ah the transformer! I wasn’t checking the first link cause I thought you talk about licensing this thread under some specific license (called gpt-2). I was quite confused
had some fun with that site before by feeding in an incomplete story, cherry picking the output by hand and feeding it back again and so on. It’s the first text generator I’ve tried that could be used as a truly meaningful writer companion, very neat!
Sort of, but you’re limited by implementation details, and generally have to replace the graphics in a way that’s hard to revert without re-installing the plugin. Most in that thread are just recolors. Also, not every module author is cool with it (personally I explicitly allow it, but I don’t think anyone bothered to take up that offer yet).
File this one where you’d usually file my posts in here - in the “haha only serious” joke category
Voltage Controlled Rapper
So, GPT-2 is incredibly powerful, and GPT-3 incredibly more so, and it will definitely reach our filthy hands eventually, whether OpenAI likes it or not. I think you can already coerce GPT-2 to produce rhyme and meter if you want. You can also refine the model on an existing text corpus.
Meanwhile, rap fakes are a thing. They don’t merely imitate the voice - they imitate the flow. (note: some mildly offensive content on that link)
This tech is still very very GPU hungry. So hungry it’s not realtime yet. But it’s progressing really fast!
Anyway, duct tape the two technologies together. Add a faceplate and a few jacks to it. Select your rapper and have them freestyle on the beat about the topic of your choice for 6 hours straight.
Great! What about training an ai on patches from patch storage to generate .vcv files? It would need some hard coded patch fixer that would replace nonexistent module names and ports etc, but might be fun with gpt or even char-rnn
In Reason’s Complex-1, when you drag a patchcable from an input, its built-in oscilloscope automatically shows you what the output in question is up to. After that it goes back to showing you its main mono mix.
In this picture I’m dragging out the white noise output, the oscilloscope is on the lower-right:
It’s fantastic for beginners. A module implementing similar behavior in VCV would be great.
I’m not sure about the technical feasibility of this one, at least without abusing parts of the API we’re not really supposed to. Especially since many modules disable internal processing when they detect the outputs are not plugged in.
“Copy an existing product but make it cheaper” isn’t a unique idea, it’s just a business decision to undercut a price. This isn’t a bad thing, just not really the point of this thread.
Podolski VCO - not an original idea, just something I only recently discovered from the free Podolski synth by u-he and think it’s neat.
The VCO can be blended from triangle to saw to pointy needles and can then be mixed with a phase delayed and/or inverted version of itself, which creates many useful shapes.
and while I’m here: @contemporaryinsanity is the poly version of Vector Victor still coming? I would still love to have a poly note data looper.