🔍 Read the full analysis: AI In Production: The Hard-Won Lessons Behind My First Video on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says his first full video in a new AI-generated, hand-drawn style was slow and visually dull. After measuring frame coverage, stillness and camera movement, he rebuilt the engine and made those measures production checks. The account also describes code review, human approval and a separate legal-model experiment that warns against treating task completion as proof of quality.
Thorsten Meyer says his first full video made with a new AI-assisted drawing engine was slow and visually dull, prompting him to measure the problems and add checks before production. In an account published on ThorstenMeyerAI.com and updated November 9, Meyer describes a workflow combining AI code review, human approval and production metrics rather than relying on a demo or subjective judgment alone.
Meyer says the idea began with a hand-drawn explainer video whose style he admired. To avoid copying it, he says he set a clean-room rule: record a short list of broad traits, such as hand-drawn strokes and a continuous illustrated world, while using no original drawings, assets, layouts or drawing code. Claude built the engine, according to Meyer, while OpenAI’s Codex reviewed the code without building it. He says five findings from the first review were fixed that evening, each with a test that failed against the earlier code.
A roughly one-minute test video received Meyer’s approval and went online that evening, he writes. The first full video made the next morning did not meet his expectations: he found too much white space, icons that did not support the narration and completed drawings that remained still. Meyer says that version remains online unchanged as an example of what the team wanted to improve.
Rather than adjusting the video only by feel, Meyer says the team tracked how much of the frame contained content and how long it stayed completely still. On the first three scenes of the rebuilt video, frame coverage 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 now act as gates: checks can stop scenes that are too empty, while the build can flag an object that does not illustrate the narration.
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 Production Gates
The account describes a practical distinction for teams putting AI-assisted work into production: a system completing a task does not establish that its output is good. Meyer’s video example shows how specific criticism—empty frames, unsupported icons and static images—was translated into measurements and checks that could catch similar issues before rendering.
That approach may make review more repeatable, but the figures are from Meyer’s own project, not an independent evaluation of AI video tools. They cover three scenes, and the article does not report a controlled comparison or explain how the metrics were validated. The account is useful as a description of one production process, not evidence that the same gates will improve every video or workflow.
Meyer also places a human decision in the process: a demo that runs is not, by itself, approval to publish. The combination he describes—automated checks and accountable human review—matters because technical completion can miss problems of relevance, clarity or presentation.
Products Worth Considering
Amazon Kindle Paperwhite (2024, 16 GB) – 7-inch Glare-Free Display, Black
Logitech MX Master 4 for Mac Wireless Ergonomic Mouse, White Silver
Amazon Kindle Scribe (3rd Generation), 64GB, 11-inch Display, Graphite, with Premium Pen
As an affiliate, we earn on qualifying purchases.
A Clean-Room Build and Review
Meyer presents the video as part of a broader account of how he moves AI ideas into production. He says the style was distilled into general characteristics rather than reproduced from the video that inspired it. The engine used his own design system, with blue for imagery and orange to highlight a key point, and an openly licensed handwriting font, according to the post.
His described division of labor gives Claude the implementation role and Codex a separate review role. Meyer says the reviewer’s findings were paired with tests written to fail on the old code. That order, he argues, helps show that a test can detect the original problem rather than merely pass after a change. The article does not provide an independent audit of the code or the review process.
The post also cites an unrelated legal-model experiment from the same week. Meyer says a model answered all 10 published Lexassay legal tasks after a weight-edit experiment, up from four, while its legal score remained zero. In the November 9 update, he says the figures refer to those 10 published tasks and that the runs are public on lexassay.com. The result is his report of an experiment, and the unchanged score illustrates why task completion alone can be a misleading measure.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Measurements Do Not Show
The post reports changes in three scenes, but it does not say whether those measurements were independently checked, how the thresholds were chosen or whether viewers found the rebuilt video more engaging. It also does not provide a broader sample of videos or comparisons with other production methods. The figures should be read as project-specific observations, not general benchmarks.
Meyer says the improved-look video went live a few hours after his critique and identifies it as a video about the Nasdaq’s worst day in 19 months. The post does not provide enough information in the supplied account to independently establish that market comparison. The legal experiment is also distinct from the video work; its reported task count and score do not establish performance on legal work beyond the published benchmark.
The account is a first-person report, and no outside reviewer or viewer feedback is included. It remains unclear how often the new gates reject scenes, whether they catch problems beyond the measured ones, and how much human review each production requires.
How the Workflow Will Be Tested
Meyer says the video engine now uses the measured checks during production, so the next step described in the post is applying the same gates to later scenes and videos. The article does not set out a release schedule, a planned independent evaluation or a target for improving viewer response.
The published Lexassay runs provide a separate way to inspect the legal-task experiment, according to Meyer’s update. More evidence about both efforts would require results across additional videos or benchmark tasks, along with clearer reporting on review outcomes and failures. For now, the confirmed development is a change in one creator’s workflow: measure visible problems, make checks repeatable, and retain human approval.
Key Questions
What did Meyer publish?
He published a first-person account of building and revising an AI-assisted engine for hand-drawn explainer videos, including the production checks he says now prevent some earlier problems.
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 when complete. He says that version remains online unchanged.
What changed in the rebuilt video?
For the first three scenes, Meyer reports that frame coverage rose from 15% to 66%, stillness fell from 29 seconds to zero, and camera moves rose from three to 27. He says measurements became checks in the production process.
Does the report prove the workflow improves AI videos generally?
No. The figures describe Meyer’s own project and three scenes. The post does not report an independent evaluation, viewer study or comparison across a larger set of videos.
What is the legal-model experiment mentioned in the post?
Meyer says a model answered 10 published Lexassay tasks after a weight-edit experiment, compared with four before, while its legal score remained zero. He says the runs are public; the experiment is separate from the video engine.
Source: ThorstenMeyerAI.com
