Robomuffin Radio — Stations Made by Machines (and Me)
Robomuffin Radio isn’t just a collection of stations — it’s a living, automated music factory.
Instead of manually curating every playlist or hand-picking every track, I’ve built a system that can take a single idea and turn it into an entire station: albums, songs, artwork, tags, everything.
This is more than streaming music. It’s an experiment in creative automation, AI music generation, and hands-off station building — with some human taste sprinkled in when I want it.
Why Robomuffin Radio Is Different
So you may be asking yourself: Why are a bunch of radio stations on a site unique or special?
Well, like most things, it’s in the creation. You can, of course, do these things by hand all day long and make the same results manually — but that was not the case here.
I’ve made music my entire life with technology and have kept up with it from the first DAWs to the incredible AI-driven tools and models we have now. I wanted to engineer a process where I could feed in my concept — in as little or as much detail as I wanted — and have it create an entire radio station’s worth of content, labeled, tagged, and ready to add to a station. This is essentially how stations are made at Robomuffin Radio, unless it’s something I choose to curate myself.
The Automation Engine
It’s all driven by an automation process via the n8n software I have running on a VPS. You can also run it locally if you’re curious. If you haven’t explored n8n and you’re into automation, I highly recommend investigating it — there’s a lot of power here.
Essentially, I give it a JSON file containing my ideas. I typically use a local or SaaS LLM to take my ideas from the JSON, analyze them based on my SYSTEM and USER prompts, and then generate albums, songs, styles of music, etc., placing them into a database.
Depending on the situation, I usually do this one album at a time. In some cases, I disconnect that part of the workflow from my song-generation sub-workflow so I can prebuild the database first and then have it generate many albums for many stations in a single run.
Handling the Limitations
Given the lack of truly usable and functional APIs from most solid music models, I run a more complicated macro workflow via a series of VMs. This took a lot of trial and error to perfect — especially to account for errors, since macro usage can be messy and inconsistent compared to API calls.
My little “farm” of VMs works through any database entries not marked “Completed,” then sends me a Discord message when finished (along with any errors along the way).
Branding and Graphics
Once the music is done, I use a simple GPT-4o prompt to generate the main station background and station logo. Yes, this could be automated too, but I don’t want to put all my eggs in that basket. Wan and Qwen are huge competitors right now, and if I can help it, I’d rather not lean too heavily on Big AI. As the tech grows and open-source models get better, I like the idea of running everything locally.
You’d be amazed at what you can do locally with just a decent NVIDIA card with a good amount of VRAM.
Storytelling and Local Models
I may do a walkthrough video of the process as I revisit it, as I’m currently reworking my old Worldbuilding / Story Narration creator (originally built in Python) into n8n so it’s easier to tweak and eventually deploy.
That’s the tool I used for the initial Robomuffin Radio Narration / Stories station — all narrations and backing tracks were done with local models. From worldbuilding and story writing to series/season/episode images, everything was created locally on a couple of refurbished gaming machines.
A Word of Caution
The reason I don’t make my workflows for this project (or even my Worldbuilder application) public is because of the potential for harm. These systems can take simple, creative concepts and produce polished, engaging, even faith-based content — but they could also be used for ill purposes, like hate propaganda or misinformation.
For me, this was a fun project built with good intentions, but I always consider more than just my own use case. I want to ensure I’m not making it easier for someone to do something they’ll regret.