI system, a human decision to publish a brief test, and a more detailed evaluation after the first full video appeared.
The improved-look video went live a few hours after Meyer’s critique, according to his account. The article also links to a roughly four-and-a-half-minute explainer about the build and its rules. Meyer says its narration uses an AI voice clone of himself. The source does not identify an independent reviewer or provide external verification of the production figures.
Meyer mentions a separate experiment from the same week to make a broader point about evaluating AI systems. In that test, he said a model completed 12 of 15 legal benchmark answers after a weight edit, compared with 4 beforehand, while its legal score remained zero. That result is distinct from the video project and does not establish that the video engine is reliable; it underscores the author’s warning that a better-looking metric can coexist with a failed outcome.
“It was slow and boring.”
— Thorsten Meyer, describing the first full video
What the Measurements Do Not Show
The source is Meyer’s own account, and the reported measurements come from his build notes and experiments. It does not explain in detail how frame coverage or still time were calculated, whether the same criteria were applied to the original and rebuilt scenes, or how the figures compare with other video projects. The 15% to 66% frame-coverage result applies to the first three scenes only.
It is also unclear whether the new gates have been tested across a substantial number of videos, how often they stop a render, or whether they catch problems beyond the specific visual issues Meyer described. The account does not report an external assessment of the videos’ quality or accuracy. Its claims about the code review, tests and publication decisions are attributable to Meyer, not independently established in the source material.
How the New Gates Perform
Meyer says the checks are now part of the engine’s production process, but his account does not give results from a longer run of videos using them. The next useful evidence would be whether those gates consistently flag sparse scenes or irrelevant illustrations, and whether human reviewers still find issues the automated checks miss.
The improved video is live, and the earlier full video remains online unchanged as Meyer’s example of what the new process was meant to address. He has not announced a formal evaluation schedule or a release plan for the engine. For now, the documented next step is continued use of the measurements and gates in future builds, alongside human review before publication.
Key Questions
What did Thorsten Meyer build?
He says he built an engine for producing hand-drawn explainer videos, with a continuous illustrated scene and moving camera. He describes the build as an evening project.
Why did the first full video need changes?
Meyer said it had too much white space, icons that did not support the narration and drawings that stopped moving once complete. He measured frame coverage, still time and camera moves before revising the engine.
What changed in the first three rebuilt scenes?
According to Meyer, filled-frame coverage increased from 15% to 66%, still time dropped from 29 seconds to zero, and camera moves rose from 3 to 27. Those measurements cover the first three scenes, not a general sample of videos.
Which AI systems were involved?
Meyer says Claude built the engine and OpenAI’s Codex reviewed its code. He describes Codex as a reviewer rather than a builder and says he personally approved the short test video before publication.
Does the account prove the new process works for every video?
No. The account reports one project and measurements from three scenes. It does not provide independent evaluation or results across a larger set of videos, and Meyer says human review remains part of the process.
Source: ThorstenMeyerAI.com
