Feeding the previous frame back with an FFT-driven amount makes visual trails grow with the audio build

The feedback-bloom visual transition is triggered when rising energy calls for the image to smear or tunnel as the audio builds. The current frame is fed back into itself via .modulate(src(o0), …), where the modulation amount is driven by an FFT bin — typically low-mid (a.fft[1]) so the bloom tracks the build. On the drop, the .modulate(src(o0)…) line is dropped for a hard snap back to clean geometry. The transition creates visual tension paralleling the audio build without requiring a source change.

Examples

osc(20, 0.1, 0.8).rotate(0.1).modulate(src(o0), () => 0.01 + a.fft[1] * 0.04).out() — trails grow with low-mid energy and snap clean when the .modulate line is dropped on the drop.

Assessment

Which FFT bin is recommended for driving a feedback bloom during a build, and why? What happens to the feedback on the drop?