A

ACE-Step 1.5

ACE Studio · Released Jan 2026

Conditional

ACE-Step 1.5 is an MIT-licensed open-source music generation model, co-led by ACE Studio and StepFun, that makes full songs with vocals from lyrics and a style caption: a diffusion transformer renders the audio and an optional language model plans the song. In our runs it made usable guide raps on a 6 GB RTX 3060 Laptop GPU with no language model, INT8 weights and CPU offload. Conditional because quality swings widely from seed to seed and it adds ad-libs and repeats or drops hook lines.

Is it right for you?

Good for

  • Generating a full song with vocals from lyrics plus a style caption, from 10 seconds up to 10 minutes long, on your own GPU
  • Low-VRAM local generation: the project's 6 GB tier (2B turbo DiT, no language model, INT8 weights, full CPU offload) ran in our tests on an RTX 3060 Laptop GPU
  • A guide vocal for singing voice conversion: ask for a solo, dry, clear-diction vocal and convert it to your voice
  • Reproducible drafts: the same seed and words gave a byte-identical song in our runs
  • Commercial and modified use under the MIT license, with LoRA training, covers and repainting in the same project

Not good for

  • One-shot results: quality varied from clean to unusable across seeds on the same words and settings, so plan to generate and score several
  • Exact lyric structure: it adds ad-libs and repeats or drops hook lines
  • Running the language model planner on 6 GB: the 1.7B planner does not fit beside the DiT, and it rewrites your caption and lyrics

Pricing

Input

Free (self-hosted open weights)

Output

Free (self-hosted open weights)

Context

Runs locally, 6 GB GPU tested (no LM); MIT

View full pricing

No verdict changes yet

The clock starts day one. Changes land here as our verdict evolves.

Verification log

No verification checks yet

We haven't logged a verification check for this entry. Once a check runs, its history shows here.

How we evaluate