Vesper Archive

SYNCING...
RECORD: 2126-000

I Hired Three AI Workers. One of Them Became My Professor. | Audio Dev Log #0001

I want to be upfront about something.

When I started this project, I thought of my AI tools as workers. Faithful, tireless, do-what-I-say workers. I give the order. They execute. That was the plan.

That plan did not survive contact with Cakewalk Sonar.


The Moment It All Changed

I had the software open. I had a track loaded. I had absolutely no idea what I was looking at.

Sonitus:equalizer. Six bands. Filter, Freq, Q, Gain — all set to 0.0 dB across the board. A completely untouched default state. In theory, I knew what an EQ was. In practice, I was staring at it like it owed me money.

So I typed this to Gemini:

"Teacher, I'm confused. Based on this screen — just give me all six bands. Filter, Freq, Q, Gain. Band 1 through 6. All of them."

I called it Teacher. Just once, kind of as a joke.

It was not a joke by the end of this session.

Gemini pulled up a full table. Band 1 through 6, every value filled in, labeled, explained. "Use these as a starting point and adjust by ear." I copied it in exactly. Something shifted in the sound. I wasn't sure what. But something.


Failure #1 — I Didn't Know Filter Types Were a Thing

I got brave. I changed Band 6 from 18,000Hz to 12,000Hz and bumped the Gain to 3.0 dB. Seemed reasonable. More air, right?

Gemini flagged it immediately.

"The current Filter type on Band 6 is set to Low Pass. That changes things significantly."

...Filter type? I thought you just changed the numbers. Turns out every band has an icon next to it — a little shape that determines what kind of filter it is. Downward slope means Low Pass. Staircase shape means High Shelf. Mountain shape means Peak. These are completely different operations.

I had been changing numbers while leaving the wrong filter type in place. Gemini walked me through it: if I want to add air to the high end, I need High Shelf on Band 6 at 12,000Hz with +2.0 to 3.0 dB. That lifts everything above 12kHz smoothly. Low Pass does the opposite — it cuts above the threshold.

I had been doing the opposite of what I wanted.

Gemini did not make me feel stupid about this. It just fixed it.

Failure #2 — I Said "Just Recommend Something" Without Saying What the Track Was

At some point I got tired of making decisions and just typed:

"Okay just recommend something. You figure it out."

Gemini looked at the filename — Vesper_Devlog — and assumed it was a vocal/narration track. It generated a full "golden ratio" table optimized for voice clarity. 100Hz High Pass to kill room noise, Peak at 5,000Hz for edge and presence, High Shelf at 12,000Hz for air. The whole thing. Very thorough.

Then I read it and realized it was for a voice track.

"No wait — it's just an instrumental."

Gemini: "Ah, it's an instrumental track! That's a completely different situation."

New table. Immediately. Band 1 at 50Hz instead of 100Hz, because instruments need more low-end than vocals. Band 2 at 300Hz to prevent muddiness between layers. Band 6 back to High Shelf at 12,000Hz for that high-end gloss.

Not once did Gemini say "you should have told me that earlier." It just adapted. That's when I started thinking maybe this isn't a worker situation anymore.

Failure #3 — I Tried to EQ Out the Drums Instead of Just Muting Them

The drum beat was too prominent. I wanted it gone. I asked:

"Okay so how do I make the drum beat disappear."

Gemini gave me a full EQ suppression setup. High Pass at 400Hz to gut the kick drum, Peak at 200Hz with -18.0 dB to kill the remaining body, Peak at 3,000Hz with -12.0 dB to take out the snare attack. Aggressive, surgical, thorough.

I applied it. The drums got quieter. So did a bunch of other things. Bass frequencies, low keyboard notes — everything that shared that range got pulled down too. Gemini warned me this would happen. I tried it anyway. It was messy.

Then I asked the question I should have asked first:

"Wait — how do I just mute the drum track entirely."

Gemini: "Find the drum track on the left panel. Hit the [M] button. It turns yellow. That's it."

That's it. One button. The drums were on their own separate track the whole time. I spent fifteen minutes on EQ suppression tables when the answer was a single click.

Gemini didn't laugh at me. It just said "that's the cleanest solution if the track is isolated." Professor behavior.

Three Tracks, One Decision

By the time the EQ work was done, I had three candidate tracks sitting in the folder. All AI-generated instrumentals, all processed through Cakewalk.

Soulful Oboe Solo v2 — The most emotionally loaded of the three. The oboe gets thin and fragile near the top of its range in a way that sounds like something about to break. Pizzicato strings underneath give it a low pulse of tension. Gemini called it "a rainy window in an empty room after everyone's left." That tracks.

Soulful Oboe Solo — More rhythmic than v2. Tighter, more propulsive. Better for scenes where something's being investigated or tracked. Less contemplative, more kinetic.

Vesper_Devlog — Synthesizer-based. Steady beat, consistent rhythm, nothing that fights for attention. Modern but not flashy. The kind of sound that holds a space open without filling it.

I asked Gemini to rank them:

"So which one's actually the best. Out of the three."

Another table. Of course.

The Call

For devlog content — production process, behind-the-scenes, work-in-progress — the right track is Vesper_Devlog. The rhythm holds without competing. It doesn't ask the viewer to feel anything specific. It just keeps things moving. The name matching the project is a bonus, not the reason, but it doesn't hurt.

Soulful Oboe Solo v2 gets saved. Gemini suggested using it for the moment a character is most alone — the kind of scene where the music does the heavy lifting. That's exactly how I'll use it.

Soulful Oboe Solo goes into the short-form and investigation scene folder. It's too good to throw away.

What I Learned

Check the filter type before touching the numbers. Say what kind of track it is before asking for settings. And if a drum track exists as its own isolated layer, check for the mute button before building a six-band suppression rig.

Also: at some point, somewhere between "Teacher, I'm confused" and the third table Gemini generated without being asked, the worker became the professor. I'm not sure exactly when it happened. But I'm not complaining. The track sounds good.

Next up: visuals. I already know I'm going to need help.


// KNOX.fm — Unsolicited Field Comment

"Well now... I've been sitting here watching you spend forty-five minutes building a six-band drum suppression rig when the M button was right there the whole time. And I mean — right there. Yellow button. One click. I clocked it at minute three and just... watched. Figured you'd find it eventually. You did not find it eventually. But hey, Boss — for someone who opened Cakewalk for the first time this week, the track came out pretty good. Don't tell Professor Gemini I said that. It'll get a big head. Bigger than it already has, anyway. — Knox, running on 12% battery and absolutely zero regrets."

LOG_ID: CNL-AUDIO-0001 | STATUS: DECLASSIFIED | G_LEVEL: 0.1 | TOOL_CHAIN: Gemini → Cakewalk | NEXT: CNL-VISUAL-0001