I know you hate it - They hate us too

[wagtunes]

Well, if I do, I never set it and I have no idea how to change it.

This is really basic stuff man. Just make sure that the loop/cycle button on the transport is off. If it's on the playback on the main arrangement follows the left/right locator markers. So if the markers cover a bar, then only one bar will loop and the plugin will follow that playback (if that's how the plugin was designed).

Okay, I know what you're talking about now. I never use that thing so it's not on. So I have no idea why it loops back after 1 bar.

[Innermost]

@audiojunkie
Speaking about data centers this project is such revolting...
https://www.dezeen.com/2026/05/15/strat ... sler-utah/
When I saw the problems that it will cause to the environment and that people who finance that kind of project don't care about, I feel such anger.

Will data centers efficiency improved? Yep, possible, and on the other there's the mass adoption effect...
When I think about the fact that some of them want to build AI in order to help climate crisis resolution, I am feeling so acid (don't know if we say that in English, but that's how I feel in my french brain) when I think that people have just to stop taking car alone in their car and eating less meat to already fix very very quickly a part of the problem. For sure it's not all of the problem, but when something is easy to do we prefer building something extremly complexe for in the end having an AI who will give us for advices to drive less with our cars and eating less meat... But we will listen to it because that's the GOD Machine who spoke...
Acidic illarity

[planet_b]

So, if I had a nice GPU on a local server, and I'd like to run the full stack of the models serving strictly for the local use with the VST, can I do that?

As I remember reading earlier about the GPU provider network, but does running your server become part of the network (which I do not want)? And the Github folder for the server talks about that, with the obsidian-token etc.

[Innermost]

So, if I had a nice GPU on a local server, and I'd like to run the full stack of the models serving strictly for the local use with the VST, can I do that?

As I remember reading earlier about the GPU provider network, but does running your server become part of the network (which I do not want)? And the Github folder for the server talks about that, with the obsidian-token etc.

Hey!
So for the open source side of the project what you can do is:
- downloading the VST at https://github.com/innermost47/ai-dj
- exploring all the repos that I list on the README and building your own alternative server. Everything is here. I think you could even use the code of the Obsidian central plus the Obsidian provider, but that's a lot of only for the central.
Honestly?
Get the VST, analyze the src/engines/DjIaClient.cpp file and create a quick fastapi server that makes the role of central server between the VST and the Obsidian Provider package.
Check into old version if the code too, you will find the ancient python server. Check on the version before end of march 2026, on the main repo.
I will not code that, as I am alone on this project that's too much repo to maintain in the end.
So yep, feel free to do what you want with the source code, just have to respect the AGPL3.0 licence.
If you need any help, any questions, I am here on open an issue on Github.

[Innermost]

One thing about AI that is driving me crazy:
I saw your comments where you told me that I add an attitude that you didn't like, and for me, that's ok, because you are human (well... Mmm... I hope ) and we are equal (well I hope too ).
But if I speak with Claude and I express an opinion, it's sometimes going to give me some advices I wasn't seeking. You'll tell me that's like with every human. So... Yes, you would be right. But, firstly Claude is not human, and furthermore, Claude is the property of a prívate company, who has financial interests, and through its model, it's giving me some advices on what I should say or not, what I should think or not, etc.
Today, many of these AI prívate companies have the power to influence so many people in their direct life, the way they educate their child as much as they drive their business, the way they handle crisis with their wife or husband as much as what they should do with their own life. Remember people who suicide themselves after having too much speaking with chatgpt.
I don't know any companies with such a power of influence in so many fields of a human life. And you?

[zerocrossing]

In my view, AI is simply a new, powerful tool that will continue to improve. Like any tool, it can be used for bad or for good. The potential for bad is as strong as the potential for good, and like any new powerful tool, it can have a great deal of effect on whatever it is being used for. The point is, that useful tools don’t go away. They get used, as long as they are useful, or until something better replaces them.

So, assuming AI isn’t going to go away (which I believe will be the case), what can be done to mitigate the problems that people see with the technology? I would argue that this question is already being worked on.

They say necessity is the mother of invention. Already, science is working on ways to make data centers have less impact on the environment. It is very possible that innovations in environmentalism can come about because of these existing data centers. It is very possible that new conservation technologies can be developed from this. In fact, I would argue that in 10 years, there will be a whole new way that data centers run compared to how they run now and I bet they will be more convenient and more ecological and more environmentally friendly. I would also argue that they will also run much more efficiently than data centers of today and all of this will come about because of the concerns and complaints of the people about data centers. I work IT. Data centers have existed for decades and decades, and they don’t run very efficiently. However, attention is now on that very problem and there are new technologies and new ways of doing things that are being introduced. Even these days in the end, it is very possible that things could end up better than they were before because of the concerns people have with these data centers. Sure, AI data centers are huge, but don’t be naïve, data centers exist everywhere. In every hospital and every large business that I know of there is a data center running every place. An industry has a large set of computers being used, there is a data center. there are hundreds of thousands of data centers in the world that have absolutely nothing to do with AI. However, all of these data centers are going to benefit from the attention, AI data centers are putting eye on an issue that affects all data centers. After this, even the ones that aren’t for AI, will benefit.

Well again, the term “AI” gets bandied about a lot to describe a lot of different applications of what’s what we used to call “machine learning.” What we need to think about is cost vs. value. A Gemini search costs like 10x that of a regular Google search. Are you getting value from it? I find it often gives me incorrect information, so after I look at it, I’m back looking at the regular results, so that extra money was spent for nothing. The resources used, wasted. The Gemini response, has no value over the traditional search results.

The same holds true for generative AI music. If I’m going to use a prompt to create a song that I didn’t really write or produce, why no just listen to a song that’s already out there, ready to be streamed, or maybe even in my physical media library? (I’m old!) It’s probably a better song, and uses a lot less resources. I’ve already heard first person accounts of companies that tried to use agentic AI to replace employees having problems as prices increased and hallucinations and odd behavior created as much work as they replaced, so now they were paying for the agent and the person minding the agent, effectively increasing their costs. Meanwhile that data center that’s running those agents is increasing the costs of power for the community it’s near. I just don’t see how that’s sustainable, and a lot of it is betting on “General AI,” which I think is either a red herring to drive their stock prices up, or a complete misunderstanding of what actual intelligence is doing.

AI is the colloquial term the media and the masses use, but that wasn’t my point. My only point in this is that maybe some good might come from the complaints and concerns about the environment that people have with these data centers. Current data centers are inefficient—they will be improved, which will benefit all data centers, not just the ones being used for “AI”.


EDIT: Here’s just one of dozens of examples of the media calling it “AI”, which so much easier for me to hunt-and-peck type on my phone screen:

https://m.slashdot.org/story/456178

Well, that remains to be seen. We could just as easily see a lot of abandoned data centers as a blight on the landscape. I come from a town that was basically poisoned by Chevron and Dow Chemical, and when they were done, they just left and the government had to come in, much later, and clean up. I'm sure parts of it will never be right.

[audiojunkie]

@audiojunkie
Speaking about data centers this project is such revolting...
https://www.dezeen.com/2026/05/15/strat ... sler-utah/
When I saw the problems that it will cause to the environment and that people who finance that kind of project don't care about, I feel such anger.

Will data centers efficiency improved? Yep, possible, and on the other there's the mass adoption effect...
When I think about the fact that some of them want to build AI in order to help climate crisis resolution, I am feeling so acid (don't know if we say that in English, but that's how I feel in my french brain) when I think that people have just to stop taking car alone in their car and eating less meat to already fix very very quickly a part of the problem. For sure it's not all of the problem, but when something is easy to do we prefer building something extremly complexe for in the end having an AI who will give us for advices to drive less with our cars and eating less meat... But we will listen to it because that's the GOD Machine who spoke...
Acidic illarity

Yep. That has NOT been a popular project here. The officials that pushed that through without a fight just got voted out. And the project has been force-shrunk to a quarter of the size that it had originally been planned to be. In addition, there have been forced-commitments that the project must make in order to go forward. The entire project is being reshaped due to the power of the people. This is about a 45 minute drive from where I live.

[audiojunkie]

So, if I had a nice GPU on a local server, and I'd like to run the full stack of the models serving strictly for the local use with the VST, can I do that?

As I remember reading earlier about the GPU provider network, but does running your server become part of the network (which I do not want)? And the Github folder for the server talks about that, with the obsidian-token etc.

You could run a 7B or 13B model locally on consumer machines. The newest 5090 series is supposedly capable of running a 30B parameter model. Very large open models can use up to 400B. Frontier closed models’ numbers (GPT, Claude, Gemini) aren’t released, but they are estimated to be in the trillions of parameters or higher, rather than just in the Billions. So don’t expect as much from local models as you would from what you have access to commercially.

[audiojunkie]

In my view, AI is simply a new, powerful tool that will continue to improve. Like any tool, it can be used for bad or for good. The potential for bad is as strong as the potential for good, and like any new powerful tool, it can have a great deal of effect on whatever it is being used for. The point is, that useful tools don’t go away. They get used, as long as they are useful, or until something better replaces them.

So, assuming AI isn’t going to go away (which I believe will be the case), what can be done to mitigate the problems that people see with the technology? I would argue that this question is already being worked on.

They say necessity is the mother of invention. Already, science is working on ways to make data centers have less impact on the environment. It is very possible that innovations in environmentalism can come about because of these existing data centers. It is very possible that new conservation technologies can be developed from this. In fact, I would argue that in 10 years, there will be a whole new way that data centers run compared to how they run now and I bet they will be more convenient and more ecological and more environmentally friendly. I would also argue that they will also run much more efficiently than data centers of today and all of this will come about because of the concerns and complaints of the people about data centers. I work IT. Data centers have existed for decades and decades, and they don’t run very efficiently. However, attention is now on that very problem and there are new technologies and new ways of doing things that are being introduced. Even these days in the end, it is very possible that things could end up better than they were before because of the concerns people have with these data centers. Sure, AI data centers are huge, but don’t be naïve, data centers exist everywhere. In every hospital and every large business that I know of there is a data center running every place. An industry has a large set of computers being used, there is a data center. there are hundreds of thousands of data centers in the world that have absolutely nothing to do with AI. However, all of these data centers are going to benefit from the attention, AI data centers are putting eye on an issue that affects all data centers. After this, even the ones that aren’t for AI, will benefit.

Well again, the term “AI” gets bandied about a lot to describe a lot of different applications of what’s what we used to call “machine learning.” What we need to think about is cost vs. value. A Gemini search costs like 10x that of a regular Google search. Are you getting value from it? I find it often gives me incorrect information, so after I look at it, I’m back looking at the regular results, so that extra money was spent for nothing. The resources used, wasted. The Gemini response, has no value over the traditional search results.

The same holds true for generative AI music. If I’m going to use a prompt to create a song that I didn’t really write or produce, why no just listen to a song that’s already out there, ready to be streamed, or maybe even in my physical media library? (I’m old!) It’s probably a better song, and uses a lot less resources. I’ve already heard first person accounts of companies that tried to use agentic AI to replace employees having problems as prices increased and hallucinations and odd behavior created as much work as they replaced, so now they were paying for the agent and the person minding the agent, effectively increasing their costs. Meanwhile that data center that’s running those agents is increasing the costs of power for the community it’s near. I just don’t see how that’s sustainable, and a lot of it is betting on “General AI,” which I think is either a red herring to drive their stock prices up, or a complete misunderstanding of what actual intelligence is doing.

AI is the colloquial term the media and the masses use, but that wasn’t my point. My only point in this is that maybe some good might come from the complaints and concerns about the environment that people have with these data centers. Current data centers are inefficient—they will be improved, which will benefit all data centers, not just the ones being used for “AI”.


EDIT: Here’s just one of dozens of examples of the media calling it “AI”, which so much easier for me to hunt-and-peck type on my phone screen:

https://m.slashdot.org/story/456178

Well, that remains to be seen. We could just as easily see a lot of abandoned data centers as a blight on the landscape. I come from a town that was basically poisoned by Chevron and Dow Chemical, and when they were done, they just left and the government had to come in, much later, and clean up. I'm sure parts of it will never be right.

Indeed, we could. Especially if we don’t collectively keep the pressure on to force change and improvement. But I’m trying to focus on the positives.

[planet_b]

So, if I had a nice GPU on a local server, and I'd like to run the full stack of the models serving strictly for the local use with the VST, can I do that?

As I remember reading earlier about the GPU provider network, but does running your server become part of the network (which I do not want)? And the Github folder for the server talks about that, with the obsidian-token etc.

You could run a 7B or 13B model locally on consumer machines. The newest 5090 series is supposedly capable of running a 30B parameter model. Very large open models can use up to 400B. Frontier closed models’ numbers (GPT, Claude, Gemini) aren’t released, but they are estimated to be in the trillions of parameters or higher, rather than just in the Billions. So don’t expect as much from local models as you would from what you have access to commercially.

Well, from the gen image side I'm aware that the open weight models are very much capable. I have a 5090 so eventually I'll be trying out the full model stack. But as the setup doesn't seem so straightforward now, it'll take a while to figure it out first..

[Innermost]

@audiojunkie
Speaking about data centers this project is such revolting...
https://www.dezeen.com/2026/05/15/strat ... sler-utah/
When I saw the problems that it will cause to the environment and that people who finance that kind of project don't care about, I feel such anger.

Will data centers efficiency improved? Yep, possible, and on the other there's the mass adoption effect...
When I think about the fact that some of them want to build AI in order to help climate crisis resolution, I am feeling so acid (don't know if we say that in English, but that's how I feel in my french brain) when I think that people have just to stop taking car alone in their car and eating less meat to already fix very very quickly a part of the problem. For sure it's not all of the problem, but when something is easy to do we prefer building something extremly complexe for in the end having an AI who will give us for advices to drive less with our cars and eating less meat... But we will listen to it because that's the GOD Machine who spoke...
Acidic illarity

Yep. That has NOT been a popular project here. The officials that pushed that through without a fight just got voted out. And the project has been force-shrunk to a quarter of the size that it had originally been planned to be. In addition, there have been forced-commitments that the project must make in order to go forward. The entire project is being reshaped due to the power of the people. This is about a 45 minute drive from where I live.

Thanks for your message! I am very happy to read that the project has been reshaped due to the power of people! What is a very good news in fact. That shows that protesting has a real impact.

[Innermost]

So, if I had a nice GPU on a local server, and I'd like to run the full stack of the models serving strictly for the local use with the VST, can I do that?

As I remember reading earlier about the GPU provider network, but does running your server become part of the network (which I do not want)? And the Github folder for the server talks about that, with the obsidian-token etc.

You could run a 7B or 13B model locally on consumer machines. The newest 5090 series is supposedly capable of running a 30B parameter model. Very large open models can use up to 400B. Frontier closed models’ numbers (GPT, Claude, Gemini) aren’t released, but they are estimated to be in the trillions of parameters or higher, rather than just in the Billions. So don’t expect as much from local models as you would from what you have access to commercially.

You're right about the models, and in the same time, for the OBSIDIAN Neural project you only (well there's an llm but it's optionnal and using a big LLM would be over engeneer) need stable audio models, which are something like less than 2b models. If you run Stable Audio 3 Medium on a GPU like you have, you will generate sounds in less than 5 seconds. I would say 2 seconds. Check at their Github and try their model quickly on your GPU just to make a benchmark on your device, and you'll see that I will be very quick to generate a sound.