What happens when you give a guy with little coding experience an AI Server? Just Another AI DAW (JAAD)

Speaker 0: Talk about a a DAW that he made, which I think is really cool that key is making music Run also, like, making this little Takes music himself. So he's, like, the user of the software RTX, like, in in the space. Speaker 1: Yeah. RTX AI, Celeste. Speaker 2: So, yeah, this is, just another AI DAW. I've seen a few out there. I didn't like any of them, so I made my own. This all A DAW? A DAW is a digital audio workstation, kind of like RTX Loops or FL Studio RTX, they have, online studios that are out there now. Speaker 0: Yeah. Speaker 2: And, anyways, if if you were here a few months ago, you probably saw my friend Zach's presentation on NewsBot. He has a Discord server that, I've been involved with for years. Run, originally, whenever AI started becoming more and more popular for, like, the, at home, you know, home, home AI server, local Fair or large language model, he was the 1st person I knew that was like, I wanna build build of those at home. And I was like, I have a bunch of old computer RTX. You can just have to start that up. Speaker 2: So I just kind of threw HINTS way. And whenever I saw what he was building with it, it just really intrigued me into building my own AI server Just like, everyone over here was discussing build their own AI servers. So with the help of, you know, my own, AI server, building out my own workflows with comfy UI and Open WebUI and all that stuff, it really had me wanting to do more and more and more. And, I found myself using Open code for my local models, but there was some stuff that AI was like, well, maybe some of these other IDEs can give me more of a Run. And I realized that when I bought this Google Pixel 10 Pro that I got a year subscription of Gemini Owner, and I was like, oh, well, I'll just take advantage of that while I have it, you know, since I got a free, year of Pro. Speaker 2: So I downloaded, Osmantic Gravity, and that's what helped me build this. I have, it running live as well right here on GitHub on my GitHub Takes. If you guys wanna go ahead and play with it yourself, by all means. AI currently have it, running Google Gemini's, Lyra on the back end that's making the music generation as well as Gemini, for the Copilot that's running on it. So this all started because when I got, to play around with, like, the, a step 1.5 workflows that Cache was building on his Discord server, You can just put in, like, a silly RTX, and it would spit out a song, like, 60 seconds Date. Speaker 2: And, they were fun to listen to, but then I found that there were higher, you know, frontier models that are out there. I'm sure some some of you have used Suno, for Using Generation. It's pretty cool. And if you pay for their monthly subscription up to $30 a month, which I thought was RTX, you know, you're looking at the possibility of playing with their studio, which is their online DAW. So, but you have to pay $30 a month to get access to that audio. Speaker 2: I thought that was outrageous. But I did like some of the functionality that they had, so I kind of adopted some of it, such as, you can highlight a certain area of a track that you, AI. And then and I will say that this is probably not the fastest computer for me to run this on. This does use, WebUI GPU, so it's gonna run faster and smoother. I'll have the computer with master GPU. Speaker 2: I'm also working on building a power RTX a performance, enhancement sharing, so it RTX removes all the extra assets that aren't necessary for visuals. But you can highlight an area. You can choose an RTX, and I will let you know that currently AI Google Gemini API key is not plugged into this because I'm using the production build that I'm sharing on Little, so it's not gonna actually generate any music for me right now. But if you guys wanna download this repo onto your home computer, set up your own dot key n v with your API key on it, by all means, that's up to you. It's you don't even have to go as far as that. Speaker 2: There's actually in my settings, you can go here and just paste your API key Fair, and it will just plug it in for you. And Qwen, yeah, you can just type in a lot like Using. You know, you can type in your style, your lyrics, and then hit generate, and it will just generate out whatever lyrics you put in it based on the amount that you highlighted. And if you just choose an actual, just a track itself and then AI it out, it'll fill out that amount of space as well. So you can also change the temperatures of certain things, while you're, and again, this is all really based off of, Google AI, so some of it's kind of buggy. Speaker 2: But my long term goal, my future goal is to have this so that people can run this locally. Right? I want it so that whenever a step gets better and there's better local models, like we said, these local models, they're roughly 9 months or so build. But, whenever they're just as good as the current RTX models and people wanna make music generate music from home, I plan on having it so that you can plug it into your, Comfy UI as the back end. Just drop your workflow JSON onto City, and then just hit run, and you don't have to worry about using a Gemini API key at all. Speaker 2: I've already got a working development build that whole how you do that, on my home computer, but I'm not sharing that yet. Currently, you can also drag and drop all of the Windows native VSTs, if you know what that means. There's they're AI, dials that you can change, different, frequencies and settings for Ariso the songs that you highlight, because it uses AI on the back end on another server and plugs into it, which is pretty cool. But yeah. And it will also use AI on the back end to Event the tempo. Speaker 2: I also have a tempo detector to find the BPM as a fallback if the AI doesn't Net give, but the AI does a pretty good job of getting it for you. This will also roll over to another project I have called Just Another AI, Audio, which will allow you to generate music there, move it over to this, and then you can edit that as well. But it's that one's more built towards, like, karaoke purposes. Built, yeah, this whole song here, you know, is I don't know if there's audio coming through. Generated with AI. Speaker 2: You can use this like a regular DAW. You can edit it as so. You can add plugins. You can JAAD this cool shortcut that allows you to delete any parts of the stems that, aren't being used, to build of clear up those frequencies there. Another cool feature that I added that I enjoyed is AI. Speaker 2: It's the music can be recognized, that it's made by AI with AI detection tools because they kind of have a custom at the RTX. It's like their watermark. Right? So when you generate a song with SuNo, it's going to have, like, an audio cutoff as you can see right here. It's ICM right at the tok/s. Speaker 2: It's gone. Like, that frequency just shows right here that it won't go any higher than that on the spectrogram, and that's intentional. That's Time a watermark that's built by Suno that says, hey. This song was made with AI, and this is how we prove it. In ASTEP and all the other generation software, it does the same thing. Speaker 2: 1 workaround I've been kind of building on that is, and you have to do System by stem right now, unfortunately, but AI City, it's a, build that I added into it that allows you to key, the AI code is like the it'll be detected up to a 30% right now on this stem that this was generated by AI. And the only reason I built this is because some of the music that I have on here, whenever you wanna share stuff with, like, Spotify or something like that, you have to go through a distributor. And some, some platforms like Spotify and Deezer, etcetera, they'll let you upload their songs even if it's AI, but some distributor, companies are like, no. You can't file, AI title's like, you can't have AI in our system. But RTX, whenever you go through a distribution, like BandLab or SoundCloud, they just send it out to all of them. Speaker 2: And if 1 of them says no, it takes it down for the rest. So what I've done is create a tool where you can kind of remove AI adding invisible frequencies you can't hear by ear to where, to, yeah, to the, the parts where it cuts off on the watermark that it kinda helps it bypass that little, frequency threading. And, it'll usually get your and if if you wanna be truthful to yourself and go to Tidal and just take it down, you can. But here's the thing 2026. Tidal will run scans on your stuff all the time. Speaker 2: They'll recognize it as AI and take it down anyways, but it just kinda gets your foot in the door to get your stuff out to those other platforms. So, but yeah. Any questions? Suck. None at all whatsoever. Speaker 2: A lot. Yeah. I actually have, a SoundCloud if you guys wanna check out some of the stuff I've made. It's, Sorry. I don't know if it's under Kayla Music using you know what? Speaker 2: It's okay. Let's go to soundcloud.com. I'm probably logged in. Yeah. So, yeah. Speaker 2: That's my profile. You can check it out. It's 100% me. I'm not from Stockholm, Sweden. It's just a BS profile I made, for my AI generated music. Speaker 2: I made a persona of a RTX DAW author, an artist, and Qdrant has RTX similar style using all the way through the album, and he also has the same vocals because you can make vocal profiles as well with Suno, which kinda helped me. So, yeah, if you wanna check that out, you can. Go ahead. Speaker 0: 1st of all, that's super cool. Yeah. I like it. DAW you so I know that there's, like, the popular services AI Pangram for, like, large language model auto-detected, like, to detect if it's generated by a large language model. It's tok/s. Speaker 0: But do you know how, like, they leverage like, they detect audio? Is it they RAG Max actual machine sharing, like, classifier model, or is it just based on, like Speaker 2: Yeah. It it's mostly based on frequencies and metadata. Right? So the majority is metadata. That's how those, detection tools work. Speaker 2: You can wipe the metadata from managing, as anyone knows. That's 1 step you can take which is 1 that I took when I uploaded my AI, for RTX. You can actually find some of, you can find the, Velvet and Chrome album from this on Spotify. It's there. It Net me get it through. Speaker 2: But Speaker 0: RTX good Date, like, kind of evading the detectors. Yeah. I'd be I guess you have a I'm sure it's, like, AI kinda native RTX, but you World also just upload regular samples and stuff too Yeah. Instead of generating Mhmm. Built you have, like, a way to cache AI make each RTX or stem or whatever you said Just Net. Speaker 2: Yeah. And the more stems that you edit that way, that you AI, like, remove some of that AI detection, the and then you, push it all out, like, you export little whole album itself RTX the whole track itself will, when it's RTX, will show a significant decrease in in ad detection as a tool. Before, I was just going into, like, different DAWs and, like, removing chunks of the song Run then mainly adding those 8 insights and stuff myself and then making it more and more and more human, and only keeping, like, the lyrics as the AI generated side of it. But, I started getting lazy and wanted something to help me AI that as well. So, yeah. Speaker 2: Go ahead. Speaker 0: Yeah. How's Time, like, insights, output stuff? Cache you add a hardware to to your System or do a like, you wanna plug it in your car, play through it? An app. Speaker 2: Yeah. Yeah. It has MIDI detection. You cache, it also has, MIDI generation as well with Lyra, which is shocking to me. I didn't dungeon could do that, until I plugged it in. Speaker 2: So, yeah, you can do everything on that end as well. If you wanna plug in your own keyboard, guitar, whatever, that'll work also. Speaker 0: The way you're using this, furloughed your own music, are you, Run going back and forth between this and, like, another, like, RTX DAW, or are you just trying to do Speaker 2: I'm trying to do everything in my own DAW at this point. Yeah. I haven't used the only other DAW that AI, Hague used on the side every now and then is BandLab. They have their own online web UI DAW as World, but that's only because they have, like, a library of, like, authentic sounds that you can use. I like to grab some stuff from that every now and then. Speaker 2: But yeah. Speaker 0: Yeah. I feel like Qwen Run showing, AI guess, is, I used to be RTX the agent. Don't research too much RTX, but 1 of the things AI when we were recording albums, it was the the master process was only like another step that I feel like was really necessary to get the music to sound crisp and sort of like radio level music rather than me just playing guitar in my garage. Literally, you wouldn't add, like, a master a, like, component to this. I mean, it's it's different because Ariso Speaker 2: there is is in the mixer tab. I actually have a, fix my mix as well as Date. Yeah. It'll actually automatically do that for you, but, there's also with the CloudVST, you can also just add and fine tune your, mastery at the end as well. Did you a free sample of the whole? Speaker 2: Not necessarily. No. Yeah. Not necessarily. No. Speaker 2: City won't do all of it. That's that's that's Net really how it works. It will, more than anything, just kind of, it it'll make the sounds sound better to your ear. Like, it'll it'll change the decibel and, frequency for some things, but, it's not gonna clear up those hard cuts that are obvious to directory that it's AI generated because it's it's a pretty solid watermark, and it was kinda hard for me to figure out how to get that workaround, and to get it to work as well. But I'm still trying to master it. Speaker 2: You know? It agent always work. So oh, both? Yeah. Yeah. Speaker 2: Both. Speaker 3: The as I haven't been through the RTX, DAW, I AI Ahmad City, but if RTX were some look at some, amount of time that you World do. Are you Just, like, sharing, sort RTX know, time AI audio this? Speaker 2: So you cache actually generate a whole using, and then I built a stem separation tool in here that will split the stems for you. And that uses AI as well, from AI, and it will split the guitar from the bass, the bass from the keyboard, etcetera. And then you World edit each part of those stems separately if you want. So just like, so if you go to Lyra, you can actually generate a full song if you want, which that j, JAS tool that I told you I was working on, the studio 1 that I'm working on, that will allow you to actually just generate a full length using, and then you can export it into this Run then split the stems and then work off of that. Yeah. Speaker 2: How much was it to build at all? AI have no idea. Yeah. Yeah. A lot of it came from a Gemini Pro subscription Fair sure. Speaker 2: A lot of it was also local hosted, like LLC 4.7 AI on open code. So yeah. 1 more. Last 1. Speaker 3: How much of this did you build with GLM Vercel Speaker 2: AI would I would say AI was honestly probably about 85% of it. The rest of it was local hosted stuff. Oh 0, no. RTX. AI was like, am I still plugged in? Speaker 2: Yeah. But, and then I would say because I haven't touched Claude code until, like, maybe a month ago, and I have recently been using Cloud RTX, AI, hard in the back end of it, because I do plan on allowing people to use, Firebase as the cloud, platform that they can store their projects, remotely so they don't have to have everything, like, on their local machine. It will store everything, on your local browser, all of your files, your assets, everything. So if you have to refresh your page, it'll automatically keep it there. But if you have to go over to a new computer or workstation, you can save everything to Firebase, and it you can pull it dungeon another location. Speaker 2: So yep. Thanks, City Boy. Oh Date,