🔍 Read the full analysis: The Beginner’s Path From AI Idea To Production Video on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer describes building a hand-drawn AI video engine, then revising it after the first full video looked slow and empty. His account says the improved workflow uses independent code review, human approval and measurable checks; the results are a personal case study, not an independent evaluation.
Thorsten Meyer has described how he turned an idea for a hand-drawn AI video style into a production workflow, after the first full-length result proved too slow and visually static. In an account published on ThorstenMeyerAI.com and updated November 9, Meyer says the revised process combines separate AI code review, human approval and checks that can stop a render when scenes fail specified criteria.
Meyer says the initial concept came from a video whose style he admired. To avoid reproducing that work, he says he wrote down a limited set of traits—hand-drawn lines, a continuous visual world and large numbers used as color accents—and built a new engine without borrowing the original’s drawings, assets, layouts or drawing code. The project used Claude to build the engine and OpenAI’s Codex to review it, with the second system assigned to inspect code rather than create it.
According to Meyer, Codex identified five issues in its first review. He says each was fixed with a test designed to fail against the earlier code. A human review was also part of the release process: Meyer watched a roughly one-minute test video and approved it for publication. The following morning, however, the first full video in the style struck him as slow and boring. He kept that version online as an example of the shortcomings rather than replacing it.
Meyer says he then measured frame coverage and the time images remained completely still, rather than relying only on subjective tweaks. For the first three scenes of a rebuilt video, he reports that frame fill rose from 15% to 66%, still time fell from 29 seconds to zero, and camera moves increased from 3 to 27. These are figures for those scenes in his own project, not a general benchmark. He says the measurements were turned into build checks that can flag empty scenes or objects that do 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
How Quality Checks Shape AI Video
The account highlights a practical issue for creators using generative tools: a system can produce a playable video without producing one that is clear, engaging or ready to publish. Meyer’s example puts human judgment alongside automated checks, rather than treating a successful render as proof of quality. That distinction matters for teams deciding how to review AI-assisted work before it reaches an audience.
It also shows how subjective feedback can be translated into specific production criteria. Frame coverage, still-image duration and camera movement do not capture every aspect of storytelling, but Meyer’s approach makes some complaints testable and repeatable. His figures reflect one project and a small set of scenes; they do not establish that the same thresholds will suit other formats or improve audience response.
The process described has implications beyond video. Meyer’s separate legal-model experiment, included as a comparison, illustrates why one metric can give a misleading impression of progress. He says a model answered more benchmark tasks after a weight-edit experiment, while its legal score remained zero. The example reinforces his stated point that completion is not the same as correctness.
Products Worth Considering
Samsung 32 Odyssey QD-OLED G8 Gaming Monitor – 4K, 240Hz
FLEXISPOT Electric Adjustable Workbench, 48 x 30 in, Maple
Logitech MX Mechanical Wireless Illuminated Keyboard, Tactile Quiet, Graphite
As an affiliate, we earn on qualifying purchases.
From Style Reference to Original Engine
The project began with Meyer seeing a hand-drawn explainer he liked and deciding to create a video engine with some similar high-level visual qualities. In his account, the first safeguard was a clean-room description: a short list of general style traits, rather than copied artwork or layouts. He says the design used his own color system, with blue for the main image and orange to emphasize one point per scene; the handwriting font was openly licensed.
Meyer describes a workflow in which the AI tools have distinct roles. Claude wrote the engine, while Codex reviewed it and reported potential problems. He says tests were written to reproduce the bugs before fixes were accepted, so the checks would show whether they could detect the original issue. He also stresses that a person responsible for the result watched and approved the test video before publication.
The account was first published on November 6, 2026, and updated November 9. The update says the figures now refer to 10 published Lexassay tasks and that the runs are public on Lexassay.com. Meyer presents the material as a personal account based on his build notes, incident notes and experiments published on his site, not as an independent audit.
“A demo that runs is not a decision.”
— Thorsten Meyer
What the Project Figures Do Not Show
The figures are Meyer’s own measurements, and the source does not describe an outside assessment or comparison with other video-generation workflows. It also does not report audience testing, viewer retention or other evidence showing whether the revised video performed better with viewers. The reported changes cover the first three scenes, so they should not be read as measurements across every scene or every project.
It is also unclear how the quality gates are calibrated, how often they produce false alarms, or whether the checks catch problems that require narrative judgment. Frame fill and camera movement can be measured, but their presence alone does not establish that a video communicates accurately or well. The article’s legal-model example is likewise a reported experiment; the supplied account does not establish that its results generalize to other models or legal tasks.
The Next Test Is Broader Use
Meyer says the improved video look was used for a piece about the Nasdaq’s worst day in 19 months, published a few hours after he critiqued the first full version. The account says later videos use the same production method. The immediate next step described is continued use of the checks: scenes that are too empty can be stopped before rendering, while objects without a clear connection to the narration can trigger a build warning.
The source does not give a schedule for further testing or say whether the workflow will be independently evaluated. To judge how well it works beyond this example, readers would need broader results across videos, along with evidence about error rates and viewer response. For now, the published account documents one creator’s iterative process: build, review, identify a failure, measure it and add a check before the next release.
Key Questions
What is the development described in the article?
Thorsten Meyer reports building a hand-drawn AI video engine and adding code review, human approval and production checks after the first full video fell short of his expectations.
What did Meyer say changed in the rebuilt video?
For the first three scenes, Meyer reports frame fill increasing from 15% to 66%, still time dropping from 29 seconds to zero, and camera moves rising from 3 to 27. Those figures apply to the scenes he measured.
Which AI tools did the workflow use?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He describes Codex as a reviewer, not a builder, and says a human watched and approved the test video.
Are the results independently verified?
The source presents the figures as Meyer’s own project measurements and notes, not as an independent evaluation. It does not report audience testing or establish that the results apply to other video projects.
Source: ThorstenMeyerAI.com
