Sign in
Sign in to your account
No password. We'll never spam you.
Free to start: try up to 2 songs/day. No credit card required.
Run the full ACE-Step 1.5 stack online — describe your track or paste lyrics, then hit generate. No install, no GPU, no waitlist.
Free to start. Try up to 2 songs/day. No credit card required.
WHY ACESTEP.IO
ACE-Step 1.5 is a powerful open-source music model — but running it locally means managing Python environments, multi-gigabyte checkpoints and 12 GB+ of VRAM. We do the heavy lifting for you, so every ACE-Step 1.5 feature is a single click away.
Speed
acestep.io runs the ACE-Step 1.5 Turbo and XL pipelines on cloud GPUs, so a full song renders in seconds instead of minutes. Skip CUDA setup, model downloads, and VRAM tuning — just type a prompt and press Generate.
Capabilities
Text-to-music, lyrics-to-song, cover generation, repaint, reference audio, vocal-to-BGM, stem separation and BPM/key extraction — every core ACE-Step 1.5 feature is exposed through a friendly web UI.
Output
Powered by the 2B and 4B XL DiT decoders, our ACE-Step 1.5 generator delivers clean vocals, tight mixes, and musically coherent arrangements that sit comfortably between Suno v4.5 and v5 in blind tests.
Coverage
Write lyrics in English, Chinese, Japanese, Korean, Spanish, Arabic and dozens more. Pair them with over 1,000 instruments and style tags to cover any genre from lo-fi to symphonic metal.
ACE-Step 1.5 is the latest open-source music foundation model from ACE Studio and StepFun. It pairs a Language Model planner with a Diffusion Transformer (DiT) decoder, so short prompts expand into structured song blueprints — lyrics, sections, metadata and all — that drive high-fidelity audio synthesis.
The LM acts as an omni planner (query rewriting, Chain-of-Thought metadata, lyric alignment) while the DiT handles raw waveform generation. Intrinsic reinforcement learning keeps both halves aligned without human preference data, producing stable output across tempos and genres.
Running ACE-Step 1.5 locally needs Python 3.11+, a capable CUDA or ROCm GPU, and multi-gigabyte model downloads. acestep.io hosts the whole stack — 2B turbo/sft, XL 4B, and all 5Hz LM tiers — so you can try every model with one click from any browser.
ACE-Step 1.5 is an open-source AI music foundation model that combines a Language Model planner with a Diffusion Transformer audio decoder. It can generate full songs with vocals from a text prompt, turn lyrics into melodies, and perform editing tasks such as cover, repaint and vocal-to-BGM.
Version 1.5 introduces the hybrid LM + DiT design, the 4B XL decoder for higher fidelity, faster turbo scheduling, native reference-audio conditioning and unified tools for cover, repaint, stem separation and LRC generation — all of which are available on acestep.io.
No. acestep.io hosts the full ACE-Step 1.5 stack on cloud GPUs. You only need a modern browser. Local installation is only necessary if you want to fine-tune your own LoRA or host the model offline.
Yes. Paste structured lyrics (verse / chorus / bridge) directly into the Create page. You can also let the ACE-Step 1.5 LM planner auto-write lyrics from a one-line idea if you prefer.
Every generation is exportable as MP3 or WAV. Studio projects additionally expose LRC lyric timestamps and stem files when you enable the separation workflow.
ACE-Step 1.5 supports durations from 10 seconds up to 10 minutes in a single pass. Longer arrangements can be built by stitching repaint or lego segments in the Studio.
Yes. All audio you generate on acestep.io is royalty-free and cleared for personal and commercial projects. You are responsible for the lyrics and references you supply.
Launch the Create studio and have your first ACE-Step 1.5 song — vocals, lyrics, mix and all — ready for download in under a minute.