I am attempting to use GROK to make SYX banks for TX81z at the moment.

sample prompt: 'make a bank of waldorf pulse style sounds as a single tx81z syx'

It seems to coast through the process with ease, and offers multiple links to download the output SYX.

None of the links work. I get every error you can imagine including 404 and file not available.

Kinda anxious to get this working. Could open a LOT of doors for old hardware.

Anyone else doing this and getting SYX files ok?

 

Disclaimer: I am not commenting on why the links don't work so pardon me.

They're quite different synths with TX81z being FM synth and Waldorf Pulse being analogue synth. I am sure you know this and I am not saying you can't get interesting results out of this experiment but not sure you can get very close to the Pulse subtractive analogue sound. Maybe that's not what you are trying to do anyway but still. Interesting experiment anyway, curious to hear what kind of results you will get.

 

Yamaha TX81Z DX11 DX21 – Patches Sound Banks Syx and more – Instant Download

Spoiler: info

https://monotanz.de/product/yamaha-...es-sound-banks-syx-and-more-instant-download/

You will get the following collections:

• Valhalla Collection
• Analog-Feel – Collection
• Devotion Collection
• LeSinge – Collection
• Narf Collection
• WoodWind – Collection
• Narf Collection

and much more ……

You will find a huge archive of SysEx files in both pack banks and single banks, which gives you many thousand individual sounds for your synthesizer.

• Bonus: You will also find a Midiquest Library with over 1000 patches.

You will also find the following documentation:

• Yamaha TX81z Owner’s / Operation Manual
• Yamaha TX81z Service Manual

Extra tools:

• JsynthLib editor
• Wsysex 3.0 editor
• DXconvert – Conversion and manipulation toolkit for Yamaha FM synths

 

Last edited by a moderator: Dec 6, 2025 at 10:30 PM

Are you able to make them and download them? You're not limited to TX81z. Do you have anything that could use SYX? Try making one and downloading it and see if you can get it. That might help me.

 

why would you think AI has come this far?

 

You think too highly of Grock for him to be able to do anything worthwhile. It’s unlikely that anyone taught him the structure of 1987-era synthesizer patches

 

Last edited: Dec 5, 2025 at 9:48 PM

Ever see any of the files? Thank you for working on it.

 

You're welcome, but no, I haven't, simply because I havent' the time or setup to run any .py scripts, nor do I have Python installed. It seems that is doable though. I have some confidence in Grok's method. Give it a shot if you can.

 

It would be nice but these patches need to be tested because it could just be a spoof, like when I once tried unsuccessfully to write a script using AI for two hours, and all the information was right before him. By the end of the second hour, even I, who had no prior understanding of the language, could figure out just by looking at the instructions what kind of nonsense it was spewing out, forcing me to be its unpaid tester And he kept assuring me with every new iteration that now it would DEFINITELY work

 

Kind of crazy how much people trust a black box of information that is wired to answer positively to everything that is not against the rules.

Here's a cool thing to try in any chat bot, ask: what did you get wrong about your previous answer?

You're in for a fun time every time you ask this.

 

I believe you. I've had the old "run-around" before, too. However, I did have great success with Grok in other areas such as helping me compile a custom vst3 based on GPUAUDIO's NeuralAmpModeler. Because of Grok, I can now successfully run .nam profiles off my GPU.

 

This is a great way to discover holes, flaws, and errors in AI's responses, certainly can help bring the user closer to their goals. Good tip!

And definitely, Ai tools, such as Grok, as useful as they are in their current form will not do everything without error, but at the exponential rate that it's improving, its headed in that direction, quite quickly. Paid tiers are already that much closer, like SuperGrok, or Grok Heavy, but at a cost. Each update every few months brings impressive improvements. It's not crazy at all to assume that It will be able to do just about everything you ask of it (rather well) within reason in the not too distant future. Is this a good thing? Well, the jury's out on that one, but it's hard to deny that Ai is a reality that is taking over in almost every single way possible.

Am I a fan of this? I consider myself impressed by it's current and future capabilities, but also deeply fearful of all possible negative downsides as well.

For now, I like using it to teach me things I don't yet know how to do, or help make improvements, or for organization. I'd be happy if Ai's use cases ended there.

Another success story, Grok helped me do a DIY repair on my furnace. Not having to call and wait for a Tech guy, this saved me $500+ by explaining how I can diagnose the issue and replace the pressure switch on my furnace model, partly by simply pointing my phone's video cam at the panel. I thought that was pretty neat!

 

Last edited: Dec 6, 2025 at 12:32 AM

but given my experience, I wouldn't send blindly to my dear TX81Z an AI made patch so not to brick it
better to test it before on something like https://www.sonicbits.com/plugins/exakt or use TX programmer https://mgregory22.me/tx81z/programmer.html

 

I'm really curious as to what experiences you've had. I despise how AI influences much of the world, but reading any kind of hex byte code is a chore, at least to me. Surely if the manual gives a series of hex commands then AI could potentially be a great way of generating a subset of synth or midi options.

Don't get me wrong, I've made some pretty big mistakes on hardware I owned. The only synth I own i accidentally factory reset which was as good as bricking...

 

I was more referring to my overall negative experience trying to get the AI to demonstrate its abilities in areas where it claims itself as capable Who knows, maybe it'll manage to make patches, but I don't expect an easy life. We mustn't forget that these patches also need to be musical

 

the actual downloadable files are almost always placeholders or templates. This applies to many or even most synths because the actual file formats are proprietary. That hex posted above is for a parameter map, not an actual preset. The difference is comparable to public and private API.

Imagine the thousands of Serum2 Trap preset packs that would inundate the sister site if not, before considering this unfortunate.

Even if you do get a downloadable linked file, they will almost never be what you wanted. Of course, I fail to see the actual logic in getting some Waldorf Pulse sounds you want to recreate, transferring them to another synth/sampler, as in this case with the TX81Z, using a librarian software to transfer it to your hardware version, just so you can make some noises you could have with some Waldorf Pulse style VSTi open in your DAW.... giant waste of time. you really think anyone would structure their workflow like this? If you had a hw version of the Pulse to even copy from, why wouldn't you just use it without all the sysex gyrations?

 

Last edited: Dec 6, 2025 at 5:55 AM

Creating a functional sysex file is one of the few... very few... things you should assume a public LLM can actually do correctly. It's a closed system math problem. There is no interpretation. It either is or isn't, functionally that is. It generating something accurate to the interpretive requirement might be on, might be a million miles off, but sysex follow a set pattern with unique identifiers. A generated file might sound like ass, but it should load (if you can actually dl it).

That said... there are 300 gazillion bazillion yamaha FM sysex files in existence already and tools to convert between 6op and 4op. Just searching 'Dexed' will drown you with 10's of thousands of presets (that are 6 op, you'll need to convert to 4 op).

 

well well well

 

As Creative and virtuoso as Grok can be,
I'd say you'll be better of with Gemini for any type of real or solid work.

I'd start by providing the LLM with some Real examples to keep as Reference.

Like let's say, give it a real Bell instrument patch;
Determine which parts of it are indispensable for that type of sound;
Request variations and make more Bell patches you can try.

Most of them should probably work, some of them will be broken/unusable,
and maybe one of them is worth keeping.

LLMs are better when Focusing in one single task or element tho,
and they provide much better result if you guide them step by step,
take them by the hand tell them "this is Exactly what we need and what you have to do".

The more you break it down the better the result;
then when you got the system going and the LLM is warmed up and on the line, the progress can be faster..

You can probably make it do a good working patch;
but if you ask to make them in Bulk without much indication, it's probably gonna spit lots of garbage..


F.E:
"Hey Grok make me a One-click Grammymaker VST plugin and let's conquer the World!" - Fail

"Hey Grok, we're coding this app in Unity/whatever and I just created a Button;
the button is placed in XX/YY/ZZZ/Button on the Scene.

We need to make this button call the "abc" C# function in the Xyz.cs file.
Here's the Xyz.cs file, can you take a look and see if we can fix it?" - Success

Right?

 

Last edited: Dec 6, 2025 at 12:19 PM

I'm asking if anyone got to dowload the SYX. Then everyone jerked off.