ACE-Step 1.5
A highly efficient open-source music foundation model combining a planning Language Model with a Diffusion Transformer for rapid, high-quality track generation.
Overview
What is ACE-Step 1.5
ACE-Step 1.5 is a state-of-the-art open-source music generation model released in January 2026 by ACE Studio and StepFun. It uses a hybrid architecture—a planning Language Model that drafts song blueprints via Chain-of-Thought, guiding a 3.5-billion-parameter Diffusion Transformer to synthesize full tracks with vocals and instrumentals.
Running it privately on Venice
On Venice, you can generate high-quality audio tracks with ACE-Step 1.5 under our anonymized privacy tier, ensuring your creative prompts and lyrics are never stored or used for profiling. While Venice routes requests through anonymized third-party inference providers, we strip all identifying metadata, giving you a permissionless, zero-retention environment to compose music without surveillance.
Assessment
Strengths and limitations
- Extremely fast inference, generating a full song in under 2 seconds on an A100 GPU or under 10 seconds on consumer hardware (RTX 3090).
- Highly efficient local footprint, requiring less than 4GB of VRAM to run, making commercial-grade music AI accessible on consumer PCs.
- Hybrid architecture uses Chain-of-Thought planning to align lyrics, metadata, and composition before diffusion, improving prompt adherence.
- Supports lightweight personalization, allowing users to train custom LoRAs from just a few songs to capture specific musical styles.
- Versatile audio editing capabilities, including cover generation, audio repainting, and vocal-to-BGM conversion.
- Can introduce high-frequency artifacts, particularly in vocal formants, if diffusion step counts are set too low.
- Standard 3.5B model is optimized for speed; the higher-fidelity 4B XL variant requires significantly more VRAM (12GB+).
- Third-party hosted on Venice, meaning requests are anonymized but processed off-infrastructure, unlike Venice's locally hosted TEE models.
Samples
Sample outputs
Generated on Venice with our standard prompt suite — the same prompts we run through every model of this type, so you can judge it like-for-like.
“An uplifting cinematic orchestral build with soaring strings, warm brass, and a hopeful resolution.”
“A mellow lo-fi hip-hop beat with a soft jazzy piano loop, vinyl crackle, and a relaxed late-night mood.”
Specifications
Datasheet
- Maker
- ACE Studio & StepFun
- Released
- January 28, 2026
- Modality
- Text-to-music, audio editing
- Architecture
- Hybrid (LM Planner + 3.5B Parameter Diffusion Transformer)
- VRAM Requirement
- <4GB VRAM (for local deployment)
- Supported Languages
- 50+ languages
- License
- MIT License (Open Source)
- Privacy on Venice
- Anonymized — prompts not stored
- Available on Venice since
- Feb 2026
API
Call it from your code
Venice exposes this model through the REST API. Queue a generation with the model id.
curl https://api.venice.ai/api/v1/audio/queue \
-H "Authorization: Bearer $VENICE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "ace-step-15",
"prompt": "An uplifting cinematic orchestral build with soaring strings"
}'
# Use the returned queue_id with https://api.venice.ai/api/v1/audio/retrieve.
# Call /audio/complete after downloading if needed.Pricing
What it costs on Venice
Pay per track on Venice — price scales with length, from $0.03.
New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.
Alternatives
How it compares
| Model | Best for | Architecture | Open weights | VRAM (Local) | Price (Venice) |
|---|---|---|---|---|---|
| ACE-Step 1.5 | Fast, highly efficient open-source music model with strong editing features. | Hybrid LM + 3.5B DiT | No | <4GB | from $0.03 / track |
| MiniMax Music 2.5 | High-fidelity commercial alternative, but significantly more expensive per track. | Proprietary | No | N/A (API only) | $0.24 / track |
| ElevenLabs Music | Premium audio quality and vocals, but carries a high cost per generation. | Proprietary | No | N/A (API only) | from $0.87 / track |
| Lyria 3 Pro | Google-backed music generation model, closed-source. | Proprietary | No | N/A (API only) | $0.10 / track |
Fast, highly efficient open-source music model with strong editing features.
Use cases
What it is good for
- 01Rapid prototyping of background tracks and BGM for content creators and indie game developers.
- 02Creating custom vocal and instrumental tracks in over 50 languages from simple text prompts.
- 03Fine-tuning personalized music styles using LoRAs for consistent artist workflows.
- 04Audio editing tasks like repainting sections of a track or generating cover variations.
Prompting
Getting better results
Provide clear genre, tempo, and mood descriptors to help the LM planner draft a strong song blueprint.
Include structured lyrics or specify 'instrumental' if you want to skip vocal generation entirely.
For local generation, keep VRAM low by using the standard 3.5B model before scaling up to the XL version.
Version history
Initial open-source release.
Current standard model with hybrid LM planner and 3.5B DiT.
Upgraded variant featuring a 4B parameter DiT decoder.
FAQ
Frequently asked questions
ACE-Step 1.5 is an open-source music foundation model developed by ACE Studio and StepFun. Released in January 2026, it uses a hybrid architecture combining a Language Model planner and a 3.5-billion-parameter Diffusion Transformer to generate high-quality tracks with vocals and instrumentals.
On Venice, ACE-Step 1.5 is highly affordable, starting from $0.03 per track. Pricing scales with track length: a 60-second clip costs $0.03, while a 210-second clip costs $0.08.
Yes, the model weights are open-source and released under the permissive MIT license. You can download and run it locally for free, or use Venice's hosted API for convenient, pay-per-track generation.
ACE-Step 1.5 is fully open-source, runs locally on consumer hardware (<4GB VRAM), and costs from $0.03/track on Venice. ElevenLabs Music is a closed, proprietary model that starts at a much higher price of $0.87/track, though it offers highly polished commercial vocals.
Yes. Beyond text-to-music generation, ACE-Step 1.5 supports advanced editing features such as cover generation, audio repainting, and vocal-to-BGM conversion.
Venice processes your requests under an anonymized privacy tier. Your prompts, lyrics, and generated audio are never stored, profiled, or used to train models. Note that inference is served via a third-party provider, but Venice strips all identifying data before forwarding.
The model maintains strict prompt and lyric adherence across more than 50 languages, making it highly versatile for global music production.
Run ACE-Step 1.5 privately
No prompt logging. No data used for training.