I’ve already shared some thoughts on working with video generators and the difficulty of controlling their output through prompts. Even when you feed the generator the exact dialogue, characters often overact, introducing unwanted emotion and unnecessary gestures.

Another limitation is that the generated clips are very short — typically around five seconds — which is rarely enough to cover the timing of narration or dialogue. As a result, instead of fully staging a scene, you often end up suggesting it.

There was also an additional challenge specific to this project. The story is built around Beethoven’s Bagatelle “Für Elise,” so I had to balance two layers at once: the narrative describing the music and its effect on the characters, and the music itself, which also needs to affect the viewer.

I’m interested in whether that balance comes through — and where it breaks.