Iβm curious, whatβs the probability of a AI music getting to the frontpage? Is it like 1 in 100?
Trust Not The System.
β
Also known as "Questtrek".
β
I make art and occasionally games.
β
π©·ππ
β
I always try to give the best feedback possible, and am open to playtesting if you want another opinion.
β
Death to AI.
β
Feel free to PM me about anyting.
California
Joined on 4/8/21
- Level:
- 19
- Exp Points:
- 3,746 / 4,010
- Exp Rank:
- 15,372
- Vote Power:
- 6.08 votes
- Rank:
- Portal Security
- Global Rank:
- 20,310
- Blams:
- 40
- Saves:
- 457
- B/P Bonus:
- 8%
- Whistle:
- Deity
- Trophies:
- 11
- Medals:
- 128
- Supporter:
- 2m
Comments
Rarer than that. There's been about a dozen that I know of through 2024 into 2025. I don't know the exact count, and there might have been more that were not caught.
Thank you so much for contributing to the future of the audio portal too! I'm still waiting for you to get back online and help find AI shit on this audio portal too!
Thanks for sharing this, Quest! I had no clue AI is getting this convincing. If you're just casually listening to it, you may not even realize an audio track is AI-generated. We definitely need to keep an eye on this.
@Platzsoldier In 2024, almost 100 AI audio received Frontpaged, some of them probably even remained, but the audio that was Frontpaged is very much and it's not so easy to search.
Excellent use of opportunity, and yet another well written guide for the public. Awesome!
@destinyart Sometimes, AI music is downright scary for me -- especially with the grainier genres like lo-fi hip-hop, or chillhop as it's commonly referred to. There seems to be a subset of instruments it adheres to for everything else. I recommend listening to a bit of it yourself on purpose, i.e. via prompt, and getting used to how it sounds. You'll spot it immediately thereafter, with almost 100% accuracy.
Also, somehow I was not already following you. Well, fixed that!
@ADR3-N Thanks for the advice! I think that's a good idea - if I listen to the AI music carefully, I'll start to spot how it sounds when I hear it. I think it's tougher when it's ambient kind of music like the genres you mentioned. Lo-fi and stuff like that can be easy for AI to imitate, because there's not many dynamics and the song stays pretty much the same. But I think it's harder for AI to sound genuine and spontaneous in the way the human voice and human-written songs sound.
Congratulations on User of the Day!
@destinyart Oh absolutely. And not only is it hard for AI to accurately emulate the musicality of a human voice -- it's also hard for it to emulate quality, conscious mixing decisions made by a real human, since the input audio is compressed mp3s -- a lot lower quality than artists typically release.
@ADR3-N Absolutely! I like working with WAV files more myself, but if I have to, I can work with mp3s as well. There's definitely a sound quality difference, though. AI definitely can't properly mix a song in the way a human could, because that takes nuance and an ability to actually analyze what you hear. AI just knows how to imitate from all the inputs it gets from the internet, but I don't think it can truly make its own decisions in mixing a song, which is why it probably doesn't sound that good. So I agree, AI can't really compare to the human voice and human creativity.
Congratz to becoming User of the Day, Quest!
You know, when I see "I am User of the Day!"-posts I usually cheekily point out that this is a once-in-a-lifetime opportunity for them to capitalize in the extra attention they are getting and that they could do something more meaningful that just step up to the microphone and announce "I am User of the Day! That is all."
But you are _actually_ doing exactly that: Doing your part in saving the world!
And I am very proud to share a community with you! :)
Very glad to share a community with you as well (:
@destinyart You are once again 100% correct. AI does not mix well at all. It works based on isolated stems fed into it of different tracks, which as you know sound like ass to begin with. There are artefacts everywhere. A lot of times, this is what we hear.
Then too, you know from crushing mp3s from 320kbps to 190kbps or whatever, it blows and compresses the mix to absolute ass. AI music is never going to sound good with that kind of quality discrepancy, no matter what it exports at.
@ADR3-N Yes, exactly, when you look at it from that view, you see why AI music is not going to sound that good. You lose quality when you compress a mix that much, so you're not going to hear something that sounds pleasant to the ear. That's probably another sign for spotting AI - if it's working on isolated stems that already sound pretty awful, its "song" is not going to have a great sound!
For what it's worth, you can use AI to detect AI music: https://youtu.be/QVXfcIb3OKo?si=IJPGtXSF80ugpRSP
Seen this video already. I already reached out to him about both potentially contributing data and using the software. There's another AI music detection program that I recently heard about, but I found it to be more time consuming than just listening to it by ear and checking metadata, and it doesn't even work yet with Suno v4.5.
This is the other program:
https://www.submithub.com/ai-song-checker
Thanks for the 200th follow btw.
Did Benn respond about their app or was he unresponsive? Seems like really useful tech, they should make it open source.
No response. I offered to give him thousands of AI generated tracks from several different models to improve his dataset. Every track I have flagged I have also downloaded, as that is how I prove it is AI (spectrogram analysis and metadata). I asked him to open source it as well, and made a sugestion to him. I sugested that if it becomes a paid product that he decreases the price or lets Newgrounds use it for free. I am in no place to bargain on the behalf of Newgrounds, but it felt useful to at least sugest it to him. Newgrounds was the start for a lot of artists, mucisians, and animators, and in his position I would not hesitate to give back to it. But again, he never responded, so I have no idea if he wants my data or if he plans on open-sourcing it.
interesting post, although I do have to wonder, what if, in theory, someone were to recreate a suno/udio song to the highest quality, like no way of telling from audio spectrograms or file metadata that it's ai?
I don't fully understand what you mean, but AI does not produce high quality music, and it won't be for a little while. This is because most of the data on the internet is in a compressed form. Spotify, YouTube, Bandcamp, Soundcloud, etc. all compress music, because it is more affordable to stream it to people that way. AI rarely can make high quality music because it is trained on this compressed audio. Consumer-level models are never greater than the sum of their parts. Also, even at the point the quality becomes indistinguishable from music created by people, computers will still be able to pick out quirks in the spectrogram that can't be seen by the human eye. Also, people can just tell with enough exposure to it. I may use a spectrogram to prove that something is AI to others, but for myself I can tell by ear now.
i want to help!!!! (: (also are you by chance going to use my music "artificial sunset in any games?)
If you really want to help, please don't publish stolen audio. It is arguably more dangerous for Newgrounds than AI, and is such a pain for the mods to deal with. I am respectfully not going to use "At The Speed of Light but reversed, slowed, and on a single channel".
If you want to get into actual music production for free, try Cakewalk or LMMS. You can create anything in them. If you want something simpler for chiptune and whatnot, try Bosca Ceoil Blue or Beepbox. There are thousands of tutorials to get you started on YouTube. Anyone can learn.
If you would rather help hunt AI, read this guide:
https://www.newgrounds.com/wiki/help-information/site-moderation/how-to-detect-ai-audio
You can also look at the Audio Portal Cleanup thread, where me and several others post the AI tracks we find. Itβs useful as a learning tool if you want to get better at recognizing it.
@Quest what I mean is recreating a suno/udio song with vst plugins, midi, samples, etc in a DAW, not making the ai song itself high quality
From my experience, there's a lot of similarities in structure between Suno tracks, especially ones from the same genre of music, so it's possible that a recreation could still trigger AI detection algorithms and people who have gotten good at detecting AI. But I am not sure.
My question though is why would anyone do that? At that point just take inspiration from other people who will create better music than the model ever will. Honestly, the legal aspects of inspiration and sampling are not my forte, but getting inspiration from real people just sounds like a much better use of your time for the quality of the end result. AI is essentially an aggregation and average of millions of songs, so it tends to sound bland and samey.
@ADR3-N Is there anything you want to add to this?
bl00dy-seraphim617
thank you so much. ππππ