Let’s talk about Stable Audio AI

Home Forums Commentary Let’s talk about Stable Audio AI

Viewing 12 posts - 1 through 12 (of 12 total)
  • Author
  • #43681
    Art Munson

    Audiosparx recently announced an affiliation/patnership with stableaudio.com so I had to give it a go. I looked at all the prompts and listened to all of the examples. My take away was, in some instances, it might be useful but overall it struck me as very homogeneous.

    So, I thought I would try an experiment. I took one of our more musical cues, of a very specific genre (Italian) and modified our description a bit to use as the prompt. I realize that this cue is not the Zeitgeist of the moment but I was curious. BTW, this cue has been licensed many times.

    “A campy quirky Italian flavored cue with a circus, comedic vibe. Featuring electric guitar, mandolin, accordion, bells, drums, percussion and tuba. 120 BPM key C Minor Light, Playful, Innocent, Positive, Energetic, Comedy”

    Here’s the original.

    Tipsy In Tuscany 30 Second

    Here’s the result.


    Yikes, I think I broke Stable Audio!

    As I mentioned, there are probably instances where the music created by Stable Audio could be useful. Personally, at this point, I’d rather be writing my own.

    Love to hear everyone’s thoughts.

    Art Munson

    Bumpty, bump.


    It’s just the dumbest investment and waste of time taking place in AI music space right now. The loss of time and money for any and all companies investing into generative AI Music is going to be massive. What thinking and emotional human being will want an AI track supporting their film, video, show, or ad? I tried a couple of keyworded descriptions and the music coming back at me is just a mish mosh of total garbage. There are no dynamics, there is no arrangement, there is no “feel”…I just do not understand why people are investing in this space?

    AI works for generating drum grooves, perhaps bass lines, horn sections, string section parts and lines…but wow…

    Invest in great search engines. What music users want is to type in a desription and then get great sounding “human made” compositions to consider. Even the $0 budget crowd will not download this garbage. Why invest in a garbage commodity that no one wants, ever asked for, and can not sound as good as a human made track.

    If they can not top ACID, and Garage Band….don’t bother investing into this noise. At least with the older AI programs such as ACID and garage band you get some vibe and human feel out of the arrangements that are essentially “drawn” and assembled together.


    Holy cow.Sounded like background music for those weird food eating AI videos on YouTube. Just no.


    The AI sounds a bit like a Danny Elfman score!


    Wow. It’s like something I’ve heard before, yet like nothing I’ve ever heard, but not in a good way.


    FWIW, Crucial has announced that it will “disallow the use of AI as a compositional tool for any music submitted to our catalog. Anything that is partially, substantially, or wholly composed by means of AI will be rejected”. Other libraries have made similar decisions.


    Generative AI is essentially a regurgitation machine spitting out remixes of it’s dataset re-assembled at sample level. There is no ‘thought’ or ‘intention’ behind it and there probably won’t be until someone is able to create an AGI ( artificial general intelligence ) which is still in the realms of Science Fiction.


    Haha!! ^^

    Art – did you try modifying your prompt to try to make it more similar to your popular cue? I’d like to see where it ended up after 4 or 5 revisions prompting it to be more like your cue.

    Art Munson

    @LAwriter. I did not but that’s a good thought. Maybe I’ll ask ChatGPT, or Bard, to modify it for me. 🙂 On the other hand I still think I’d rather spend my time writing music than prompts. Couple that with the fact that a growing number of libraries will not accept any music with any element of AI in it.


    Hi All, I have mentioned previously in various interviews, in our most recent communique, etc. that Stable Audio is, at least in this current initial version, labeled as “experimental” by Stability AI. Based on my own substantial exploration and use of it, it’s MUCH BETTER at creating instrument tracks with one or two instruments than it is in creating full ensemble music containing many instruments. Since it’s better at creating instrument parts, this version is more useful to composer/producers than it is to end clients looking for complete, finished songs.

    While I do not have any info about the details, I’m certain that there are changes coming in Stable Audio that will begin to resolve the shortcomings of the initial release. Stay tuned!

    Kery Michael

    Interesting… you can hear it “learning”. The algorithm plays the main theme three times, transitions to subdominant (I think), comes back to tonic. And then the rest falls completely apart.

    It’s got a long way to go. But you can hear some melody and basic song construction.

    With the amount of brain power and money being thrown at this… I don’t know. I don’t know how good it’s going to get.

Viewing 12 posts - 1 through 12 (of 12 total)
  • You must be logged in to reply to this topic.

Forgot Password?

Join Us