The key to great results is a descriptive and detailed prompt. The model understands nuance, so the more information you provide, the closer the output will be to your vision. Here are some best practices:
-
Be Specific with Genre and Style: Instead of
rock music, tryenergetic 1980s synth-pop with a driving drum machine beat and male vocals. -
Layer Multiple Descriptors: Combine mood, instrumentation, tempo, and use case.
- Example:
A slow, melancholic piano melody over ambient synth textures, suitable for a tragic film scene.
- Example:
-
Define Instrumentation: Call out the specific instruments you want to hear.
- Example:
Upbeat funk track with a prominent slap bass line, funky rhythm guitar, and a horn section.
- Example:
- Use the "Include/Exclude Styles" Feature: Refine your output by explicitly including or excluding certain tags like acoustic, repetitive structure, or four-on-the-floor kick.
- Build Section by Section: To have the most control when creating a full song, generate the Intro first. Once you're happy with that first section, you can click the "+" sign and specify the style for the next part. Then, use the "Continue the conversation..." prompt box to generate the Main Groove or Chorus, building your track piece by piece.
- Iterate and Refine: If the first generation isn't perfect, don't start over! Adjust your prompt and regenerate. Small changes can have a big impact.
For a complete list of tips and examples, please see our official Prompting Guide.