🔍 Read the full analysis: Why I Started My AI Production Project With A Boring Video on ThorstenMeyerAI.com
Get smart everyday buys 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 describes how he turned an evening experiment in hand-drawn AI video into a production workflow after the first full video felt slow and boring. He says the team measured frame fill, still time and camera moves, then made those measures checks in the build process.
Thorsten Meyer says a first full video made with his new AI drawing engine was too slow and visually dull, prompting him to measure what was wrong and add production checks before later videos could be rendered. In an account published November 6 on ThorstenMeyerAI.com, Meyer described a workflow that combines AI code review, human approval and measurable scene requirements rather than relying on a working demo or subjective tweaks alone.
Meyer said the idea began after he watched a hand-drawn explainer and wanted to create a distinct visual approach of his own. He wrote down a short list of style traits—a hand-drawn line, one continuous visual world and large numbers used as color accents—and said the project used no drawings, assets or layouts from the video that inspired it. Claude built the engine, while a separate AI system, OpenAI’s Codex, reviewed the code. Meyer said Codex identified five issues in its first review, all of which were fixed that evening.
According to Meyer, each finding received a test designed to fail against the old code before the fix was made. He said that approach helped establish that a test could detect the original problem, rather than simply pass after the code had changed. Meyer also described watching a roughly one-minute test video and approving it for publication. That short version appeared in an article about local votes against data centers, he said.
The first full video, produced the following morning, did not meet his expectations. Meyer identified excessive white space, icons that did not support the narration and finished drawings that remained still. He said the team measured frame coverage and the duration of static images, then changed the engine. Across the first three scenes of the rebuilt video, he reported 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 figures apply to those three scenes; the account does not provide a broader sample or independent assessment.
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 Scene Metrics Became Build Gates
The account’s main production point is that Meyer did not treat a more appealing second version as enough. He said the measured targets became checks that can stop a scene before rendering: a scene judged too empty is blocked, and the build flags an object that does not illustrate something being said at that moment. That moves the response to a weak output from a one-time editorial fix toward a repeatable production rule.
The distinction matters for teams using generative tools to make publishable material. A system may produce a file and pass its own tests without producing a useful result. Meyer’s example separates technical completion from editorial approval: a human watched the test video and made the publication decision, while automated checks addressed specific code and scene requirements. His account is a description of one project, not evidence that the same thresholds will work for other formats or teams.
Meyer also says the first full video remains online unchanged as an example of what did not work. Keeping it visible makes the development easier to assess, but the article does not report external audience testing, comparative viewership or other evidence that the revised approach performed better with viewers.
Popular right now
Ninja Foodi DZ550 10-Qt DualZone Smart XL Air Fryer, 6-in-1, Black
Sony WH-1000XM4 Wireless Noise Cancelling Headphones, Black
Bose QuietComfort 45 SE Wireless Noise-Cancelling Over-Ear Headphones
As an affiliate, we earn on qualifying purchases.
From Evening Prototype to Published Video
Meyer’s timeline began with an idea in the evening, followed by a working drawing engine and a short test video. The test was approved and published that night, according to his account. The first full video arrived the next morning and was criticized by Meyer for its pacing and visual composition. He said a revised version, using the new measurements and rules, went live a few hours after that critique.
The source describes the updated visual system as using Meyer’s own design colors, with blue carrying the main picture and orange highlighting a key point. It also says the handwriting font was openly licensed. The explainer linked in the account is about four and a half minutes long and uses an AI voice clone of Meyer, as identified in the source material.
Meyer situated the video work alongside a separate experiment that week involving a local model and legal questions. He said an AI-assisted weight edit increased the number of completed answers in a legal benchmark from 4 of 15 to 12 of 15, while the model’s legal score remained zero. The account offers this as another example of why task completion alone should not be mistaken for quality; it does not provide enough benchmark detail to evaluate the test independently.
“And honestly, it was slow and boring.”
— Thorsten Meyer, describing the first full video
What the Published Metrics Do Not Show
The reported video measurements cover the first three scenes of the rebuilt video. Meyer’s account does not specify how the measurements were calculated, whether they were independently checked, or whether the same results held across later scenes or other videos. It also does not give viewer research showing whether the revised pacing or visuals were more effective for audiences.
The source does not identify the original video that inspired the style, and it does not provide a detailed technical description of the drawing engine. Meyer says the project did not reuse that video’s drawings, assets or layouts, but the account does not include an independent review of the implementation. The legal-model experiment is also described without enough benchmark information to judge how the test was constructed or how broadly its result applies.
How Future Videos Will Be Checked
Meyer says the scene measurements now operate as production gates, so later builds can be stopped when a scene is too empty or an object does not match the narration. The next step described in the account is continued use of those rules in videos made with the revised visual system. The source does not announce a release schedule, additional benchmark results or plans to publish a broader evaluation of the workflow.
For now, the clearest evidence is the project’s own before-and-after account: a short test video was approved, the first full version drew criticism, and a revised build followed after specific measurements were added. Whether those gates improve consistency across a larger body of work remains to be seen.
Key Questions
What happened with Thorsten Meyer’s AI video project?
Meyer said he built a hand-drawn video engine, published a short test and then found the first full video too slow and visually dull. He measured scene characteristics and added checks before later renders.
What changed in the rebuilt video?
For the first three scenes, Meyer reported that frame fill rose from 15% to 66%, still time dropped from 29 seconds to zero, and camera moves increased from 3 to 27. The figures describe those scenes, not a general result across all videos.
Which AI tools did Meyer say he used?
Meyer said Claude built the drawing engine and OpenAI’s Codex reviewed the code. He described Codex as a reviewer rather than a builder, with findings addressed through tests.
Did the revised video’s audience response improve?
The source does not report audience testing, viewership comparisons or other evidence of viewer response. Meyer reports changes to production measurements and says the revised video went live.
What remains uncertain about the results?
The article does not explain the measurement method in detail or establish whether the reported changes apply beyond the first three scenes. It also does not provide an independent technical review or broader evaluation of the workflow.
Source: ThorstenMeyerAI.com
