This isn’t even my final form

@shah_bu_land @cloneofsimo Yes

@itschloebubble @sureailabs Once you’re at one step is more of a GAN (Gender-Agnostic Nomenclature)

@somewheresy What do you think I’ve spent the last three years doing?

Do you want to do cutting edge research on generative music production tools? Do you want to publish papers and also release open weights and code? Apply to be a research intern on our team!

@matdryhurst @dadabots @robotson @hollyplus_ Yeah we got you on this. Probably not lyrics-conditioned though. The data scale on that is tricky and voice cloning capabilities require extra considerations. The team’s working hard on better musical controllability as well, text prompts are just the start.

@cloneofsimo Yeah I used to think I didn’t need shifting when making longer audio outputs since the scale of details doesn’t change like it does with different image resolutions. But @zaptrem pointed out that with longer audio comes more repetition, hence more redundancy.

@somewheresy @killthenoise @dadabots Fun fact: I started my AI music journey by helping Kill The Noise get the Jukebox Colab working. Met dadabots that week.

@ClementDelangue @ada_rob @GoogleAI @answerdotai Please, we need better base text encoders for diffusion models

@joesparks @archi_tradition I stayed there last month. It’s huge! They put me on one of those floating floors and I had to ask to be moved to a lower floor. The fear of heights hit hard.

@RoyalCities @_lyraaaa_ @YoungPhlo_ Yeah I really recommend that latter approach. Works great!

@_lyraaaa_ @RoyalCities @YoungPhlo_ Pretty sure the PyTorch Lightning ModelCheckpoint callback in has options to do that

@killthenoise For real, I keep seeing this video and I’m like “those are the most fun people to be with in the crowd”

@dadabots @somewheresy Group effort 😄

RT @RoyalCities: 📢ATTN Producers & Musicians📢 Today is the release of the most capable FREE open source sample generator tailored for EDM…

@dadabots @_lyraaaa_ @encanti @mrbillstunes I’m down!

@unfamiliartunes @VividFeverDrms Agreed. Training and inference at scale consume a lot of energy, but the whole “each AI image is 12 gallons of water” thing is absurd. It’s like 5 seconds on a consumer GPU. If that were the problem then Call of Duty players would be boiling the world.

RT @dadabots: Prompt Jockey (n) - it's DJing but harder. With DJing you're playing usually other peoples' tracks. But with this, the tracks…

@DaveyJ_ @killthenoise HYATT

@cloneofsimo I set my EMA values one time in 2022 and I’ve never changed them since. This haunts me too.