Can Djay pro develop AI integration into our library? Happy to help the developers with that
I am an AI engineer. As as a small experiment, I developed a small application that will analyse my entire music library.
With that, it can give each track energy, danceability, and mode score (and key, BPM).
Using this, the app can build a playlist using different strategies (for example, ascending energy, or matching key tracks based on the Camelot wheel.
But I think the best part is using AI chat to analyse your library and suggest playlists. In an interactive chat, you can ask anything about your music (genres, top artists, etc) or help in creating playlists.
It is a simple app, based on whatâs upcoming from VDJ and other ideas I collected around.
The music Library view (you add tracks, it will analyse those)
The AI chat doesnât require compute power. The LLM (the AI model that runs your chat) can be accessed remotely. I use Gemini.
If you are worried about AI costs, there are open source models that can run on your laptop without the need for a powerful machine. (For example QWEN 4 or Google Gemma).
My experience with such applications so far has been massively underwhelming. (Lexicon/Mixed in key)
I guess theyâre ok if you only play generic electronic dance/club music made in the last decade because that is all they are designed to interpret.
Once you throw 60s, 70s, 80s Pop, Rock, Reggae, Funk, Soul, Disco in the pot these programs lose control, go all over the place and you end up regrading your tracks anyway.
Auto analysing programs could work better if you could personalise the levels and gradings to your own style and fields in which you operate.
Not everyone wants to be a club/festival DJ
In general, for Open Format / Mobile DJ styles (Weddings, Latin, Ballroom, etc.), there is still a lot of room for improvement. Itâs not just about the Key or Beat Grids (which can be tricky enough with live drumming in older tracks); the main issue lies with the âEnergyâ and âDanceabilityâ metrics.
I actually looked into tools like Mixed In Key specifically to leverage their Energy/Popularity detection, but I found that for these specific genres, the algorithms simply donât interpret the âvibeâ correctly. A numerical value often fails to reflect the real reaction on the dancefloor.
Because of this, I decided to stick to my own manual tagging method, which is the only way to be 100% sure.
That being said, I think AI integration is a fantastic suggestion and definitely the future, so you have my vote! Hopefully, a dedicated AI model could eventually learn these nuances better than standard algorithms.
I donât disagree with you all about the song analysis part.
From my own testing it sometimes works, sometimes sort of works. Tagging songs energy can be subjective and an algorithm canât simply be always correct.
The part you might be missing is the ability to understand your library better by using AI:
Chat to discover tracks buried in your folders
Chat to understand similar tracks from the same period or same genre.
Chat to get ideas for popular songs, themes, or even hidden gems
If you have not yet tried to using chatGPT/Gemini/Claude or other AI/LLMs to brainstorm on music and playlists ideas, I think you are missing out. I know (and agree) - curating a playlist (or mixing on the fly) is an art. But you might find yourself falling into the same patterns/known tracks you use over and over.
AI gives you an unfair advantage, and I think thatâs why VDJ are integrating it to their product.
The solution I experiment with is context aware meaning it has access to your library. Integrated into DJay pro, it can also have access to your playlist history and itâs a great way to get next track suggestions or playlist suggestion.
You hit the nail on the head. I completely see your point now.
In fact, I went down that rabbit hole myself: I used to extract data from Tunebat, use AI to calculate a custom âaverageâ metric based on those values, and then manually inject that into my files using Mp3Tag. I did all of that just to try to organize myself better, but as we agreed, those metrics are still subjective.
However, the âContext Awareâ aspect you mentioned is what really interests me. Iâve often wanted to export my history or a specific playlist to an external LLM (like ChatGPT) to ask for analysis or suggestions, but Djay doesnât make exporting data/playlists for this purpose very easy.
Having that integrated directlyâwhere the AI knows my library and historyâwould be a game changer. Anything that facilitates my workflow outside the club (prep and discovery) is more than welcome.
Iâm an AI skeptic, but the rate itâs developing is staggering. If this is something you tried 6 months ago, itâs old news. Something you try today could be old news next week. Someone is going to have something that works sooner rather than later.
Since joining this thread Iâve had a play with some AI playlist creation on chat gpt. Iâve thrown quite a few ideas at it such as âmake me a 3 hour nineties dance dj set keeping bpm and harmonic mixingâ etc with music from my own library and also asking for suggestions
Itâs ok but I donât find it mind blowing or game changing and it did get quite a lot of stuff wrong.
A bit like the previously mentioned energy rating etc, by the time i had tweaked and corrected everything I might as well have done it myself manually.
Maybe itâs just me.
Thanks for sharing @daniel_curley. My experience has been similar. I find this most helpful with genres, countries or venues/events that Iâm less familiar with. In those cases, itâs a big time saver.
The big question is: How do you go about exporting a playlist or folder so the AI can analyze it?
If I recall correctly, Rekordbox makes this easy by allowing you to export lists directly to a text format.
Does Djay have a similar native function that Iâm overlooking to get that text data out, or are you guys using a specific workaround (like a script or copying/pasting) to feed the list to the AI
@Albert_Maro I custom built this solution to scan through my local folders for music. Itâs not part of DJay pro.
Edit: sorry I just realized youâre asking about the other post using chatgpt.
I think chatGPT and alike wonât be as good as a dedicated solution that has access to your music library and playlist history. Having this access will provide more accurate curation of content.
I use lexicon standard subscription which lets you export a few file types. Thereâs a free version which is handy but Iâm not sure if itâll let you export
Iâve actually been testing that workflow. The main hurdle Iâm facing is that the AI models donât always interpret the raw CSV structure correctly when uploaded directly.
To fix this, I tried converting the file to XLS (Excel) to make it âcleanerâ for the AI to read. However, even when manually defining the delimiters, the columns are not separating correctly, resulting in a bit of a data mess.
Iâm going to give it another try and tweak the import settings to see if I can get a clean table, but right now it requires quite a few extra steps to get the data ready for the AI."
