Local AI RecipesLocal music generation

ACE-Step 1.5 vs YuE2-3B: local music recipe

A practical split between a commercial-friendly product path and a more experimental YuE2 path with model-weight licensing limits for companies.

ACE-Step product path · YuE2 research pathReviewed 2026-09-26
Commercial statusACE-Step 1.5 is MIT. YuE2 code is Apache-2.0, while model weights are CC BY-NC 4.0 with additional creator permission; companies need a commercial model license.
HardwareUse the model/runtime combination that fits your GPU. Consumer GPUs favor smaller or quantized local paths; larger high-quality stacks need more VRAM or offload.
RuntimeACE-Step 1.5 is the current AILocalabs commercial product path. YuE2 is worth testing separately for symbolic planning, covers and editing.

For a company shipping a paid local music application, licensing is part of the engineering decision, not a footnote.

AILocalabs currently prefers ACE-Step 1.5 for the commercial MusicKiln path. YuE2 is technically interesting, but its model-weight terms make it a separate research/creator recipe unless a company obtains a commercial license.

DIY recipe

Build it yourself

  1. Start with ACE-Step when you want a commercial-friendly local baseline and a straightforward product path.
  2. Try YuE2 when its planning, cover or agentic editing workflow is the actual reason for the experiment.
  3. Benchmark generation time, VRAM, prompt adherence and output quality on the same prompts before switching a production backend.
  4. Pin the exact model and runtime revision; music stacks change quickly and a “same model name” benchmark can become meaningless.
Risks

What can go wrong

  • Do not confuse an open-source code license with the model-weight license.
  • Output copyright/style questions remain separate from the software license.
  • A better demo is not enough reason to migrate a production product without regression and packaging tests.
AILocalabs decision

What we ship — and what we do not

MusicKiln follows the ACE-Step commercial path. We publish YuE2 as a comparison/research route rather than silently bundling it.

Recipe discussion

Tried it? Leave results, questions and improvements.

Discussion stays with this recipe so future builders can learn from it. Never post passwords, API keys, license keys or private files.

Loading discussion…

Join the discussion