
My First Experiences with AI Video: What Works—and What Doesn't
My first hands-on experiments with AI video—and what actually happens when Higgsfield, Codex, strong reference images and careful final review come together.

SERIES · 10 PARTS
How well can AI help you create your own videos? In this series of personal experiments, I document my experience with AI video — from choosing a model and suitable reference images to costs, common errors and post-production. The articles show real results and explain what worked and what didn’t.

My first hands-on experiments with AI video—and what actually happens when Higgsfield, Codex, strong reference images and careful final review come together.

Seedance, Kling, and Wan in practice: What affordable models surprisingly solve well – and why a high credit price doesn't guarantee approval.

Two equally priced Seedance productions demonstrate what reference images truly achieve—and why multiple keyframes still don't guarantee reliable direction.

The woman should have emerged from the bathroom at the end of a long hallway. Instead, she appeared from the floor and wall – turning a spatial error into a better horror idea.

Why code that looks plausible is still not C# – and why even the technically corrected version of the Operus Manifest failed to convince.

Lizzy was supposed to rap in German, but it sounded surprisingly French. Why exact pronunciation instructions aren't a guarantee – and when a mistake can become a character trait.

Two failed Lily attempts and one approved creature reveal why playback alone isn't enough and when an AI video can truly be released.

Not every mistake requires a new generation. Three real-world cases show when I use the result directly, edit it precisely, regenerate it, or discard it entirely.

A successful Seedance clip was still unusable for Instagram. The original, the Remotion attempt, and the Genjutsu correction show how far a targeted rescue can go.

Two weeks with video AI: my results with Seedance and other models, credit usage and the limitations I encountered when making AI videos.