🔍 Read the full analysis: Why My First AI Video Was Boring—and Why That Was Useful on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer says his first full video from a new hand-drawn animation engine was slow and boring, despite a successful one-minute test. He measured composition and motion, rebuilt the engine, and turned those measures into checks that can stop unsuitable scenes before rendering. The account is a personal report, and its results come from the creator’s own notes and experiments.
Thorsten Meyer says the first full video made with his new AI-assisted, hand-drawn animation engine was slow and boring, even though a shorter test had impressed him. In a report published November 6, he describes measuring the problems, rebuilding the engine and adding checks intended to stop empty or poorly matched scenes from reaching a final render.
Meyer says the idea began with a hand-drawn explainer whose style he liked. To avoid copying it, he says he wrote down only broad visual traits: drawn strokes, a continuous illustrated world and large numbers used as accents. He reports that the team used no drawings, assets or layouts from the video that inspired the idea. The new engine was built from scratch, with colours from his own design system and an openly licensed handwriting font.
According to Meyer, Claude built the engine and OpenAI’s Codex reviewed its code without building it. Codex identified five issues in the initial review; Meyer says each was fixed with a test designed to fail against the old code. He also says he watched and approved a roughly one-minute test video before it went live that evening. The following morning, however, the first full video had too much white space, illustrations that did not support the narration, and long stretches of still imagery.
Meyer says he measured frame coverage and still-image duration rather than relying only on subjective tweaking. In the first three scenes of the rebuilt video, he reports that filled-frame coverage rose from 15% to 66%, stillness fell from 29 seconds to zero, and camera moves increased from three to 27. He says those measures became build gates: scenes that are too empty can be stopped before rendering, while objects that do not illustrate the narration can trigger a warning.
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 Visual Checks Changed the Build
The account highlights a practical distinction for people using generative tools: a functioning demo is not the same as a finished piece that works for viewers. Meyer’s first short test looked promising to him, but the longer video exposed pacing and composition problems that were not apparent in the demonstration.
Turning those observations into measurable checks could make revisions more repeatable than relying on taste alone. The reported figures are specific to three scenes in Meyer’s project, not a general benchmark for AI video tools. Still, the approach offers a concrete example of treating visual quality as part of production review, rather than assuming that a successful render is ready to publish.
Popular right now
Breville Bambino Espresso Machine, Brushed Stainless Steel
Breville BES870XL Barista Express Espresso Machine
Dyson Big Ball Multi Floor Canister Vacuum, Yellow
As an affiliate, we earn on qualifying purchases.
From Evening Prototype to Published Video
Meyer describes the work as a personal account of taking an AI idea into production. The short test went live within an article about local votes against data centers, he says. The first full video, which he describes as boring, remains online unchanged as an example of the problem. A later video using the revised approach went live a few hours after he wrote his critique, according to his account.
The article also connects the video work to a separate experiment involving a local model and legal questions. Meyer says a model completed 12 of 15 answers after a weight-editing experiment, compared with four beforehand, but its legal score remained zero. He presents that result as another warning that improved task completion does not by itself show improved performance. These are the author’s own project results; the supplied material does not establish an independent evaluation.
“And honestly, it was slow and boring.”
— Thorsten Meyer, describing the first full video
Limits of the Reported Results
The measurements are Meyer’s own results for the first three scenes of one video. The report does not provide an independent assessment, a full account of how frame coverage was calculated, or evidence that the same gates improve videos across other projects. It also does not say how often the checks produce false alarms or miss a problem.
The supplied material identifies the tools used and describes the review process, but it does not include a separate response from Anthropic or OpenAI about their roles. The legal-model experiment is also presented without enough benchmark detail to assess the test’s scope or generalizability. The account supports conclusions about Meyer’s workflow, not broad claims about AI video quality or legal-model capability.
Whether the Gates Hold Up
Meyer says the checks now stop scenes that are too empty and flag objects that do not illustrate what is being said. The next test is whether those gates continue to catch problems in later videos without blocking acceptable work. The report does not specify a schedule for further testing or provide results across a larger collection of videos.
For readers following the project, the most useful next evidence would be additional examples, clear measurement methods and results showing how often the gates catch issues before publication. Until then, the reported improvements remain a documented account of one creator’s build and revision process.
Key Questions
What was wrong with the first full AI video?
Meyer says it had too much white space, illustrations that did not support the narration and long stretches where the image remained still.
What changed in the revised video?
For the first three scenes, Meyer reports that filled-frame coverage rose from 15% to 66%, stillness fell from 29 seconds to zero, and camera moves increased from three to 27.
How was the AI engine reviewed?
Meyer says Claude built the engine and OpenAI’s Codex reviewed the code. He reports that five review findings were fixed and paired with tests that failed on the earlier code.
Do these results prove the method works for all AI videos?
No. The figures come from Meyer’s own project and cover the first three scenes of one video. The supplied account does not include independent testing or results across other projects.
Source: ThorstenMeyerAI.com
