Please rate this song.

The "music singularity" continues, with each song being used as input to refine and infer the next. Now, it is possible to use two songs as input with the latest models, so one can be an a capella vocal, and the other a stellar instrumental mix. The result is extraordinary:

Ghost of You

Released July 21, 2025

The reverb, delay, stereo effects, and dynamic range in this song easily surpass almost all commercial releases - it would be quite difficult to do this in a standard workflow - probably requiring Dolby's software for creating Atmos spatial audio and placing sounds as objects.

This is also the first time that I've been able to achieve flawless processed male vocals, which for some reason were impossible in previous models. And, it was done in half the time that previous songs required.

The model somehow understood that the word "ghost" should result in the haunting "ooooooh" vocals in the intro and backgrounds in many places. I don't understand how it would associate vocals with specific sounds when the prompt didn't call for them, but it did. Everything in this song contributes to the atmosphere perfectly - there are even three hooks (the "ghost of you" vocals, the oboe/saxophone riff, and I would argue that the reverb itself is a hook.)

After hearing this, why would anyone hire a human singer for a house track anymore? Human vocalists are now only superior in singer-songwriter settings, and only barely.

The "knee of the curve" has now been reached and these models are rapidly going to surpass unassisted human artists by the end of the year.