OBSIDIAN-Neural v1.0.0 Beta - Local TFLite models, free API keys, seeking macOS testers

[Innermost]

Update on OBSIDIAN-Neural AI sampler VST:

v1.0.0 Beta features:

• Local TFLite inference - runs offline with reduced model size (quality trade-off vs server models)
• Free API keys available for users without compatible GPUs - DM for access
• Most stable release to date, actively seeking user feedback

macOS Testing: Looking for macOS users willing to test compatibility. Build is unsigned due to code signing costs - happy to work with testers on installation.

Video Documentation: Compiling examples of live AI-human collaboration sessions. These hybrid performance formats are still emerging - interested in documenting early implementations.

Installation Note: Windows may flag the installer due to unsigned certificate - common issue with free software from independent developers. Source code fully available for review.

Free API Access: DM for complimentary API keys (50 generations each). Server runs French hours (9h-24h CET). Simply download latest VST from releases page and you're set.

The reception has been... interesting across different communities. Curious about KVR's perspective on AI-assisted sampling tools in production workflows.

GitHub: https://github.com/innermost47/ai-dj
Releases: https://github.com/innermost47/ai-dj/releases

[Lind0n]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

[Innermost]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

The code is completely open-source on GitHub. If you found "nefarious patterns", please specify which lines contain malware.

[Innermost]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Technical Reality: PyInstaller False Positives

Your "malware" concerns show a misunderstanding of how PyInstaller works. This is a well-documented issue in the Python community:

Even a simple "Hello World" compiled with PyInstaller triggers antivirus false positives - as discussed on the official Python forums.

Why this happens:

• PyInstaller bundles Python interpreter + dependencies
• Creates high-entropy executable sections (compressed bytecode)
• Runtime unpacking behavior mimics malware droppers
• Unsigned executables always trigger heuristic detection

Regarding your other claims:
- "GitHub sites blocked for port scanning" - Technically nonsensical. GitHub hosts source code, not network scanners.
- My VST uses standard HTTP requests (like your browser). This isn't "port scanning."

Code transparency: Everything is open-source. You can read every line, run the Python scripts directly, or build it yourself.

[Innermost]

Update: Taking Security Concerns Seriously

Thanks to those who raised concerns about antivirus detection - you were right to be cautious, and I appreciate the feedback.

I've submitted both executables to Microsoft's official security analysis for false positive review:



This is a known issue with PyInstaller (Python compiler) where legitimate applications trigger heuristic detection due to their packaging method. Microsoft's analysis team will verify the software is clean and update their detection signatures accordingly.

I always encourage users to be security-conscious, so thanks for keeping the community safe. The open-source code remains available for inspection, and these official reviews should resolve any remaining concerns.

Feedback like this helps improve the project for everyone.

[Innermost]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Curious that a fellow developer would confuse HTTP requests with port scanning and PyInstaller packaging with malware. As a plugin developer, feel free to point out the exact lines of "nefarious code". If you've identified specific security issues in the source code, please feel free to open a GitHub issue with the details.

[EnsoLake]

AI is the technique and aesthetic of fascism.

[Innermost]

AI is the technique and aesthetic of fascism.

You know what? You're absolutely right to be concerned. As a musician myself, I've felt that same anxiety watching tech companies steamroll creative industries with promises of "democratization'" while extracting all the value.
I'm genuinely curious about your perspective - what would you actually WANT from AI music tools if they had to exist? What would make them feel like they're serving musicians instead of replacing us? Better compensation models? Mandatory artist consent for training data? Local control instead of cloud dependence?
My plugin was my attempt to put the power back in artists' hands rather than rent it from corporations, but clearly that's not addressing the deeper issues you're raising. What needs to change from your viewpoint? Because honestly, the current trajectory where a few tech giants control everything while musicians get scraps... that IS pretty fascistic when you put it that way.

[ThomasHelzle]

That is one massive topic for a KVR post.

• All AI training data is basically stolen, nobody got asked beforehand.
  There is a massive difference between downloading a picture/track for listening to it or watching it VS. using it to build a "replication machine" for ALL pictures, videos, tracks and code ever created.
  There simply isn't any kind of copyright that could cover that properly and no artist ever gave his/her okay to being replaced by slop.
• The people crating LLMs are usually not creatives, so they mostly miss the point of art and creativity: It is not about the result or product, but about the process, the "doing it", the playing together, the joy of making something, the dancing.
• Many of the top people in the AI/corporate world are deeply traumatised individuals who were abused by humans. So now they dream of replacing humans/nature with machines and AI...
• As we can see these days, many "non creatives" carry a deep resentment for artists and now feel like they can do it themselves without investing the time and effort.
  Similar to how NFTs did introduce the "proof of work" concept in a totally backwards way - it used to be the artists work, sweat and tears that made things valuable, not wasting power on useless computations.
  These distortions matter.
• The resources and money wasted on all this AI stuff is unfathomable. You could have solved many real challenges for humanity and the planet with those resources and braincycles.
• More isn't better. Easier isn't better. Automatic isn't better.
• Working with LLMS makes people (even) more stupid.
  Teachers find themselves correcting the output of ChatGPT instead of students work.
  I see programmer-chats these days where they do not discuss how to solve a problem and write good code but how to get this or that assistant to do something.
  Instead of learning to code and getting better at that, they wrestle with assistants.
  Muscles you don't use atrophy.
• If you let an LLM create music, what is the actual point of it, for you as a human being?
  Do you learn something new? Do you get better at something?
  Do you understand more about the world, yourself, your community?

And I won't even go into hallucinations, fake news generation, social unrest, unemployment etc.

So yes, the whole underlying assumptions and ideas of the big corporations pushing this stuff into the world as if there's no tomorrow (sic) is dehumanising, trauma-driven, wasteful and making people poorer and easier to goad - we see that on every level ATM.

And when there is a power failure, how do you even open the door?

[Lind0n]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Curious that a fellow developer would confuse HTTP requests with port scanning and PyInstaller packaging with malware. As a plugin developer, feel free to point out the exact lines of "nefarious code". If you've identified specific security issues in the source code, please feel free to open a GitHub issue with the details.

Well this is what Malwarebytes says of your github references:

Malwarebytes Browser Guard Logo
Website blocked due to a risky pattern
Website blocked: https://github.com/innermost47/ai-dj
v3.0.20 | Heuristics: a risky pattern | ID: 22977144

and I didnt confuse HTTP requests with anything.

You might also want to do a quick scan of the interwebs to see if anyone else has a plugin called Obsidian, to reduce the confusion;

https://www.waproduction.com/plugins/vi ... yzoWjtd7OJ

Good luck with it.

[Lind0n]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Technical Reality: PyInstaller False Positives

Your "malware" concerns show a misunderstanding of how PyInstaller works. This is a well-documented issue in the Python community:

Even a simple "Hello World" compiled with PyInstaller triggers antivirus false positives - as discussed on the official Python forums.

Why this happens:

• PyInstaller bundles Python interpreter + dependencies
• Creates high-entropy executable sections (compressed bytecode)
• Runtime unpacking behavior mimics malware droppers
• Unsigned executables always trigger heuristic detection

Regarding your other claims:
- "GitHub sites blocked for port scanning" - Technically nonsensical. GitHub hosts source code, not network scanners.
- My VST uses standard HTTP requests (like your browser). This isn't "port scanning."

Ok well a couple of points:

1. if you knew beforehand about this PyInstaller issue - why didnt you mention it to your perspective beta testers - who in all probability are not coders? Not that all coders are using or have experience in the Python world as you seem to infer here.

2. I never said your GitHub site was port scanning I said(and I quote): "the sites listed in google are blocked for port scanning" - so a number of other sites listing in Google searches for your plugin are doing this - Github sites I can find myself without the help of Google...

[enCiphered]

That is one massive topic for a KVR post.

• All AI training data is basically stolen, nobody got asked beforehand.
  There is a massive difference between downloading a picture/track for listening to it or watching it VS. using it to build a "replication machine" for ALL pictures, videos, tracks and code ever created.
  There simply isn't any kind of copyright that could cover that properly and no artist ever gave his/her okay to being replaced by slop.
• The people crating LLMs are usually not creatives, so they mostly miss the point of art and creativity: It is not about the result or product, but about the process, the "doing it", the playing together, the joy of making something, the dancing.
• Many of the top people in the AI/corporate world are deeply traumatised individuals who were abused by humans. So now they dream of replacing humans/nature with machines and AI...
• As we can see these days, many "non creatives" carry a deep resentment for artists and now feel like they can do it themselves without investing the time and effort.
  Similar to how NFTs did introduce the "proof of work" concept in a totally backwards way - it used to be the artists work, sweat and tears that made things valuable, not wasting power on useless computations.
  These distortions matter.
• The resources and money wasted on all this AI stuff is unfathomable. You could have solved many real challenges for humanity and the planet with those resources and braincycles.
• More isn't better. Easier isn't better. Automatic isn't better.
• Working with LLMS makes people (even) more stupid.
  Teachers find themselves correcting the output of ChatGPT instead of students work.
  I see programmer-chats these days where they do not discuss how to solve a problem and write good code but how to get this or that assistant to do something.
  Instead of learning to code and getting better at that, they wrestle with assistants.
  Muscles you don't use atrophy.
• If you let an LLM create music, what is the actual point of it, for you as a human being?
  Do you learn something new? Do you get better at something?
  Do you understand more about the world, yourself, your community?

And I won't even go into hallucinations, fake news generation, social unrest, unemployment etc.

So yes, the whole underlying assumptions and ideas of the big corporations pushing this stuff into the world as if there's no tomorrow (sic) is dehumanising, trauma-driven, wasteful and making people poorer and easier to goad - we see that on every level ATM.

And when there is a power failure, how do you even open the door?

Claiming all training data is stolen oversimplifies a complex legal and technical issue. Many models are trained on public domain, licensed, or user-contributed data.
Saying developers are deeply traumatised is an ad hominem and attacks the people, not the ideas or implementations, don´t you think?
And assuming that using AI inherently makes people more stupid is a deterministic view that ignores context. The same argument was made about calculators, DAWs and even books at one point.

You're a sound designer, you know that tools don’t replace creativity. So when someone makes a blanket dismissal of AI tools without considering how real artists are using them today (imperfectly, experimentally, often critically), it’s worth questioning.

I'm not saying you are wrong to be concerned but your framing leaves no room for nuance or exploration. That’s anti-creative in its own way.

Yes, there are real ethical concerns around training data, copyright, and creative labor. And those debates need nuance, not absolutism. But reducing all AI-assisted creativity to “slop” or trauma-driven tech utopianism ignores how many artists are actively experimenting with these tools, not to replace themselves, but to expand what’s possible.
Tools have always abstracted effort. From paintbrushes to DAWs to samplers.
I’ve seen AI-assisted workflows spark genuinely original results not because the model “created” them, but because it introduced friction or randomness that the artist could sculpt. Isn’t that part of the joy too?

You ask what’s the point of using an LLM to create music. Maybe it’s the same point as using any other instrument: discovery.

[ThomasHelzle]

You've already fallen into the trap, accepted the brutal normalisation of AI and how it is pushed onto humanity.
5% "good" don't make the 95% of "highly problematic" go away.
I'm not saying that I'm right, but you asked for discussion and views.
The above is mine.

Cheers,

Tom

[Innermost]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Technical Reality: PyInstaller False Positives

Your "malware" concerns show a misunderstanding of how PyInstaller works. This is a well-documented issue in the Python community:

Even a simple "Hello World" compiled with PyInstaller triggers antivirus false positives - as discussed on the official Python forums.

Why this happens:

• PyInstaller bundles Python interpreter + dependencies
• Creates high-entropy executable sections (compressed bytecode)
• Runtime unpacking behavior mimics malware droppers
• Unsigned executables always trigger heuristic detection

Regarding your other claims:
- "GitHub sites blocked for port scanning" - Technically nonsensical. GitHub hosts source code, not network scanners.
- My VST uses standard HTTP requests (like your browser). This isn't "port scanning."

Ok well a couple of points:

1. if you knew beforehand about this PyInstaller issue - why didnt you mention it to your perspective beta testers - who in all probability are not coders? Not that all coders are using or have experience in the Python world as you seem to infer here.

2. I never said your GitHub site was port scanning I said(and I quote): "the sites listed in google are blocked for port scanning" - so a number of other sites listing in Google searches for your plugin are doing this - Github sites I can find myself without the help of Google...

Lind0n, thank you for the clarification and feedback - you're absolutely right on both points.

1. Security disclosure: You're correct that I should have mentioned the PyInstaller detection issue upfront in my original post. That's an oversight on my part - security concerns should always be addressed proactively, especially for beta software. I've now added this information prominently to the SECURITY.md and will include it in future posts.

2. Search results: My apologies for misunderstanding your comment about Google search results. You're right that third-party sites indexing the project could potentially have issues - that's outside my control but worth noting for users doing their research.

Regarding the Malwarebytes detection: This is exactly the PyInstaller heuristic issue I mentioned. The pattern detection is flagging the packaging method, not malicious code. Microsoft's security team is reviewing the executables, and users can always run the Python scripts directly to avoid any compiled binaries.

Name conflict: Good catch on the existing Obsidian plugins. The full name "OBSIDIAN-Neural" was chosen to differentiate from existing products, but I understand the potential confusion.
Your feedback as a fellow developer helps improve both the project and how I communicate about it. Beta testing is all about finding these kinds of issues - technical and otherwise.
Thanks for keeping the community vigilant about security.

Top

[Innermost]

Your GitHub sites are blocked for nefarious patterns in the code (malware) the sites listed in google are blocked for port scanning,or refuse to connect... - this all looks very very dodgy from the outside....

Technical Reality: PyInstaller False Positives

Your "malware" concerns show a misunderstanding of how PyInstaller works. This is a well-documented issue in the Python community:

Even a simple "Hello World" compiled with PyInstaller triggers antivirus false positives - as discussed on the official Python forums.

Why this happens:

• PyInstaller bundles Python interpreter + dependencies
• Creates high-entropy executable sections (compressed bytecode)
• Runtime unpacking behavior mimics malware droppers
• Unsigned executables always trigger heuristic detection

Regarding your other claims:
- "GitHub sites blocked for port scanning" - Technically nonsensical. GitHub hosts source code, not network scanners.
- My VST uses standard HTTP requests (like your browser). This isn't "port scanning."

Ok well a couple of points:

1. if you knew beforehand about this PyInstaller issue - why didnt you mention it to your perspective beta testers - who in all probability are not coders? Not that all coders are using or have experience in the Python world as you seem to infer here.

2. I never said your GitHub site was port scanning I said(and I quote): "the sites listed in google are blocked for port scanning" - so a number of other sites listing in Google searches for your plugin are doing this - Github sites I can find myself without the help of Google...

Update: I can confirm Malwarebytes is indeed blocking the GitHub repo due to PyInstaller heuristics. I'm reporting this as a false positive to Malwarebytes and have removed any third-party distribution links from the README and still working on the executable installer file.

Plot twist: If you visit the official PyInstaller GitHub repository itself, you'll find that Malwarebytes blocks it too. So we have the amusing situation where the antivirus software blocks both the tool used to create executables AND the executables it creates.

The irony is not lost on me that PyInstaller - a legitimate, widely-used Python packaging tool with 12.5k+ GitHub stars - gets the same treatment as projects using it. This highlights just how widespread these heuristic false positives have become in the Python ecosystem.