Design an ambient sound bed that sells a specific location without a single spoken word

Why this works

Sound design prompts fail most often when they're written like music prompts — asking for a genre and mood produces a melodic ambient track with a recognizable musical shape, which is the wrong output entirely for a scene bed that needs to feel like environment rather than composition, so explicitly ruling out verse/chorus structure and naming this as "textural, non-melodic-forward" redirects Suno v5 away from its song-generation defaults toward something closer to a field-recording-plus-drone hybrid. Listing specific sound layers individually (the train rumble, the fluorescent hum, the metallic drips, the synth drone) rather than a single mood adjective works because each named element gives the model a distinct textural layer to generate and blend, which is how real sound designers build a scene bed — layer by layer, not as one homogenous pad. Asking for an emotional undercurrent "without being on-the-nose or dramatic about it" is a deliberate constraint against Suno's tendency to resolve ambiguous mood prompts into more conventional, obviously emotional cues (swelling strings, minor-key stabs) that would announce the mood instead of letting it sit underneath a scene quietly. The explicit looping requirement in the runtime note matters practically for game audio and installations specifically, where a bed that builds to an ending and stops is unusable — a mostly static loop with one subtle swell gives just enough movement to avoid sounding robotic on repeat without breaking the seamless-loop requirement.