Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I installed it and everything went surprisingly fine and easily. It used about 8GB or VRAM max on a nvidia A30 and takes about 30s to generate 10s of audio. The max duration seems to be 30s in the frontend but the quality is a lot lower.

Mixing genres do not really work and the model doesn’t seem to be trained on band names. However it does perform well to create music using existing styles.

I generated some Eurovision crap and minimalist techno that were very much believable. But mixing death metal with lofi ambient isn’t the best, nor the epic progressive rock guitar solo I asked.

I think the examples on the website are cherry picked but with some experience in prompt engineering and many tentatives, it should be possible to generate great samples.

It’s also excellent at generating boards of Canada like music. The audio artéfacts, the low fidelity, the weird sounds, the detuned synths, this model does that very well and it does sound great to me.

Thanks a lot to the authors.



> But mixing death metal with lofi ambient isn’t the best

Would you expect that to be good if a human did it?


(Not the original commenter) I don't know of any death metal examples, but chill-lofi-beats+djent music already exists and it's pretty good. This isn't the only example but it's my personal favorite one:

https://youtu.be/-Yeqp2-dF5M


I don't think you can have death metal without double bass hammering. I don't think you can have lofi with double bass hammering.


Depends on the human. If it’s myself, not really. But if it’s well done, yes I would enjoy death metal sonorities in lofi ambiant once in a while. But I admit the prompt is perhaps a bit challenging.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: