Yes, code module is polyphonic, think about it as being duplicated to match the current polyphony count in the gate signal. That's also why you have no control over individual voices from inside Code Module.
Can’t think of a reason why it would be country restricted. What you probably need is to be logged in to a google account, so they can steal your most inner thoughts. Google already knows everything about me, and I’m already logged because of my personal gmail, so I never encountered this problem
As a side note, the original giku prompt for scene generates some parts inside a code interleaved with others outside. You can still copy it all, but I wonder what causes that behavior? The shader prompt doesn’t do that. I noticed the same behavior in gemini and chatgpt. Just puzzled, no big deal
Thanks Ben, could we also bother you to share your Shader Gem text in the same manner? (I know @giku provided a shader AI contract file, but even if yours is only slightly different, it could mean huge difference in results)
I've tried it locally with open weight models gemma4 and Qwen3.6 and the hit rate was not too bad, depending on how much complexity you're asking for.
Gemma4 was better in sticking to the prompt and rules in a conservative but boring way.
Qwen3.6 was better in creative, more advanced designs but the miss rate was considerably higher. Who cares though, I had the model create 200 versions then I quickly went through them and picked the best.
Does anyone here have a recent M3/M4/M5 Macbook with 48GB or more RAM?
Pretty cool! I have been experimenting with random cv generators like this since code module is available, but your code already gave me some further ideas :) thanks for sharing
this could be done with a lot of conventional modules, and I did in the past, but it’s so much neater and convenient to have all the “mess” encapsulated in a single code module
Comments
I understand you guys not being so passionate about krell patches as I am, but I have a question:
How does increasing CV polyphony work? How does the code module process all voices? I guess it’s polyphonic?
Better
Yes, code module is polyphonic, think about it as being duplicated to match the current polyphony count in the gate signal. That's also why you have no control over individual voices from inside Code Module.
Makes sense, thanks. Was baffled at first since there isn’t even a gate, so how is the code module working polyphonically
and in hindsight I remember you mentioning it before, I think I get it now, you have no control over voices because each is a separate instance
Does your link still work? It seems to go to a default page
It does work (for me) - are you signed into Google in your browser?
I'm signed in (not your account of course) and it doesn't work here either. Is it something that cannot be pasted as plain text here?
Working fine here
I just went in and re-copied the share link. It’s set to anyone with a link. Not sure what else to try.
Weird. All I get from your link is this:
It might be restricted to U.S. only?
Still works for me in germany, just tried on an android phone.
Sorry, I thought the pic would just be embedded. Anyway not much to see, but Gemini telling it's been instructed to write code for the code module...
Working for me in the UK.
Can’t think of a reason why it would be country restricted. What you probably need is to be logged in to a google account, so they can steal your most inner thoughts. Google already knows everything about me, and I’m already logged because of my personal gmail, so I never encountered this problem
Likely you used the clip instead of the picture icon to upload. Embedding pictures works just fine, here
A little proof of concept
Thanks @pedro, I think that was it(:
I can post the text I used for the instructions, it’s nothing ground breaking.
That would be great, also for use with other models 😉
Post it as an attachment txt instead of putting it the message body. I mean just for neatness please
As a side note, the original giku prompt for scene generates some parts inside a code interleaved with others outside. You can still copy it all, but I wonder what causes that behavior? The shader prompt doesn’t do that. I noticed the same behavior in gemini and chatgpt. Just puzzled, no big deal
Thanks Ben, could we also bother you to share your Shader Gem text in the same manner? (I know @giku provided a shader AI contract file, but even if yours is only slightly different, it could mean huge difference in results)
Here's mine, in case you're interested.
I've tried it locally with open weight models gemma4 and Qwen3.6 and the hit rate was not too bad, depending on how much complexity you're asking for.
Gemma4 was better in sticking to the prompt and rules in a conservative but boring way.
Qwen3.6 was better in creative, more advanced designs but the miss rate was considerably higher. Who cares though, I had the model create 200 versions then I quickly went through them and picked the best.
Does anyone here have a recent M3/M4/M5 Macbook with 48GB or more RAM?
Of course I’m interested 🤓 will definitely try it out!
48GB or more RAM
that costs like a brazilian dollars these days
Yeah prices are crazy. But these larger Nvidia cards aren't cheap either.
i followed giku's advice and asked gemini to make a script for me you us:)
it's a random walk with levy flight.
i uploaded a demo patch.
it was error free working on first attempt.:)
has a trigger input, three parameters (maxstep, jumpchance,jumpscale), output is -1...1
press play to hear a melody, or hit a pad to make a step.
and anotherone...
it's a bytebeat synthesizer.
2 parameters:
speed
formula (8 formulas)
i uploaded a demo patch
https://patchstorage.com/bytebeat-2/
Pretty cool! I have been experimenting with random cv generators like this since code module is available, but your code already gave me some further ideas :) thanks for sharing
this could be done with a lot of conventional modules, and I did in the past, but it’s so much neater and convenient to have all the “mess” encapsulated in a single code module
hey and thanks for bringing https://en.wikipedia.org/wiki/L%C3%A9vy_flight to my attention! I’ve been doing that, didn’t know it had a name 😅
ps: I will check your bytebeat one later when I have a chance, I see we have some overlapping fringe interests 😂
hey @pedro
happy to hear you can make some use of this.
here's where i learnt about the levy flight: https://thecodingtrain.com/challenges/53-random-walker-with-vectors-and-levy-flight
the random walk, or drunk walk is fascinating me for a long time now. (back when max was midi only, pd)
it makes some nice visualisation too...
it's also an option in the lfo module lately, but i wanted to have more control.
and it seemed to be a good start into the code module and the ai stuff.
hey, here's another one i always wish to have on ios.
the chebyshev 4n bandpass digital filter from renoise native effects.
i uploaded a demopatch.
press play and check out:)
perhaps @giku may have a look at the code in terms of cpu optimisation. the ai couldnt come up with anything useful, while retaining the sound.