I haven’t tried the sequencer yet, so can’t help much, but…isn’t it doing some slower frame-by-frame export at high quality? The visualizers are made for real-time use, at “normal” simulation frame-rates.
A possible path is that we can bake the analysis of samples, instead of doing it real-time. Again something I haven’t tried yet, but should be able to store a dataset with freq+magnitude values, which we can query with timestamps or similar.
Now instead of sequencer failing to get realtime audio, we’d have it query the dataset every frame, and it will get back the correct value.
Would probably also have to decouple the current smoothing setup from frame-time, and replace it with something that works step by step in the way sequencer runs logic.