TL;DR
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
Thorsten Meyer says he built a hand-drawn video engine in one evening, but its first full video was slow and visually weak. After measuring the problems and fixing them, he set review, approval and quality checks as gates for future videos.
Thorsten Meyer says he built a new hand-drawn video engine in one evening, then found its first full video slow and visually weak the next morning. In a post published November 6, 2026, he describes how a second AI reviewer, his own approval and measurements of the video’s first three scenes shaped a workflow for putting AI-assisted creative work into production.
Meyer says the style began with a video he liked, but he wanted to avoid copying its artwork or layouts. He describes a clean-room approach: recording a short list of style traits, such as hand-drawn strokes and a continuous visual world, without using the original video’s drawings, assets or layouts. Claude built the engine, according to Meyer, using his own design system and an openly licensed handwriting font.
In the same evening’s review, OpenAI’s Codex checked the code while Claude handled the build. Meyer says Codex identified five findings that were fixed, each with a test designed to fail against the earlier code. He then watched a roughly one-minute test video and approved it for publication. The short version appeared in an article about local votes against data centers, he says.
The first full video in the new look did not meet his expectations. Meyer describes too much empty space, icons that did not support the narration and images that stopped moving after they were drawn. He says measurements of the first three scenes showed the frame filled with content 15% of the time, the picture standing still for 29 seconds and three camera moves. After changes, those figures were 66% frame fill, no standing-still time and 27 camera moves. He says two checks now stop a build if a scene is too empty or its icons do not illustrate what is being said.
One Evening, One AI Video And the Rules Learned the Hard Way
Thorsten Meyer built a hand-drawn video engine in one evening. The first full video exposed weak pacing and visuals, leading to measured checks, code review and human approval.
From style idea to test
Meyer says he recorded style traits without reusing the original video’s artwork, assets or layouts.
He noted traits such as hand-drawn strokes and a continuous visual world.
Claude built the engine; OpenAI’s Codex reviewed the code and five findings were fixed.
Meyer watched a roughly one-minute test video and approved it for publication.
What changed in the first three scenes
Meyer measured the scenes before and after revising the production rules.
| Measure | Before | After |
|---|---|---|
| Frame fill | 15% | 66% |
| Still-image time | 29 seconds | Zero |
| Camera moves | 3 | 27 |
Checks before publication
The revised workflow turns visual feedback into gates while keeping a person responsible for the final decision.
- Stop scenes that are too empty.
- Check that icons illustrate the narration.
- Test fixes against earlier failures.
- Watch the finished video.
- Approve the result before publication.
- Judge clarity and engagement beyond code checks.
From Visual Critique to Build Gates
Meyer’s account shows how he turned subjective feedback about a video into specific production checks. Frame fill, still-image duration and camera moves gave him a way to describe what felt slow, while the two build gates were intended to catch empty scenes and unhelpful icons before publication.
He also draws a line between a system that runs and one that is ready to publish. In his workflow, AI review and automated tests support the process, but a human still watches and approves the result. That distinction matters to creators adopting AI tools: passing code checks does not establish that the finished work is clear or engaging.
The figures are a record of this project, not an independent evaluation of video quality or proof that the workflow will work for other creators. The post presents them as Meyer’s own build notes and experiments. Its practical value lies in the concrete process he describes: define what should be checked, test fixes against prior failures and keep a person responsible for the final decision.
Products Worth Considering
Samsung Galaxy Tab A7 Lite 8.7-Inch 32GB Wi-Fi Tablet, Gray
Samsung Galaxy Tab A11 8.7-Inch Wi-Fi Tablet, 64GB, 4GB RAM, Silver
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
As an affiliate, we earn on qualifying purchases.
How the Video Was Reworked
The timeline in Meyer’s account spans an evening and the following day. He says the initial idea came after watching a video with a style he liked. To avoid reproducing it, he wrote down style traits rather than reusing visual material. Claude then built an engine, and Codex reviewed its code. Meyer approved a one-minute test video that evening.
The next morning’s full video exposed problems that the short test had not. Meyer recorded his critique, measured the first three scenes and changed the production rules. The before-and-after figures in the post refer only to those first three scenes; they do not describe the full video or a broader sample.
Meyer uses other examples in the post to explain why he does not treat passing tests as proof of quality. He says that in one weight-edit experiment, a model completed 12 of 15 legal answers, compared with four previously, while its legal score remained at zero. He also reports a code review where all 12 tests written by the author passed but four of a reviewer’s six checks failed. These are his examples, and the post does not provide outside validation for them.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Measurements Cover
The post does not publish the full video, detailed scoring criteria or independent assessments of the before-and-after results in the supplied material. The reported 15% to 66% frame-fill change and other figures cover the first three scenes, so it is unclear how representative they are of the whole video or future projects.
It is also unclear whether the new gates have been used on multiple videos, how often they stop a build, or whether viewers found the revised video more engaging. Meyer presents the workflow as lessons from his own project; the account does not establish that the same thresholds suit other formats or production teams.
How Future Videos Will Be Checked
Meyer says the two checks for sparse scenes and icons that do not illustrate the narration are now part of the build process. His next step, as described in the post, is to use those gates on later videos while retaining code review and human approval. The account does not give a schedule for additional releases or report results from subsequent projects.
Key Questions
What happened in the project?
Thorsten Meyer says he built a hand-drawn AI video engine in one evening, approved a short test, then revised the first full video after identifying pacing and visual problems.
What changed after the first full video?
For the first three scenes, Meyer reports that frame fill rose from 15% to 66%, still-image time fell from 29 seconds to zero, and camera moves rose from three to 27.
Which AI tools did he use?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes the reviewer as checking and reporting findings rather than building the system.
Did the AI review replace human approval?
No. Meyer says he watched a roughly one-minute test video and approved it before publication. His described process keeps a human decision after the code review.
Are the reported results independently verified?
The supplied post attributes the figures to Meyer’s own build notes. It does not provide independent verification or show whether the measurements generalize beyond the first three scenes.
Source: Thorsten Meyer AI
