ACE-Step 1.5
Explore text-conditioned songs, covers, repainting and other editing tasks with base, SFT and turbo variants.
GenerationOpen-source code
- Start here
- Compare base, SFT and turbo on one brief. For fine-tuning, inspect the base variant; for quick iteration, examine the turbo workflow.
- Training data
- The authors describe professionally licensed music plus public-domain and royalty-free sources, without an itemized recording inventory in the linked card.
- Training rights
- These are the authors' provenance claims, not an independently verified clearance. Reference recordings and lyrics retain their own rights.
- License & output terms
- Linked 1.5 release: MIT. This differs from the earlier v1 release's Apache 2.0 declaration.
- Limitations
- Capabilities and fine-tuning support vary across base, SFT and turbo models. Do not infer exact editing control from a good sample or apply v1 benchmarks to 1.5.
- Technique
- Language-model planner + diffusion transformer
- Runtime
- Local inference; DiT and language-model selection change memory needs and generation speed.
- 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.