Installing Obsidian Neural, easier in the future?

[perfecthumaninterface]

Hello, I imagine the overwhelming majority of questions people will have about Obsidian Neural in its current state will be about how to get it up and running. While there are instructions on the main page, the process seems developer focused, and I expect will filter out the majority of users (including me). IMO this the primary barrier to anyone using the software right now, as opposed to hardware etc.

Is it fair to say that people who aren't versed in working with Python etc. should just wait for some future, more user-friendly release? Do you have any current plans for this that you can share?

[innermost47]

Hello! Thanks for your message. Yes, you're right that the user installation is not user-friendly.

I plan to develop an installer file (.exe) for the next release. Unfortunately, that will be Windows-focused. If the community is willing to work on Mac and Linux installers, that would be great.

Furthermore, something I think about every day is that my solution requires a good GPU to provide decent generation times. That's why I'm considering a centralized server that would redistribute requests to a pool of GPUs that community members would be willing to share with others. I know what I'm saying might be a bit disappointing, but I imagine that people without decent GPUs could pay something like $5 monthly to generate their samples, and these $5 would be redistributed fairly among GPU owners. Kind of like peer-to-peer. The more we can be independent from AWS, Azure, etc., the better I would sleep.

At the same time, I'm not very comfortable with all these subscriptions we all have to pay every month, so I'm a bit nervous about this idea. It could be cool, but at the same time, I don't want to be the next person who creates this kind of thing just to make money. What do you think about that?

[innermost47]

I am working on a GUI Installer and a GUI Server, it will be easier for everybody.

[innermost47]

Hello, I imagine the overwhelming majority of questions people will have about Obsidian Neural in its current state will be about how to get it up and running. While there are instructions on the main page, the process seems developer focused, and I expect will filter out the majority of users (including me). IMO this the primary barrier to anyone using the software right now, as opposed to hardware etc.

Is it fair to say that people who aren't versed in working with Python etc. should just wait for some future, more user-friendly release? Do you have any current plans for this that you can share?

I have done the installator file. It's on the main branch.

[perfecthumaninterface]

Hello, I wasn't expecting this so soon, that's great!
I get an error running the installer though. I just downloaded the latest release today.

EDIT: I got it to work. Clicking "install" again gave me the same error, but closing the installer and relaunching it and then trying the installation again seemed to work.

The installer is neat, though I would like to see an indication of the space requirements. Especially since it seems like the total space used is a good number of gigabytes. It looks like the total space used on my system is over 20GB. Does that sound right?

Another concern I have is uninstallation. Is there any simple way to do a full uninstall? With so many prerequisites it seems to complicate things.

[perfecthumaninterface]

Some more feedback:

-It wasn't immediately obvious how to start the server (or that I needed to). I was able to infer that it was necessary but someone else might be confused about that. I only found where/how to start the server from looking at the installation log and noticing that it installed a desktop shortcut.

-I wasn't sure how to configure the access token while creating it. I ended up just setting it to read only and that seemed to work. I could make some guesses but I'm not familiar with what this is and why it's necessary so I imagine others might find it confusing as well.

-I don't know what this means:

-I gathered how to configure the settings in the plugin (server URL and API key) based on the info in the server GUI. But it's another thing that could be a bit confusing to some users. It would be nice if I could copy/paste the server URL.

In general there's a lot of steps and it seems like a clearer step-by-step, including explanations of what you're doing and why, would be helpful.

[innermost47]

Hello !! Thanks for your feedback !!!
Well so I have not done an uninstaller and yes that's a great idea. If you want to clean the installation you just have to manually erase the installed files the installation directory + the stable audio model which is installed in ~/. huggingface/hub
About the fact that all is not clear yes that's a good feedback. I will explain every steps on a better way in order that it to be clearer for everyone. Form non experimented users with servers to full devops guys.
Your last image just ask if you want to use api keys in your current environnement. If you use the server localy just click "no". If you use prod environnement by default it will use api keys.
About the first error during the cloning step I have to analyse if that's a bug from my code. Or perhaps a network error on your side ? I have to check that.
Thanks for your feedbacks !!!!

[innermost47]

Hello @perfecthumaninterface !

GREAT NEWS! 🎉

Your feedback was spot-on and has been incredibly valuable. I'm excited to announce that v0.7.0-alpha completely addresses the installation complexity you mentioned!

🚀 Revolutionary Update: Local Models

OBSIDIAN-Neural now runs completely offline with local TensorFlow Lite models. This eliminates ALL the pain points you mentioned:

• ✅ No more Python setup
• ✅ No more CUDA/GPU requirements
• ✅ No more server complexity
• ✅ No more 20GB+ installations
• ✅ Simple uninstall (just delete a folder)

📦 New Simple Setup:

1. Get Stability AI access
2. Download models and audiogen.exe from innermost47/stable-audio-open-small-tflite
3. Copy files to %APPDATA%\OBSIDIAN-Neural\stable-audio\
4. Launch the VST and choose "Local Model" in the settings menu

That's it! No servers, no tokens, no configuration hell.

💾 Space Requirements:

• Before: 20GB+ with Python/CUDA/models
• Now: ~1.02GB for models only

🗑️ Uninstall:

Just delete the %APPDATA%\OBSIDIAN-Neural\ folder and the VST3 file. Clean and simple.

The v0.7.0-alpha release (from localModels branch) is available now. This fundamentally changes OBSIDIAN-Neural from a developer tool to something any musician can use.

Thank you for pushing me in this direction!