FlopCoin generates audio the same way it generates images: you describe what you want and the network renders it. The difference is that a song has two separate halves, and telling them apart is most of the skill.
Style and lyrics are two different boxes
The style is the sound: the genre, the instruments, the mood, the way it was recorded. Something like "folk ballad, fingerpicked acoustic guitar, close mic, sparse". The lyrics are the words that get sung.
You can give either one on its own:
- Style only gives you an instrumental in that style.
- Lyrics only gives you a sung track, with the model choosing its own backing.
- Both gives you those words sung in that style. This is usually what you want.
The one thing that does not work is pasting your lyrics into the style box. The model reads that box as production direction, so a verse dropped there is treated as a description of a mood and never actually gets sung.
Length
Tracks can run from 5 seconds to 4 minutes, and the default is 2 minutes. Longer tracks cost more to render and take longer to come back, so if you are still trying out ideas, generate short and only stretch it out once the style is right.
Getting the same take twice
Every generation uses a seed. Keep the prompt, the lyrics, the settings and the seed the same and you get the same track back. Change the seed and you get a different performance of the same idea. That is the knob to reach for when a track is nearly right but you want to hear another attempt.
Can I sell what I make?
For the default music model, yes. It runs on Apache-2.0 licensed weights, which permit commercial use.
Some other audio models on the network run on weights whose licence forbids commercial use. Those are marked non-commercial in the model catalog, and that marking is the authoritative answer. If you plan to sell, stream or advertise with what you generate, check that the model you picked is not one of them.
Developers can read the same information from the API - see generating audio through the API.