Why Was My Prompt Flagged? Understanding Content Moderation in Stable Audio

Summary

If a prompt you submit gets flagged or blocked, it's usually because it named a specific real person — an artist, musician, public figure, or other named individual — as a reference for style or inspiration. This is expected behavior, not a bug: it's part of how Stable Audio's content moderation harness protects against generating audio that mimics or impersonates a real person's voice, style, or likeness without consent.

Why this happens

Stable Audio's moderation system automatically screens prompt text before generation begins. When a prompt includes the name of a real, identifiable individual — for example, "in the style of [artist name]" — the system may flag or block the request. This is an intentional safeguard, not a harness bug, and applies even when the intent is purely creative (e.g., "for inspiration" rather than direct imitation).

This protects against:

• Style cloning of real people
• Misattribution of AI-generated audio to a real artist
• Rights and likeness issues tied to naming specific individuals in generated content

What to do instead

Rather than naming a specific person, describe the qualities you're going for. For example, instead of "sounds like [artist name]," try describing the genre, era, instrumentation, tempo, or production style you have in mind (e.g., "moody 1980s synth-pop with a breathy female vocal and analog drum machine"). This gives the model the creative direction you're after without triggering the moderation filter.