Gemini AI to make tones

I've come to realize that along with all the logistical issues around LLMs/GPTs (environmental, economic, societal, etc), a huge problem I have with them is they're the world's ultimate plagiarism machine.

These things literally can only do the things they do because of the all the human work that has come before. The only reason it can write code, summarize a document, or create a song is because enough humans did those things, by themselves, doing the hard work. And it's just regurgitating all that work with some sparkly random numbers mixed in to make it seem like it's new.

Every time I even think about using one of these tools, I just can't get over the fact that I didn't earn the output. It's not my work. It's everybody else's, literally, and I can't even find out who in order to thank them.

As a creative, as an engineer, and as a human being, the whole thing just makes me sick to my stomach. They've commoditized humanity, the things about us that are unique and special, into a meaningless and soulless stream of... well.. slop.
 
Last edited:
I use grok to find some explanations on some things related to the fractal world.
Even though I’m on the free plan, the expert mode is really great and grok knows how to explain sometimes difficult things in a pedagogic and simple way.

I learned a lot about compressors etc
 
I do not understand why anyone would want to deny themselves the joy that is dialing in and finding their sounds. It's not dissimilar to learning some lick you can't quite pull off at first and then finally you nail it and can consistently nail it.

There are already presets inside the machine for all the sounds I've seen referenced in this thread. Will they sound 1:1 like the record? Nope. Because there are so many other factors at play BUT they will get you to started on where you are trying to go. Especially in a cover situation/etc.

Also RE: learning -- all AI is doing is scraping data so why not just read the actual articles/etc that the data is coming from? I'm not trying to be rude I just don't understand why a composite snapshot of data is better than a well written article about it. Especially something like compressors.
 
helloiamdustinwhite said:
I do not understand why anyone would want to deny themselves the joy that is dialing in and finding their sounds. It's not dissimilar to learning some lick you can't quite pull off at first and then finally you nail it and can consistently nail it.

There are already presets inside the machine for all the sounds I've seen referenced in this thread. Will they sound 1:1 like the record? Nope. Because there are so many other factors at play BUT they will get you to started on where you are trying to go. Especially in a cover situation/etc.

Also RE: learning -- all AI is doing is scraping data so why not just read the actual articles/etc that the data is coming from? I'm not trying to be rude I just don't understand why a composite snapshot of data is better than a well written article about it. Especially something like compressors.
Click to expand...
That’s totally irrelevant.
I sometimes ask AI for help as I’ve mentioned but I spent and will still spend hours on the unit itself (not even on the editor) and love discover things.

Second :

You are rude.
Sometimes you need a quick answer on a problem that you DONT want to spend 4 hours on it.
I dont see why one would not use AI as as you said is compelling data.

I’m totally against those idea that if u use AI you are a less user or human that someone that don’t.

AI is à wonderful tool.

Don’t be a snob.

You are making generalities without knowing, shame for the judgement.
 
AI even with the jumps tech are wrong.... a lot more than I'd be comfortable with using them as a search engine. You have to double triple check.
 
Taco Bell Eating GIF by Jukebox Saints
 
I decided to try Gemini Pro 3.1. I loaded the info I felt relevant in Pic 1. Pic 2 is the result after about a minute. It's been thinking and doing the buffer spin for over 5 minutes now. So I'm unsure if it's actually still trying or if I broke it and the world will hate me. How long did it take to create the answers, those of you who got answers from it?
 

Attachments

  • Gemini 3.1 pic 1.png
    Gemini 3.1 pic 1.png
    100 KB · Views: 28
  • Gemini 3.1 pic 2.png
    Gemini 3.1 pic 2.png
    117.4 KB · Views: 28
My prompt to Gemini:
I have a Fractal FM9 guitar amp/effects modeler. If I upload an unamplified sample direct from my guitar (the DI signal) and a sample of an amped guitar, with effects, that I would like to re-create through the FM9 would Gemini be able to provide settings for a FM9 preset to obtain the desired tone?


Gemini Response (Fast Thinking mode):
Gemini cannot create a precise preset for an FM9 guitar amp/effects modeler. It cannot perform a technical "Tone Match" to generate exact parameter settings.

Capabilities
If you upload the DI and amped samples, this AI on Google Search can provide a theoretical framework for building the preset:
  • Signal Chain Suggestions: It can suggest which blocks you might need and their order.
  • Model Matching: Based on the audio characteristics, it can recommend specific FM9 amp and cabinet models that might get you close to the target sound.
  • Initial Parameters: It can provide estimated starting points for knobs, but these will be approximations.

Recommended Alternatives
For an accurate recreation of the amped tone from a DI signal, the FM9 itself and other specialized tools are more effective:
  1. Fractal Tone Match Block: The FM9 can utilize Tone Match data.
  2. Impulse Response (IR) Capture: You can capture the frequency response of your amped setup as an IR. Using the FM9’s Cab block with a custom IR is often considered more accurate than a general EQ match.
  3. Third-Party AI Tools: Specialized tools like NAM (Neural Amp Modeler) or ToneGPT are designed specifically for amp profiling and can sometimes be integrated into a Fractal signal chain via the effects loop.
Pro Tip: For the best results, upload the FM9 Owner’s Manual or the Blocks Guide as a source document. This allows the AI to reference the specific models and parameters available in your firmware version.


Gemini Response (Pro mode):
No, Gemini cannot analyze an audio file to generate a Fractal FM9 preset or precise numerical settings.

While Gemini can describe the audio (e.g., "high-gain distortion with scooped mids"), it cannot "listen" to a track and mathematically reverse-engineer the signal chain into a compatible .syx file or parameter list. Furthermore, the FM9 lacks the onboard "Tone Match" creation feature found in the flagship Axe-Fx III, meaning it cannot natively "capture" a tone from a sample by itself.

To achieve this, you must use a specific "Offline Tone Match" workflow involving a computer.

The "Offline Tone Match" Workflow
Since the FM9 cannot generate a Tone Match internally, you must create one using a DAW (Digital Audio Workstation) and import it as an Impulse Response (IR).

1. Isolate Audio
Prepare Your Tracks: In your DAW (Logic, Pro Tools, Reaper), load your Reference Track (the tone you want) and your Source Track (your dry DI guitar recording).

2. EQ Match
Apply Matching Plugin: Use an EQ matching plugin (e.g., FabFilter Pro-Q 3, Logic Match EQ, or iZotope Ozone).
  • Set the plugin to "listen" to the Reference Track as the Target.
  • Set it to "listen" to your DI Track as the Source.
  • Run the "Match" function to generate a complex EQ curve that makes your DI sound like the recording.
3. Export IR
Create the WAV: Export only the resulting EQ curve (usually by playing a white noise burst through the matching EQ) as a mono .WAV file (48kHz, 24-bit). This WAV file is your "Tone Match."

4. Import to FM9
Load via FM9-Edit: Open FM9-Edit and go to the "Manage Cabs" tool. Drag your exported WAV file into a "User Cab" slot. The software will automatically convert it to the Fractal format.
5. Build Preset
Cab Block: In your FM9 preset, place a Cab Block after your Amp Block. Select the User Cab slot you just loaded. This IR now applies the entire "tonal DNA" of the recording to your signal.

Why this is necessary
Hardware & Software Limitations

Tone Match Capability
The Axe-Fx III has a dedicated block that performs this entire process automatically in real-time. The FM9 and FM3 chips are not powerful enough to run the calculation, but they can play back the result. By doing the math on your computer (Step 2 above), you bypass the hardware limitation.

Gemini's Role
Gemini is a Large Language Model, not a Digital Signal Processor (DSP). It analyzes text and image data, not raw audio waveforms for frequency analysis. If you upload the audio to Gemini, it might say "This sounds like a Vox AC30," which can help you choose the starting Amp Model, but it cannot provide the thousands of data points needed to replicate the speaker cabinet and mic placement.
 
This is a weird topic to argue about. AI is a useful tool (like all of the other ones we use) for getting you in the ballpark and explaining what makes up a particular sound and why really fast (of course using the information shared by the pros that have figured out what knob to try to solve what particular problem). It is like reading a book or using google, aren't these just two different ways to try to solve the same problems? You can read through or watch videos on low and high cutting the cab block (the same question and answer 500 times) or let an AI engine sum it up for you and put it in context to what you are trying to do. In no way does it replace just grabbing blocks and trying different things to see where the gear takes you (and should not as there will be no new sounds) or loading up a preset and just seeing what kind of riffs that sound inspires you to play.

It is never going to replace a good set of ears but it can do this sort of thing very well to help get new users past settings paralysis. It gives you a few knobs to turn to see what it does to the sound.


1771875334495.png
 
Last edited:
It's not going away so use any tools at your disposal if you choose to accomplish a task. It's helped me for certain fractal related topics.
 
prina said:
My prompt to Gemini:
I have a Fractal FM9 guitar amp/effects modeler. If I upload an unamplified sample direct from my guitar (the DI signal) and a sample of an amped guitar, with effects, that I would like to re-create through the FM9 would Gemini be able to provide settings for a FM9 preset to obtain the desired tone?


Gemini Response (Fast Thinking mode):
Gemini cannot create a precise preset for an FM9 guitar amp/effects modeler. It cannot perform a technical "Tone Match" to generate exact parameter settings.

Capabilities
If you upload the DI and amped samples, this AI on Google Search can provide a theoretical framework for building the preset:
  • Signal Chain Suggestions: It can suggest which blocks you might need and their order.
  • Model Matching: Based on the audio characteristics, it can recommend specific FM9 amp and cabinet models that might get you close to the target sound.
  • Initial Parameters: It can provide estimated starting points for knobs, but these will be approximations.

Recommended Alternatives
For an accurate recreation of the amped tone from a DI signal, the FM9 itself and other specialized tools are more effective:
  1. Fractal Tone Match Block: The FM9 can utilize Tone Match data.
  2. Impulse Response (IR) Capture: You can capture the frequency response of your amped setup as an IR. Using the FM9’s Cab block with a custom IR is often considered more accurate than a general EQ match.
  3. Third-Party AI Tools: Specialized tools like NAM (Neural Amp Modeler) or ToneGPT are designed specifically for amp profiling and can sometimes be integrated into a Fractal signal chain via the effects loop.
Pro Tip: For the best results, upload the FM9 Owner’s Manual or the Blocks Guide as a source document. This allows the AI to reference the specific models and parameters available in your firmware version.


Gemini Response (Pro mode):
No, Gemini cannot analyze an audio file to generate a Fractal FM9 preset or precise numerical settings.

While Gemini can describe the audio (e.g., "high-gain distortion with scooped mids"), it cannot "listen" to a track and mathematically reverse-engineer the signal chain into a compatible .syx file or parameter list. Furthermore, the FM9 lacks the onboard "Tone Match" creation feature found in the flagship Axe-Fx III, meaning it cannot natively "capture" a tone from a sample by itself.

To achieve this, you must use a specific "Offline Tone Match" workflow involving a computer.

The "Offline Tone Match" Workflow
Since the FM9 cannot generate a Tone Match internally, you must create one using a DAW (Digital Audio Workstation) and import it as an Impulse Response (IR).

1. Isolate Audio
Prepare Your Tracks: In your DAW (Logic, Pro Tools, Reaper), load your Reference Track (the tone you want) and your Source Track (your dry DI guitar recording).

2. EQ Match
Apply Matching Plugin: Use an EQ matching plugin (e.g., FabFilter Pro-Q 3, Logic Match EQ, or iZotope Ozone).
  • Set the plugin to "listen" to the Reference Track as the Target.
  • Set it to "listen" to your DI Track as the Source.
  • Run the "Match" function to generate a complex EQ curve that makes your DI sound like the recording.
3. Export IR
Create the WAV: Export only the resulting EQ curve (usually by playing a white noise burst through the matching EQ) as a mono .WAV file (48kHz, 24-bit). This WAV file is your "Tone Match."

4. Import to FM9
Load via FM9-Edit: Open FM9-Edit and go to the "Manage Cabs" tool. Drag your exported WAV file into a "User Cab" slot. The software will automatically convert it to the Fractal format.
5. Build Preset
Cab Block: In your FM9 preset, place a Cab Block after your Amp Block. Select the User Cab slot you just loaded. This IR now applies the entire "tonal DNA" of the recording to your signal.

Why this is necessary
Hardware & Software Limitations

Tone Match Capability
The Axe-Fx III has a dedicated block that performs this entire process automatically in real-time. The FM9 and FM3 chips are not powerful enough to run the calculation, but they can play back the result. By doing the math on your computer (Step 2 above), you bypass the hardware limitation.

Gemini's Role
Gemini is a Large Language Model, not a Digital Signal Processor (DSP). It analyzes text and image data, not raw audio waveforms for frequency analysis. If you upload the audio to Gemini, it might say "This sounds like a Vox AC30," which can help you choose the starting Amp Model, but it cannot provide the thousands of data points needed to replicate the speaker cabinet and mic placement.
Click to expand...
I thank you for the detailed response. I have a starting point now. All the best
 
Rags said:
Testing AI proved to be fruitful. It can go and fetch the manual and WIKI's on its own as well, just be sure to specify what FW ver you need and what unit you're working with for an up to date answer. A simple question I asked produced the following result (in nested quotes); I remember catching it once (with the free version) making a simple error, but it may have been the way the question was formulated that led to an off kilter response. Otherwise, you're right, it very well may be the future for streamlined tweaking. I think the thing that intrigued me most was how it 'talked' to itself when 'thinking'. I ran some screen captures so I could analyze the output which made me chuckle in amazement to no end. The entire session is pages long.

Here's a sample ....

" ' What does the EVH 5150 III tube head equate to in a Fractal Axe 3 environment ' ?

' Great question — this is exactly where your Fractal Axe-FX III library of amp models comes into play.

The EVH 5150 III Tube Head (sometimes called the 5150 III or EVH 5150III Stealth) is modeled in the Axe-FX III under the “EVH 5150 III” family of amp blocks. Fractal lists them as:

Fractal Axe-FX III equivalents:​

  • “PVH 6160 Block” → models the Peavey 5150 / 6505, which is Eddie Van Halen’s original circuit.
  • “EVH 5153 Blue” → mid-gain channel, based on the EVH 5150 III Blue channel (great for tight rhythm with crunch).
  • “EVH 5153 Red” → high-gain channel, based on the EVH 5150 III Red channel (saturated lead tones).
  • “EVH 5153 Green” → clean channel model, taken from the same amp. ' "


It also produced some elementary block schematics that showed me a new way to tackle building a preset based on very specific questions. The end results were not bad at all.


" 👉 Since you’re building dual-amp scene layouts, you could easily set up:

  • Scene A: EVH 5153 Blue (tight rhythm)
  • Scene B: EVH 5153 Red (singing lead)
  • Blend or split with your Fender-style clean amp block for dynamics.

Would you like me to sketch a block grid for a Jason Aldean-style preset (5153 Blue + Fender Clean + delay/reverb scenes) in the same way we’ve been documenting your JVM/Fractal setups? That way you’d have a country-rock “arena” option ready to print. ' "
Click to expand...
wow... does Jason Aldean actually use a 5153 ?! :cool:

He rocks for sure, interesting country rock mix rig !
 
LiamH said:
It's not intelligent. LLM AI scrapes text from the internet and matches word patterns. A good proportion of otherwise intelligent people evidently think their AI companions are capable of conscious thought. Of course there are some media and tech billionaires that are likely already using this senseless waste of electricity to their own advantage.
Click to expand...
That's something about LLMs most users don't understand. It's not thinking. It's using probabilities to string words together based on what it scraped off the web.

To grossly oversimplify what is a complex programming task, I'm sure. The value of LLMs to the average person is marginal at best.
 
You must log in or register to reply here.
Back
Top Bottom