I’m planning an upcoming live show and want to experiment with splitting incoming tracks into stems on the fly so I can apply separate effects live.

What AI stem-separation tools (Spleeter, LALAL.AI, Demucs, etc.) support low-latency processing for live use?

Hardware specs: What GPU/CPU configurations have you used to get under 100 ms latency?

Workflow tips: How do you route stems into your DAW or mixer for minimal dropouts?

Fallbacks: Have you built any caching or buffer strategies in case of tool hiccups mid-set?

Please share your setups, command-line flags, or routing diagrams—and any gotchas you ran into during rehearsals. This will help all of us make live AI integration rock-solid!