MusicLM: Generating Music From Text abs: project page:

introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". dataset:

new music video just landed

Insane results, It's a real shame they aren't planning on releasing the model.

If anyone was hoping for to use this for themselves -- give it up. The MusicLM paper relies on another internal Google model, MuLan, developed in Aug 2022, that also has not been released and has no open source implementation or dataset.

It’s getting pretty good. Time for me to start looking for a real job soon 🫠

This is just a Queen b side

RIP music producers. Only 5.5k samples? that is crazy.

"we have no plans to release models at this point." 🙁

That is can condition on humming/whistling is going to be huge.

Wowzer. Not available to the public, but amazing. Imagine the copyright issues.😵‍💫

Great work from Google, and great to see the release of the associated MusiCaps dataset. Not to be mistaken with MusCaps, a music captioning model (