Testing the First AI Shorts Production Workflow with NotebookLM and Google Vids
Summary
Drafted a shorts script with NotebookLM, generated AI video clips with Google Vids/Veo, and added subtitles and narration to validate the first end-to-end shorts production workflow. The goal was not monetization — it was finding out whether a repeatable shorts production SOP is achievable.
Background
In the May operating plan, shorts were positioned not as a full monetization channel but as a secondary experiment to validate the production workflow first. The goal was simple: run through the full cycle at least once — from idea to script to video to upload copy.
Defining NotebookLM as the Shorts Production Workspace
The first step was defining Google’s AI research and organization tool, NotebookLM, as a dedicated workspace for shorts production. The notebook was named “Shorts Production SOP Lab | NotebookLM” and loaded with May/June operating standards, a 30-second shorts script template, a script generation prompt, a blog conversion prompt, and a performance log form.
Initially it was tempting to treat NotebookLM as just a script generator, but in practice it became closer to a knowledge base that organized the entire production process. Once operating standards, script templates, and post-production review forms were all loaded in, NotebookLM functioned less like an AI chat window and more like a workspace for building a repeatable SOP.
First Experiment Subject: Haereujil Lantern
The subject for the first experiment was a haereujil (nighttime tidal flat foraging) lantern. The initial script was built around product features and precautions, but during review the promotional language and exaggerated phrasing were trimmed and the format was shifted to a “pre-purchase checklist.” The final message became informational: “Before heading to the night sea for haereujil, check your lantern’s brightness, runtime, and battery condition.”
Video Generation with Google Vids/Veo
Video was generated using Google Vids/Veo. The first attempt was a 16:9 landscape video, and the night-sea atmosphere and headlamp inspection scenes came out more realistic than expected. A 9:16 portrait version was generated afterward to match the shorts format, but while it fit the format, the realism was noticeably lower than the 16:9 version.
This surfaced an important judgment call. Rather than always generating in 9:16 from the start, it may be more effective in some cases to generate naturally in 16:9 and then crop to portrait. Scenes with high visual information — a night sea, people, products, hand movements — tend to produce more natural framing in landscape mode.
Subtitle Direction
Subtitles were also reviewed. The initial instinct was to fit as much narration as possible into the text, but for shorts the right move was to show only the key checkpoints rather than long explanations. The final subtitle structure was short and explicit: “Check 1. Brightness,” “Check 2. Runtime,” “Check 3. Battery condition.”
Upload Channel Decision
The upload channel was also a real decision. One option was to post it as the first video on the ShoppingNotes channel. But the core of this video was not a product recommendation — it was an AI shorts production experiment. Posting it to Chulbuji Official as an “AI shorts production experiment” was the more coherent choice.
What This Experiment Confirmed
First, NotebookLM can be used not just for writing scripts but for organizing production standards and building an SOP.
Second, Google Vids/Veo is viable for generating AI video clips for shorts, but realism varies depending on aspect ratio.
Third, shorts production is less about making a perfect video in one pass and more about running through the full cycle — script, video, subtitles, upload, and review — as quickly as possible.
This experiment was not about producing a high-quality short. The point was that the content production routine was validated all the way through using AI tools. The next steps are to refine the shorts production SOP through May, then separately assess the viability of running a shopping shorts monetization channel in June.
Next Actions
- Log the production time and blockers from this experiment in the performance record form.
- Compare the 16:9-generate-then-crop approach with the 9:16 direct generation approach and reflect the result in the SOP.
- In the next shorts, maintain the “pre-purchase checklist” format rather than product recommendation to verify repeatability.
Related Insight: What Matters in AI Shorts Production Is Not the Video — It’s the Repeatable Routine