I've been trying to integrate Suno into some of our internal creative workflows for months now. After generating hundreds of tracks across every style they offer, I keep coming back to the same conclusion.
The 80s synth-pop mode is the only one where the output feels consistently coherent and usable. The drums have a proper pattern, the bassline locks in, and the melodies actually resolve. It doesn't sound like a random collage of AI music ideas.
With other genres—especially modern pop or rock—the structure falls apart. The verses and choruses blur, the instrumentation gets chaotic, and there's often a "demo" quality that's impossible to polish. It's like the model truly *understands* the constraints and tropes of 80s production, but struggles with anything less formulaic.
Has anyone else found a specific mode or set of prompts that yields usable results outside of this one niche? I'm curious if this is a limitation of the training data or if I'm just approaching the other genres incorrectly.