Okay, I have to admit, when we first started using Suno for mood music on our SaaS project, I was ready to pull my hair out. The first batch of outputs felt like pure chaos — some were weirdly upbeat for a calm onboarding flow, others had vocals when we strictly needed instrumentals.
So our design team (just three of us!) set up a real vetting process. It felt overkill at first, but honestly, it saved us *so* much time and frustration. We ended up generating about 50 clips over a week to find that one perfect background track for a key user tutorial. Here’s exactly how we did it.
First, we got super specific with the prompt. Instead of "calm tech music," we wrote: "60-second lofi beat, no vocals, BPM between 80-90, melodic synth lead, subtle vinyl crackle, for a software tutorial screen." We used that exact base prompt for 20 generations, only tweaking one element at a time (like swapping "synth lead" for "warm piano"). This let us actually compare apples to apples.
Then, we listened to everything in a dedicated FigJam board! We dropped each MP3 into a frame, tagged it with the prompt variation, and used sticky notes for quick reactions. We had columns for "Yes," "Maybe," and "No." The key was doing this together, on a call, so we could hear when someone said "Ooh, the percussion on this one fits perfectly" or "The melody here is distracting."
Our final filter was a quick, dirty user test. We took our top 5 tracks and slapped them onto a prototype in Figma. We sent the prototype link to five non-design teammates and asked one simple question: "Does the music feel appropriate for learning a new feature?" Their gut reactions were gold — and they unanimously picked the same track we were leaning toward. The data beat our opinions.
It’s a process, for sure. But now we have a template for scoring Suno outputs on fit, vibe, and usability. No more guessing! Anyone else building a system for sorting through AI audio?