This is a well-known frustration with AI music generators. The core problem is that prompting "no drums" or "exclude percussion" doesn't work reliably because the model focuses on the keyword "drums" rather than the negation, and often generates percussion anyway. Here are a few practical approaches you can try:
1. Reframe your prompt positively (avoid negations)
Instead of saying "no drums," describe
only what you want. In the Studio Context Bar's style/description field, try something like:
"Solo reggae skank organ, offbeat staccato chords, Hammond organ, clean, sparse, rhythmic keyboard only"
Emphasizing "solo" and "only" while naming the specific instrument gives Suno a clearer target. Avoid mentioning drums or percussion at all — even to exclude them.
2. Use stem separation after generating
If the generation still includes drums, you can split the generated audio into stems within Studio and isolate the keyboard/instrument part. In Studio, select the generated clip and extract stems. Then keep the keys/other stem and delete or mute the drum stem. The separation isn't always perfect, but it's often good enough for a clean organ part.
3. Try Suno Sounds for a loop
Suno Sounds is an experimental feature that lets you generate individual instrument samples and loops using text prompts. In Custom mode, select "Sounds" from the dropdown, choose "Loop," set your BPM and key, and prompt something like:
"Reggae skank organ, offbeat staccato Hammond chords"
This is designed for isolated instrument parts, so you're far less likely to get unwanted drums.
4. Generate, then extract and iterate
You can highlight a section on the timeline and ask Studio to generate a specific instrumental part over it, so if one generation is close but has light percussion bleed, you can extract the MIDI from the organ stem and use that as a starting point for a cleaner re-generation.
The Sounds loop approach is probably your most direct path to a clean, drum-free skank organ part. If that doesn't get you exactly what you need, generating a full part and then stem-separating is the reliable fallback.