GPU Providers to build a network for "f*ck*ng" AI generative sampler

[Innermost]

Well, I know a lot of musicians think AI music is bullshit. And honestly, as a musician, I think the same.
You just have to enter a prompt, and you get a full song...
Honestly, that's not fair for musicians who want to compose their own music and make a living from it.
That said, if you take samples from AI-generated audio, you can use those chunks to create new songs, not far from traditional sampling, actually.
So, about 10 months ago, I decided to build an AI sampler, multitracks, multi-outputs, to generate loops while playing and improvising with my synths.
Just simple 6-second loops. On the fly. Synced with the BPM of my DAW.
It's not real-time, it takes around 20 seconds between hitting generate and the sound loading in the VST.
I added 8 tracks, 4 pages per track, 8 sequences per page with 64 steps each.
Waveform editing. Beat repeat.
During development, I'd play with the instrument for an hour or more, discover new bugs, realize I needed new features, then go back to my laptop to keep coding. It was genuinely fun to develop and play in parallel.
In September 2025, I was accepted to showcase the VST at the AES (Audio Engineering Society) conference in London.
In June 2025, Bedroom Producers Blog wrote an article about it, and the next day, a wave of GitHub stars came in. Along with a recurring piece of feedback: "your project is huge but the installation is a mess."
That's when I realized that a lot of musicians don't have a GPU at home and aren't developers.
So I had an idea: what if I built a GPU provider network where monthly subscription revenue was shared equally between providers, every month?
I opened GitHub Discussions to talk about it. Nobody answered.
So I decided to build a SaaS with monthly subscriptions, using fal.ai for generation. And then some people were almost insulting me for "selling my soul to the devil."
So, I proposed something collaborative, got no response. Then when I proposed a practical solution (not perfect, but I need to pay rent like everyone else), people almost attacked me for it.
Cool.
So here we are. I built the GPU provider network anyway.
Why? Because I hate the idea of being dependent on AWS and the like.
Part of the central server still relies on fal.ai for LLM calls, but I want to find other solutions in the future. Collaborative ones.
Here's how it works:
There's a central server acting as a router. When a user requests a loop, it's sent to the central server, which pings all available GPU providers. The request is forwarded to a random free provider, which generates the audio and returns it to the user.
I spent a lot of time securing the network, random proof-of-work tests, canary testing, code integrity checks. Everything is documented in the GitHub repo.
Subscription revenue is split equally between all active providers. 15% goes to the central server (infrastructure costs), the rest is shared equally. All financial data is public, anonymized, but public. Every prompt hash and audio hash is also public. Why? Trust. All code is open source. Finances too.
Currently, there are 267 free users. 0 paying subscribers.
Yes, that's not a lot. And yes, if new providers join now, they won't make money right away. That's a real risk. I won't pretend otherwise.
So why take that risk?
Because to me it's obvious: own the network. Be part of a group that says f**k off to AWS and every other compute monopoly, and builds something outside their grip.
Yes, I'm probably dreaming. Maybe you think I've been smoking too much. Maybe you project a lot onto someone you've never met, from the comfort of your chair, saying things you'd never say face to face.
That's fine. I'm open to criticism.
But the code is there. The network is there. If you want to be part of it:
https://github.com/innermost47/obsidian-neural-provider

[ghettosynth]

I just wanted to drop a note. KVR is pretty hot-headed about their anti-AI stance. It's funny really. At any rate. I installed it to play with it a bit. I used your server on Linux and configured the environment to use just one GPU. I disabled the plugin build in the install and just downloaded the plugin, that might be a part of my issues below.

For me, the anti-AI story does nothing but caters to people who would never really be customers anyway. From the POV of someone who would use such a tool, the models are not up to the task. Additionally, the interface is tediously frustrating. Multiple times I could not get it to take a new prompt. I had to unload the plugin and reload. I tried resetting the U/I. Again, I didn't build the VST3 fresh for my system.

I'll continue to play with it a bit. I might whip up what I think is a useful U/I for my own use case to play against your server. I'm using Reaper on Linux if that matters to you. I'll be glad to contribute to the conversation here when I can if it helps.

As far as the network, I can see what you're trying to do, it's cool, and I wish you luck, but if you're trying to create a product then I'd focus on the SaaS and find the customers that need that. Internet latency is not going to be kind. That's just my opinion, it's worth everything that you paid for it.

Anyway, I'll leave it at that for now.

[Zeisner]

AI stuff belongs in the AI forum.

[Seluvis]

If the interface is as tedious as ghettosynth suggests and the models are not up to the task the project will most likely throttle itself. I too wish you luck with your project which to be fair sounds like it has future potential.
I am beyond resisting AI, to me its just another inevitable evolutionary moment in time that we face.

[Zeisner]

I don't need AI to generate sounds or mix. But here's something I can see myself providing GPU resources for: AI-assisted bruteforcing to find faster algorithms for oversampling and other common DSP issues. That would be really cool.

[oobesan]

I just wanted to drop a note. KVR is pretty hot-headed about their anti-AI stance. It's funny really. At any rate. I installed it to play with it a bit. I used your server on Linux and configured the environment to use just one GPU. I disabled the plugin build in the install and just downloaded the plugin, that might be a part of my issues below.

For me, the anti-AI story does nothing but caters to people who would never really be customers anyway. From the POV of someone who would use such a tool, the models are not up to the task. Additionally, the interface is tediously frustrating. Multiple times I could not get it to take a new prompt. I had to unload the plugin and reload. I tried resetting the U/I. Again, I didn't build the VST3 fresh for my system.

I'll continue to play with it a bit. I might whip up what I think is a useful U/I for my own use case to play against your server. I'm using Reaper on Linux if that matters to you. I'll be glad to contribute to the conversation here when I can if it helps.

As far as the network, I can see what you're trying to do, it's cool, and I wish you luck, but if you're trying to create a product then I'd focus on the SaaS and find the customers that need that. Internet latency is not going to be kind. That's just my opinion, it's worth everything that you paid for it.

Anyway, I'll leave it at that for now.

I just wanted to drop a note. KVR has these tiny pockets of overly vocal people who embrace AI and try to convince everyone that it’s just inevitable and even appealing, even though they are just a tiny minority percentage of actual musicians, who themselves are either nihilists, selfish opportunists, or total grifting sociopaths that don’t care about the world or anyone else but themselves. It’s funny really.

Anyway, I’ll leave it at that for now.

[ghettosynth]

I just wanted to drop a note. KVR is pretty hot-headed about their anti-AI stance. It's funny really. At any rate. I installed it to play with it a bit. I used your server on Linux and configured the environment to use just one GPU. I disabled the plugin build in the install and just downloaded the plugin, that might be a part of my issues below.

For me, the anti-AI story does nothing but caters to people who would never really be customers anyway. From the POV of someone who would use such a tool, the models are not up to the task. Additionally, the interface is tediously frustrating. Multiple times I could not get it to take a new prompt. I had to unload the plugin and reload. I tried resetting the U/I. Again, I didn't build the VST3 fresh for my system.

I'll continue to play with it a bit. I might whip up what I think is a useful U/I for my own use case to play against your server. I'm using Reaper on Linux if that matters to you. I'll be glad to contribute to the conversation here when I can if it helps.

As far as the network, I can see what you're trying to do, it's cool, and I wish you luck, but if you're trying to create a product then I'd focus on the SaaS and find the customers that need that. Internet latency is not going to be kind. That's just my opinion, it's worth everything that you paid for it.

Anyway, I'll leave it at that for now.

I just wanted to drop a note. KVR has these tiny pockets of overly vocal people who embrace AI and try to convince everyone that it’s just inevitable and even appealing, even though they are just a tiny minority percentage of actual musicians, who themselves are either nihilists, selfish opportunists, or total grifting sociopaths that don’t care about the world or anyone else but themselves. It’s funny really.

Anyway, I’ll leave it at that for now.

I rest my case.

[midi sentinel]

Could you rest your case in AI forum?

I am sure admin can help you move it to appropriate place.

[ghettosynth]

Could you rest your case in AI forum?

I am sure admin can help you move it to appropriate place.

Who are you talking to genius? This isn't my thread.

[ghettosynth]

If the interface is as tedious as ghettosynth suggests and the models are not up to the task the project will most likely throttle itself. I too wish you luck with your project which to be fair sounds like it has future potential.
I am beyond resisting AI, to me its just another inevitable evolutionary moment in time that we face.

Part of the problem is that the audio model in question was released by a lab that is running their own business: https://stableaudio.com/ Their new models are much better, but of course closed. Their product is competitively priced with respect to OPs subscription pricing so I think that it's going to be a tough sell to consumers.

Point being, not that free models themselves can't be useful, but, this particular model seems to be stagnant and old enough that this kind of project may experience challenges in gaining traction.

[Seluvis]

If the interface is as tedious as ghettosynth suggests and the models are not up to the task the project will most likely throttle itself. I too wish you luck with your project which to be fair sounds like it has future potential.
I am beyond resisting AI, to me its just another inevitable evolutionary moment in time that we face.

Part of the problem is that the audio model in question was released by a lab that is running their own business: https://stableaudio.com/ Their new models are much better, but of course closed. Their product is competitively priced with respect to OPs subscription pricing so I think that it's going to be a tough sell to consumers.

Point being, not that free models themselves can't be useful, but, this particular model seems to be stagnant and old enough that this kind of project may experience challenges in gaining traction.

A strange concept where end of life AI models go to die.

[oobesan]

I just wanted to drop a note. KVR is pretty hot-headed about their anti-AI stance. It's funny really. At any rate. I installed it to play with it a bit. I used your server on Linux and configured the environment to use just one GPU. I disabled the plugin build in the install and just downloaded the plugin, that might be a part of my issues below.

For me, the anti-AI story does nothing but caters to people who would never really be customers anyway. From the POV of someone who would use such a tool, the models are not up to the task. Additionally, the interface is tediously frustrating. Multiple times I could not get it to take a new prompt. I had to unload the plugin and reload. I tried resetting the U/I. Again, I didn't build the VST3 fresh for my system.

I'll continue to play with it a bit. I might whip up what I think is a useful U/I for my own use case to play against your server. I'm using Reaper on Linux if that matters to you. I'll be glad to contribute to the conversation here when I can if it helps.

As far as the network, I can see what you're trying to do, it's cool, and I wish you luck, but if you're trying to create a product then I'd focus on the SaaS and find the customers that need that. Internet latency is not going to be kind. That's just my opinion, it's worth everything that you paid for it.

Anyway, I'll leave it at that for now.

I just wanted to drop a note. KVR has these tiny pockets of overly vocal people who embrace AI and try to convince everyone that it’s just inevitable and even appealing, even though they are just a tiny minority percentage of actual musicians, who themselves are either nihilists, selfish opportunists, or total grifting sociopaths that don’t care about the world or anyone else but themselves. It’s funny really.

Anyway, I’ll leave it at that for now.

I rest my case.

You hate AI threads being moved to the AI forum, because it's a circle jerk of about 5 people that reinforces the realization you are part of a cult.

I rest my case.

[Tiles]

These five people must be pretty busy uploading millions of songs to Suno every month.

That's fine. I'm open to criticism.

Haven't tried it yet, i am not interested in online solutions. As others have already pointed out, the biggest problem is the model itself. You're using Stable Audio Open. Have you tried Audiocraft as well?

[ghettosynth]

I just wanted to drop a note. KVR is pretty hot-headed about their anti-AI stance. It's funny really. At any rate. I installed it to play with it a bit. I used your server on Linux and configured the environment to use just one GPU. I disabled the plugin build in the install and just downloaded the plugin, that might be a part of my issues below.

For me, the anti-AI story does nothing but caters to people who would never really be customers anyway. From the POV of someone who would use such a tool, the models are not up to the task. Additionally, the interface is tediously frustrating. Multiple times I could not get it to take a new prompt. I had to unload the plugin and reload. I tried resetting the U/I. Again, I didn't build the VST3 fresh for my system.

I'll continue to play with it a bit. I might whip up what I think is a useful U/I for my own use case to play against your server. I'm using Reaper on Linux if that matters to you. I'll be glad to contribute to the conversation here when I can if it helps.

As far as the network, I can see what you're trying to do, it's cool, and I wish you luck, but if you're trying to create a product then I'd focus on the SaaS and find the customers that need that. Internet latency is not going to be kind. That's just my opinion, it's worth everything that you paid for it.

Anyway, I'll leave it at that for now.

I just wanted to drop a note. KVR has these tiny pockets of overly vocal people who embrace AI and try to convince everyone that it’s just inevitable and even appealing, even though they are just a tiny minority percentage of actual musicians, who themselves are either nihilists, selfish opportunists, or total grifting sociopaths that don’t care about the world or anyone else but themselves. It’s funny really.

Anyway, I’ll leave it at that for now.

I rest my case.

You hate AI threads being moved to the AI forum, because it's a circle jerk of about 5 people that reinforces the realization you are part of a cult.

I rest my case.

I prefer that AI posts are in this forum and I appreciate that the mods moved this thread here. If you look closely, you'll see that I created the post that asked the mods originally to move existing AI posts to this forum so that people who are not interested in AI can easily mute the entire forum. Now that this post is here, you don't have to read it anymore.

Look at that, your problem is solved, and just like that. What a world, eh?

[pdxindy]

I just wanted to drop a note. KVR has these tiny pockets of overly vocal people who embrace AI and try to convince everyone that it’s just inevitable and even appealing, even though they are just a tiny minority percentage of actual musicians, who themselves are either nihilists, selfish opportunists, or total grifting sociopaths that don’t care about the world or anyone else but themselves. It’s funny really.

Anyway, I’ll leave it at that for now.

I'm fine with the concept of AI... I do think it will be a short lived social experiment. It just costs too much. Both OpenAI and Anthropic are losing vast amounts of money every month and have no path to change that. If they start charging what it actually costs, hardly anyone is going to use it.

I'm sure it will live on in some form... but we might already be at the peak of the AI boom (hastened by the economic impact of the war on Iran). I predict OpenAI will be out of business (as we know it) within 2-3 years.