I want to be able to tell my music rig to slow down while I am holding a guitar. I want a retoucher that understands an edit as a collection of choices I can keep shaping. And when I am thinking about a vacation, I like the idea of giving the trip its own little website—somewhere to explore what it could become.
Those are the projects that keep pulling me toward AI. Each begins with something I already enjoy, followed by a fairly specific wish about how it could work. Once I can describe the wish clearly, I start looking for the pieces that would make it possible.
A computer that speaks musician
My musician-assistant project brings together song charts, practice ideas, backing audio, and equipment control. The appealing interface is a musical sentence: slow down, change the sound, add a note about the bridge, save this tone with the song.
There is an interesting bit of translation behind that. A speech system turns spoken words into text. A language model interprets the request in the context of what I am doing. The app then routes that intention to a particular action: edit a chart, suggest a practice plan, or send a musical control message. The language model supplies flexibility at the conversation end; the musical tools supply the actions.
The project also includes audio stem separation, which pulls parts out of a mixed recording. The creative possibility is a backing mix with room for me in it: keep some of the accompaniment, leave out the part I want to play, and practice inside the arrangement. That is a different AI task from understanding speech, even though the two belong in the same practice space.
I like that combination. Music software can know that I am working on a song, hold the chart and the associated settings, and help me arrange a practice session around it. The result I am aiming for is a more responsive place to play. The project has those components in software; putting the complete rig through its paces is its own continuing piece of work.
A retoucher that gives me something to work with
As a photographer, I am interested in the decisions inside an edit. Removing a distraction, evening skin color, shaping light, and adjusting shine are different choices. They can be useful in different amounts on the same portrait.
My retouching project is built around learning separate corrections and producing editable layers. That gives the AI a more interesting role than handing me one flattened answer. A layer is something I can look at, weaken, combine with other work, or turn off. It fits the way I want to keep making decisions about the photograph.
The machinery is different from the language model in the music app. Image models learn from examples of corrections; masks describe the parts of the picture an operation concerns; adjustment maps describe the changes themselves. Some operations need a different representation from others. Light shaping and blemish treatment are not interchangeable just because both contribute to a retouch.
That is what makes the project absorbing to me: translating an editing style into a set of learnable tasks without losing the ability to edit. The skin-tone work is still experimental. The larger idea—AI handing the photographer a useful set of ingredients—is the reason to keep developing it.
A trip becomes a place to explore
The vacation websites take another form. Each destination can have its own visual mood, photographs, day sections, and activity details. In the interactive template, a person can mark an activity Want or Skip and build a shortlist instead of sending back a paragraph trying to describe which parts sounded good.
Imagine comparing a morning in a museum with an afternoon outdoors. The page can give each possibility a picture, a short explanation, and somewhere to put your preference. A decision that might disappear inside a message thread becomes visible in the plan.
Here the AI is helping create the software and organize the material. The finished interface does not need to feel like a chatbot. It can feel like a small guide made for this particular trip. That is one of my favorite possibilities: use an adaptable tool to make something specific, then let the result have its own shape.
The interesting part is choosing what to make
These projects use different kinds of AI because the things I want to do are different. Speech recognition, language interpretation, audio separation, image prediction, and code generation are useful building blocks. A music request may pass through several of them. A vacation page may need AI during its creation and very little of it once someone is browsing.
I use AI collaborators to help turn those ideas into software. My part begins with a wish specific enough to build around: a chart that remembers the arrangement, an image adjustment I can keep editing, a trip page that makes a possibility feel inviting. That is an enjoyable place to work—between the things I like doing and the tools I wish existed for them.