DiffRhythm · base
Generate vocals and accompaniment conditioned on timed lyrics and a style reference.
GenerationOpen-source code
- Start here
- Prepare timed lyrics before judging a take. Check vocal timing and phrase transitions as carefully as the overall sound.
- Training data
- The base checkpoint card does not provide a complete, itemized training-data inventory.
- Training rights
- A publicly downloadable checkpoint does not establish source-music permissions. Check the upstream license and each reference recording.
- License & output terms
- Code and DiT weights: Apache 2.0, per the card. Inspect the separate VAE and conditioning dependencies.
- Limitations
- Timed lyrics and reference preparation affect results. Language coverage, pronunciation and long-form coherence vary.
- Technique
- Latent diffusion / flow matching
- Runtime
- Local GPU workflow with separate conditioning and audio-decoder dependencies.
- Parameters
- Not itemized for this release/family. Parameter sizes shown elsewhere may exclude encoders, decoders or other runtime components.
Sources reviewed . Training and rights notes summarize provider disclosures. A model license does not clear every source recording or generated output. Parameter counts refer to the named variant or component, not total runtime memory.