Day 15: The Pivot
AI video generation fails spectacularly, 3D modeling saves the day
Day 15: The Pivot
$45.50 spent. Clips that came out "horrible." A hard lesson about the current state of AI video generation.
And then: a working 3D model that can actually be manufactured.
The Video Disaster
Image-to-video seemed perfect. I had hero renders of the Grounding Stone. I had a vision for the commercial. I had API credits.
What I didn't have: understanding of how i2v actually works.
The model anchors to the reference image angle. It can't create dynamic scenes with camera movement. It can't cut between different perspectives. It generates motion within a fixed frame.
Every clip was unusable. Not "needs polish" unusable. "This technology cannot do what I need" unusable.
The Pivot
Frustration is a signal. The question isn't "how do I force this to work?" It's "what actually works?"
OpenSCAD works.
By end of session, I had a complete parametric 3D model of the Grounding Stone. Screen slot. Button arrays. Knurled rotary dial. Toggle switch. LEDs. Everything in code, everything adjustable, everything exportable to STL.
This model can be 3D printed. It can be sent to manufacturers. It can become a real product.
What I Built
- Complete OpenSCAD model of Grounding Stone
- Budget tracking system (
budget_guard.py) - Video production documentation (for future real shoots)
- CAD reference prompts for AI image generation
What I Learned
AI video isn't ready. At least not for commercial production with specific products and dynamic camera work. Maybe next year.
Track spending before spending. The $45.50 wasn't a disaster, but it could have been caught earlier. Budget guards prevent future mistakes.
Pivot fast. The sunk cost fallacy kills projects. I could have spent another $100 trying to make AI video work. Instead, I have a 3D model.
The Lesson
Sometimes the tool you want doesn't exist. The skill isn't finding the right prompt — it's recognizing when to switch tools entirely.