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.
1,000
3.00I'm curious as to why someone would rate the song a "2."
I understand that some people might not enjoy this type of house music, but I envision a "1" as the type of noise someone starting out with a drum set might produce, and a "2" as a new band recording songs in their garage.
To me, it seems as if the models have in the past month passed what humans can create unassisted, so I'm interested as to how even a first take model-generated output, which this is far from, would achieve a 2 now.
@SteveSokolowski I wasn't the 2, but I don't see how anyone could say unjokingly that the models have, in the past month, passed what any human can create unassisted.
Just as ridiculous as someone rating this a 2 to you sounds, your statement sounds that way to me.
@No_uh This particular song is not something that would have been impossible to create. It was designed to be simple and have a coherent theme, which is why it scores high - music on the radio is simple and catchy.
If you haven't used multiple models together and the blend songs input feature, then I don't think you'll understand yet. I'll have to intentionally do this at some point - the music created will likely not be widely appealing, however.
The first thing the models have achieved that is past human ability is spatial effects. The things coming out of the models now play using the DTS Neural X or Dolby Surround algorithms far better than commercially produced music. Try "Ghost of You" on a system with 7+ speakers if it can run those algorithms.
@SteveSokolowski I can tell you right now I'd probably rate your ai songs higher if listening to them didn't come hand-in-hand with reading these multi paragraph pretentious ass writeups every time.
@SteveSokolowski I didn't really want s response. I was just giving you advice if you want higher scores, bud. Take it or leave it.
Regardless, the irony of this statement while the descriptions you write and long-winded bullcrap you go on about is disrespectful to musicians and traditional artists is so funny.
@Ziddletwix You should rate the quality of the song. I should make the titles of the polls more boring in the future.