You paste lyrics and a style prompt. DiffRhythm spits out a complete song—vocals and instrumentals in one go. No bouncing between tools for backing tracks and vocals. No stitching files together afterward.
The tech? Latent diffusion applied to full song generation. Vocals and accompaniment get synthesized simultaneously (not layered in stages like most AI music tools require). Generation happens fast. The architecture isn't autoregressive—you're not waiting around while it builds your track piece by piece. For podcasters needing custom intro jingles or YouTubers churning out videos with original music, that speed matters.
Songs can run up to four minutes on the free plan. Paid tiers unlock eight-minute generation and access to newer model versions beyond V1. Style control works through text prompts—pop, jazz, rock, ballad. Want instrumentals only? That works too.
Once you've got your track, download it as WAV or MP3. Extract stems if you need to isolate parts. Remove vocals entirely.
The free plan gives you eight songs total—enough to test the waters but not much else. After that you're looking at $6.99 monthly for the Starter plan (roughly 18,000 credits) or $59 monthly for unlimited generation. Concurrent generation limits kick in too. Free users get one at a time.
Commercial licensing comes included with paid plans. You can make your generated songs public or keep them private.
