I needed a 30–60s bed for a product demo. Budget: basically zero. Skill level: I can edit audio, I can’t arrange a band.

So I treated AI music generator tooling like any other API-ish product: input quality in, usable asset out, iterate when it misses.

What I tried

I used Doremi as the workspace. The useful bit for me wasn’t a single model—it was being able to compare Suno / Mureka / ElevenLabs on the same prompt, then keep the take that matched the cut.

Workflow that worked:

Write a prompt like a ticket: genre + mood + instruments + vocal yes/no + rough length.

Generate 2–3 variants.

Extend or remix instead of re-rolling from scratch when the vibe is close.

Export and drop into the timeline; adjust prompt only if the sync is wrong.

When I wanted the dedicated generate UI (model picker + prompt/lyrics), I used their create-music flow.

Takeaways for builders

Prompt like you’re writing acceptance criteria.…