Creative Process

The three of us meet regularly to discuss the theme and direction of each album. At this stage, AI is not involved. We begin with our own thoughts, experiences, questions, senses, and conversations.

We first decide the larger theme of the album and its art direction. Then we discuss what each track should explore, how the smaller themes should unfold, and in what order the tracks should be experienced.

The essays and songs of Level Up Audio are designed to stand on their own. But they are also planned so that, when experienced in sequence, they create a deeper emotional and conceptual flow. That structure begins at the planning stage.

Once the first round of planning is complete, we use AI to review and document the album’s structure.

In this process, we refer to multiple AI models to cross-check the logic of the concept, the connection between themes, the emotional flow, and possible areas for further development. AI sometimes offers new suggestions, but only a portion of those suggestions are actually adopted into the work.

We know that AI can be excellent at organizing ideas, comparing perspectives, and helping us see a plan more objectively. We also know that AI cannot replace our experience, intuition, context, or responsibility. So we use AI with humility, but the final judgment always remains human.

The refined plan is then documented and stored in the cloud, so each director can refer to it at any time and keep the project aligned throughout the later stages of creation.

The essay is the writer’s domain. In Level Up Audio, every source of inspiration begins here.

Because we are more comfortable thinking and writing in Korean, we often begin with Korean essays and use AI to help translate or reshape them into English. But we do not use the translated text as-is. We read it again, revise it, and work through the language until the English version carries the emotion, meaning, rhythm, and nuance of the original Korean essay as closely as possible.

This is one of the most time-consuming parts of our process. We believe that a strong essay leads to stronger lyrics, and stronger lyrics allow the music and visuals to move in a clearer direction.

The representative image for each track often appears before the essay is finished. Sometimes it emerges while writing. Sometimes it comes after receiving an essay from another writer. We place that first visual impression inside the essay document. From there, the image becomes the beginning of the music, video, cover art, and overall visual atmosphere of the track.

To create the image, we refer to our own photographs, previous works, visual references, and the feeling of the scene we have in mind. We then use AI to explore different visual directions. Sometimes we generate dozens, or even hundreds, of samples. Most of them are discarded. We select the one direction that feels most deeply connected to the essay, and then we revise and refine it.

The final image is not simply used as it was generated. We shape it through both traditional editing methods and newer forms of intuitive visual adjustment. We adjust color, texture, density, atmosphere, and visual flow until the image feels more beautiful, more meaningful, and more faithful to the essay.

Once the essay is complete, we write the lyrics based on it. This is also one of the stages where we receive the most help from AI.

Because we are more comfortable in Korean, we use AI to help refine rhythm, phrasing, and expression in English lyrics. But word choice, structure, repetition, chorus placement, density, and the overall flow of the lyrics are continuously reviewed and revised by human hands. As with the essays, we spend a great deal of time and attention on the lyrics.

The generative music tools we use rely heavily on lyrics. For that reason, we believe that good music begins with good lyrics. Lyrics are not just input for the music. They are the emotional and structural foundation of the song.

When creating music, we do not simply generate many tracks at once and choose one quickly. We listen to each version from beginning to end. We consider the mood of the song, the clarity of the lyrics, pronunciation, emotional flow, and, most importantly, whether the music is genuinely good to listen to.

So far, a song has almost never arrived perfectly on the first attempt. Most tracks go through repeated cycles of generation, listening, rejection, revision, and retrying. In many cases, it takes around twenty attempts before we arrive close to the direction we were searching for.

In the future, if the day comes when we can devote ourselves more fully to Level Up Audio, we hope to expand the role of human performance, composition, arrangement, and recording. We also look forward to collaborating with other artists. AI is a tool that currently helps make our work possible and expands what we can explore, but the direction of Level Up Audio remains open toward deeper human expression and collaboration.

We create videos that allow listeners to experience each song while also following the lyrics.

As a first step, we use AI-assisted tools to generate a draft lyric video. But we do not use that draft as-is. While listening to the music, we adjust the timing of the subtitles, the breathing of each line, the transitions, the pace, and the density of the screen. The goal is to help the viewer focus naturally on the music and the lyrics.

The background of each lyric video is usually based on the representative image of the track. That image is not simply placed behind the text. We add layers, subtle motion, color, and texture so that the visual tone supports the song without distracting from it. We aim for a background that quietly serves the music rather than calling too much attention to itself.

In the description of each YouTube video, we include a link to the full essay on our website, along with a brief explanation of the song’s background. We want the music, lyrics, image, and essay to remain connected.

One of the experimental directions Level Up Audio is exploring is the full-length music video. Every result created through the process described above has its own meaning. But among them, there are certain songs that stay with us more deeply, continue to echo in our minds, and feel especially worth expanding into a longer visual form. Around once a month, we plan and produce a full music video based on the essay and music of one of those pieces.

We call this work experimental because it reveals very clearly how we approach AI.

Traditional full-length video production usually follows a linear structure. In the industry, this is often described as a waterfall model: planning, filming, editing, and post-production are each strongly dependent on the stage before them. Once a goal has been set, changing it significantly in the middle of the process is difficult because of budget, time, and production constraints.

Working with AI changes that structure. Rather than fixing one goal from the beginning and moving toward it in a strictly linear way, we can test multiple directions, review the results, and continue adjusting the work toward a better outcome. Even the final goal is not always a fixed conclusion from the start. It can become a flexible hypothesis that is refined as the work develops.

At the moment, Level Up Audio’s full-length music videos are primarily created by Hyun S, our senior director, working alone. Over the course of about a week, he creates, reviews, discards, rebuilds, and connects many scenes into a single video. The results are not always perfectly satisfying. Still, completing one music video over seven days with AI, imagination, and the continuous judgment of one person is, in one sense, very fast, and in another sense, a process that requires a long and focused effort.

This work is not about replacing human creativity with AI. It is an experiment in how far human creativity can be expanded through collaboration with AI. It is also one of the ways Hyun S reaches out to listeners through Level Up Audio.

The Level-up.audio website was planned and built by us, from backend to frontend. We use AWS, WordPress, Cloudflare, and other tools to provide the service, and we build or customize the functions we need to create the experience we want.

Even a simple-looking page requires many small decisions if we want the experience to feel refined and graceful. In this process, AI is used for advice, code review, structural checks, and alternative suggestions. But every implementation is reviewed and tested by humans before it becomes part of the site.

For us, the website is not just a collection of links. It is the original archive of Level Up Audio, where essays, lyrics, music, images, and videos can exist together with their full context.