There's No Perfect Prompt. Only the Sense of Keep Moving.

ChulbujiRunningAIMusicVideoVibeCodingAICollaborationRunBuddyBuildingInPublic

At first, I tried to put too much into a single scene. I thought it had to include stretching, the moment of taking off, a living atmosphere, and visible emotion — all at once. But the more I loaded in, the more awkward the result became. Movements tangled, frames broke, and the scene I wanted never appeared.

Here’s the ChulbujiRunning music video we made. See for yourself.


At first, I was a little thrown off. The scene was clear in my head — but the AI wouldn’t pull it out the way I imagined. I assumed describing what I wanted in more detail would help. It didn’t always. In fact, asking for too many things at once was exactly what caused the scene to collapse. That lesson landed more clearly this time than ever before.

The breakthrough wasn’t dramatic. It was about pulling back the ambition, breaking the scene into smaller pieces, and giving only one clear instruction at a time. Not “a scene where they stretch and then take off running” — but “a smooth tracking shot of someone beginning a light jog.” Giving one precise movement at a time worked far better.


Watching this unfold, I kept thinking about how much it resembles running. When you try to run well from the very first step, your body locks up. Breathing collapses. Posture tenses. And eventually, pace breaks down. But if you keep your gaze just a little ahead, and focus on one breath, one arm swing, one stride at a time — somehow the flow comes alive. AI worked the same way. Rather than trying to extract a perfect result in one go, making the next single step precisely was what got results.


This held true whether it was long-form or short-form. For a 3-minute video, I had to keep the full arc in view. For a 15-second short, the energy had to land in the first few seconds. Even with the same content, the rhythm for YouTube and Instagram was completely different. Long video needed slow, accumulating force. Short video needed an immediate punch.

Unexpected problems came up along the way. A strange filter covered the whole frame once — that threw me. And then I’d view a 16:9 scene in 9:16 and realize: this is an entirely different image. At first, these corrections felt like friction. But as I kept going, I understood. This wasn’t failure — it was the reality that every platform runs differently. YouTube’s rhythm and Reels’ rhythm are not the same. Scenes, captions, and timing all had to be re-fitted accordingly.


What this work confirmed, again, is a very simple fact. There is no perfect prompt. You don’t start already knowing everything. You make something, it breaks, you fix it, you adjust — and that’s how the sense of it builds.

When I think about it, my relationship with running isn’t so different. I’ve never started a run with perfect form, perfect pace, and a perfect training plan in place. I run, I listen to what my body says, I slow down when it’s too fast, I push a little further when the flow is good — and that’s how I learn my own rhythm. Making this AI music video was exactly that. In the end, what moved me forward wasn’t perfect preparation — it was the repetition of pressing start, making something, and fixing it when it didn’t work.

“It doesn’t have to be perfect. One step today is enough.”

That line didn’t feel like just a lyric. This time, it was the actual way I worked. I didn’t arrive here because I had the perfect plan. I arrived here because I kept moving — even in small steps.


The biggest insight from this project is clear. AI collaboration is not a game of finding the right sentence. It’s closer to a craft of breaking scenes down, adjusting the flow, and pushing through to the end. That’s the standard I’ll use going forward for ChulbujiRunning content. Not trying to get it right in one shot — but decomposing into scene-level units, re-editing to match each platform’s rhythm, and building intuition through execution.


What I made: A ChulbujiRunning music video, built scene by scene with AI. What broke: Greed. Putting too much into one prompt collapsed the scene every time. What I learned: There’s no perfect prompt. There’s only the sense of keep moving — same as running.