Kinetic type plus AI footage: make the WORDS obey the physics of the scene behind them
Lyric videos usually fail because the type is a separate layer floating over unrelated footage. Two worlds, no conversation. The fix that leveled mine up: make the typography live IN the generated scene, obeying its light, physics, and depth.
Concretely: when the lyric is over a rainy street shot, the words get wet and reflect the neon. When it's over a windy field, the letters bend and shed like the grass. I pull a depth map from the AI shot and parent the type to it so words sit BEHIND foreground elements and in front of background ones. They inhabit the space instead of hovering on glass.
The text should feel discovered in the scene, not pasted on it. Light it with the scene's key, grade it with the scene's LUT, let it cast a shadow. Suddenly it's one world. Before/after in a reply.