U-he Zebra 3 (Alpha)

[TechHaus]

There are so many good Hive 2 skins though... some that explicitly reverse the mirroring, and even implement completely different workflows...

You are not wrong. However, skins can lose compatibility when software updates are done, and I dislike feeling reliant upon 3rd party skins. I'd rather rely on skins directly supported in the app itself. Do you REALLY like the mirrored placement of components? Or do you, like me, have to stop and think every time you go from working on one oscillator to the other to remember where the settings are? It throws off my thinking every time. Even though it is just for a few seconds, it feels like an unnecessarily "forced" change, just to fit within the shape of the gui, rather than an optimization of the workflow.

[TechHaus]

Deleted. urs you deserve better than this.

[audiouser720]

Lil Zebralette 3 sounds incredible. I can’t wait for the full stripy horse to drop. I think u-he should take feature requests lightly at this stage. I’d much rather they focus on the trademark quality than try to cram everything into it, ’cos bro cannot play keys without an arpeggio.

Speaking of which, I do wonder what’s next for u-he after Z3 and the well deserved Xmas break. Diva 2, I imagine? With an MSEG and improved internal oversampling perhaps. I’ve always wished u-he did a sampler, or more like a drum sampler. Not to compete with Kontakt but a character sampler. A proper drum synth sampler with an intuitive interface and some properly modelled vintage flavour is really missing from the plugin world since the Geist fiasco. It’s also the only thing missing from u-he’s repertoire really, just sayin’.
But anyway, let’s keep on topic and celebrate the arrival of Zebra 3, which will reach us before GTA 6 and 3I/ATLAS

EDIT: Zebralette 3 has the most beautiful and pleasing GUI I’ve seen in a long while. So there

[daily patcher]

with LLMs I'm having a field day doing sound design with my personal "scripter" generating endless uhm wavetables.

it has certainly given Hive a new lease on life for me, it has kind of a dream of mine to create a natural language input for uhm scripts and the tools are finally good enough to where I don't need much spare time to actually get something workable going.

You know, I tend to take u-He synths for granted. Today I finished a song and decided to put a splash of Hive on it, which admittedly I hadn't used in a while. Well, it was just the spice the track needed.

Getting Zebra 3 will be my best Christmas present this year.

To me Hive still has the most pleasing UI and satisfying workflow of all synths.. still my gold standard. And yes not to mention the sound and different filters available are fantastic. I wish uhm scripting was more popular, though now with LLMs I'm having a field day doing sound design with my personal "scripter" generating endless uhm wavetables.

Wait!! What?!! We must have more information!!!

I, personally, am not smart enough to benefit from the UHM scripting. I suck at understanding the intense mathematical formulas that I’ve seen tossed around when talking about it. Are you saying that LLMs are now advanced enough to feed it wav files and it can spit out the needed formulas, or even the whole UHM script??

What engine are you using? What is your prompting process? Could you post an example of one of your prompts?

I try different solutions every few weeks or so, things are always changing quickly when it comes to AI agents so the best approach really is just to build something multiple times and see what approach does it best to your liking this particular month. In this case, my favorite solution to date has been using Google AI Studio.

I never wanted to feed it wav files for uhm scripts so much as give it natural language to describe the sounds I want. However that would be a very interesting avenue to explore next once I get tired of this current approach. Maybe even add an audio preview option as well.

I'm right there with you though I don't really have any experience in DSP coding so my math leaves a lot to be desired, but I've always had an interest in uhm scripting as a means for expanding the sound sources for Hive.

If there is interest I can see about making this something to share with others, it took me such a small investment of time though I realize reality anyone could probably use AI Studio to build something similar if they spend enough time figuring out what they want to do at the outset and guiding the process.

I have a long list of instructions I have refined I give as the app's core instruction set (I don't know how much of this is necessary now, it has grown over time as LLMs have failed to generate usable scripts so starting from scratch may be better if you are just getting started). But basically as long as you are guiding it as things fail, it can learn enough from the scripts themselves, the manual, internet resources (mostly KVR as the uhm dataset if were being honest) and from there develop a good understanding of creating working uhm scripts.

A picture says a thousand words, here is my favorite iteration of the tool so far (the prompts with "Add Complexity" get a lot more interesting than the example I chose from one of the simple prompt ideas):

[IvyBirds]

Glad to see that things are moving along, can't wait for Zebra 3

[Fannon]

I try different solutions every few weeks or so, things are always changing quickly when it comes to AI agents so the best approach really is just to build something multiple times and see what approach does it best to your liking this particular month. In this case, my favorite solution to date has been using Google AI Studio.

I never wanted to feed it wav files for uhm scripts so much as give it natural language to describe the sounds I want. However that would be a very interesting avenue to explore next once I get tired of this current approach. Maybe even add an audio preview option as well.

I'm right there with you though I don't really have any experience in DSP coding so my math leaves a lot to be desired, but I've always had an interest in uhm scripting as a means for expanding the sound sources for Hive.

If there is interest I can see about making this something to share with others, it took me such a small investment of time though I realize reality anyone could probably use AI Studio to build something similar if they spend enough time figuring out what they want to do at the outset and guiding the process.

I have a long list of instructions I have refined I give as the app's core instruction set (I don't know how much of this is necessary now, it has grown over time as LLMs have failed to generate usable scripts so starting from scratch may be better if you are just getting started). But basically as long as you are guiding it as things fail, it can learn enough from the scripts themselves, the manual, internet resources (mostly KVR as the uhm dataset if were being honest) and from there develop a good understanding of creating working uhm scripts.

A picture says a thousand words, here is my favorite iteration of the tool so far (the prompts with "Add Complexity" get a lot more interesting than the example I chose from one of the simple prompt ideas):

Nice! This looks really useful and if Zebra 3 can also import the scripts, this might actually be a good way to create waveforms in natural language. Have you thought about creating this as a Skill (https://code.claude.com/docs/en/skills)? If it's just prompt engineering, it might be sufficient to publish this as a markdown file and then you can ask any LLM to learn the skill (via link) and then use it. Or is your solution more involved than that?

[audiojunkie]

with LLMs I'm having a field day doing sound design with my personal "scripter" generating endless uhm wavetables.

it has certainly given Hive a new lease on life for me, it has kind of a dream of mine to create a natural language input for uhm scripts and the tools are finally good enough to where I don't need much spare time to actually get something workable going.

You know, I tend to take u-He synths for granted. Today I finished a song and decided to put a splash of Hive on it, which admittedly I hadn't used in a while. Well, it was just the spice the track needed.

Getting Zebra 3 will be my best Christmas present this year.

To me Hive still has the most pleasing UI and satisfying workflow of all synths.. still my gold standard. And yes not to mention the sound and different filters available are fantastic. I wish uhm scripting was more popular, though now with LLMs I'm having a field day doing sound design with my personal "scripter" generating endless uhm wavetables.

Wait!! What?!! We must have more information!!!

I, personally, am not smart enough to benefit from the UHM scripting. I suck at understanding the intense mathematical formulas that I’ve seen tossed around when talking about it. Are you saying that LLMs are now advanced enough to feed it wav files and it can spit out the needed formulas, or even the whole UHM script??

What engine are you using? What is your prompting process? Could you post an example of one of your prompts?

I try different solutions every few weeks or so, things are always changing quickly when it comes to AI agents so the best approach really is just to build something multiple times and see what approach does it best to your liking this particular month. In this case, my favorite solution to date has been using Google AI Studio.

I never wanted to feed it wav files for uhm scripts so much as give it natural language to describe the sounds I want. However that would be a very interesting avenue to explore next once I get tired of this current approach. Maybe even add an audio preview option as well.

I'm right there with you though I don't really have any experience in DSP coding so my math leaves a lot to be desired, but I've always had an interest in uhm scripting as a means for expanding the sound sources for Hive.

If there is interest I can see about making this something to share with others, it took me such a small investment of time though I realize reality anyone could probably use AI Studio to build something similar if they spend enough time figuring out what they want to do at the outset and guiding the process.

I have a long list of instructions I have refined I give as the app's core instruction set (I don't know how much of this is necessary now, it has grown over time as LLMs have failed to generate usable scripts so starting from scratch may be better if you are just getting started). But basically as long as you are guiding it as things fail, it can learn enough from the scripts themselves, the manual, internet resources (mostly KVR as the uhm dataset if were being honest) and from there develop a good understanding of creating working uhm scripts.

A picture says a thousand words, here is my favorite iteration of the tool so far (the prompts with "Add Complexity" get a lot more interesting than the example I chose from one of the simple prompt ideas):

Very cool!! How did you teach it UHM scripting? That's the biggest thing that has me curious.

[Urs]

You can teach it uhm scripting by uploading the uhm language guide... I guess...?

[Urs]

Btw. one of my goals for Zebra 3 release is to add extra functionality to use the spline waves of Zebra 3 in uhm, e.g. in Hive.

And maybe there'll be a curve fitting tool, too, where a spline curve can be extracted from a wavetable.

In case it was missed, copy/paste of waveforms etc. between Z3 oscillators and/or MSEGs is not just .svg but also .uhm.

[daily patcher]

Btw. one of my goals for Zebra 3 release is to add extra functionality to use the spline waves of Zebra 3 in uhm, e.g. in Hive.

And maybe there'll be a curve fitting tool, too, where a spline curve can be extracted from a wavetable.

In case it was missed, copy/paste of waveforms etc. between Z3 oscillators and/or MSEGs is not just .svg but also .uhm.

That's amazing, I didn't know what was still on the table as far as uhm support goes but I'm very much looking forward to this.

Very cool!! How did you teach it UHM scripting? That's the biggest thing that has me curious.

As Urs mentioned, using the guide as a reference material is hugely helpful. I also took my entire folder of wavetables, pared it down to just uhm scripts, and consolidated to one .md file as learning material as well. Those coupled with access to kvr / internet resources and guiding the LLM if it fails can get you far.

Nice! This looks really useful and if Zebra 3 can also import the scripts, this might actually be a good way to create waveforms in natural language. Have you thought about creating this as a Skill (https://code.claude.com/docs/en/skills)? If it's just prompt engineering, it might be sufficient to publish this as a markdown file and then you can ask any LLM to learn the skill (via link) and then use it. Or is your solution more involved than that?

That's not a bad idea, next time I play around with Claude I may explore this. In theory most if not all of this could be published as a markdown file as that is basically the instruction set I've been carrying over each time I build an uhm generator.

[uOpt]

You can teach it uhm scripting by uploading the uhm language guide... I guess...?

You might try something like notebooklm.google.com

It is a LLM where you dump some PDFs in and it answers questions from those documents. I have good success with Google's version. However, it is hopeless when drawings or diagrams appear in the documents.

[audiojunkie]

Btw. one of my goals for Zebra 3 release is to add extra functionality to use the spline waves of Zebra 3 in uhm, e.g. in Hive.

And maybe there'll be a curve fitting tool, too, where a spline curve can be extracted from a wavetable.

In case it was missed, copy/paste of waveforms etc. between Z3 oscillators and/or MSEGs is not just .svg but also .uhm.

Cool!

[zerocrossing]

To me Hive still has the most pleasing UI and satisfying workflow of all synths..

I sold Hive 2 because I just couldn't get on with the UI. The mirroring of controls was inconsistent and I could never get a quick at-a-glance sense of where everything was or what was going on with a patch. It was decent in every other respect though.

I agree. Basing a layout on a synth name, or vise versa, is what we call "too cute by half," in my world. Let's just have a nice clean logical layout. Great sounding synth, though, but I'd definitely turn to it more if it had a logical layout.

[zerocrossing]

I basically never liked the Serum GUI, for one.

It became the gold standard for plugin synth UI because they made sure you didn't like it.

[zerocrossing]

Lil Zebralette 3 sounds incredible. I can’t wait for the full stripy horse to drop. I think u-he should take feature requests lightly at this stage. I’d much rather they focus on the trademark quality than try to cram everything into it, ’cos bro cannot play keys without an arpeggio.

Have you seen the video where Urs is talking to Nick Batt and he's going over how using the modulation system to create arpeggio type effects can be done? It's way more interesting, IMO, and if you need ol' fashioned arpeggios, your DAW probably has one built in.