From LaurieWired.
Lo-fi girl’s got nothing on this.
I built a tool that dynamically creates Infinite focus music that adapts to what’s on your screen.
By combining a few locally-run Open-Source AI models (MagentaRT + InternVL3), it’s trivial to have your own contextual, endless radio station.
Plus, I also go over some music model glitch token behavior; with absolutely terrifying results.
—
Here’s the Github to try it yourself:
https://github.com/LaurieWired/InfiniteRadio
Docker Container:
https://hub.docker.com/r/lauriewired/musicbeats
—
Timestamps:
00:00 Introduction + Model Architecture
04:03 Code + Live MagentaRT Demo
06:30 Contextual Interpretation (Phase 1)
08:38 Adding Image Classification (InternVL3)
11:28 Dynamic Music Generation Demo
14:49 Centroid Glitch Tokens (Spooky!)
—
Follow LaurieWired on Social Media:
►https://linktr.ee/lauriewired
—
Papers Cited:
MagentaRT:
https://magenta.withgoogle.com/magenta-realtime
InternVL3:
https://arxiv.org/pdf/2504.10479
CoCa:
https://arxiv.org/pdf/2205.01917
Disclaimer: This is a personal project not affiliated with my employer.


