ACE-Step 1.5

• 🎵 Open-source music foundation model generating full songs from text
• 🆕 Hybrid design: language-model planner guides a Diffusion Transformer
• 🧠 Planner uses chain-of-thought reasoning to draft song blueprints
• 🌐 Multilingual lyric and prompt support
• 🔧 Unifies cover generation, repainting, and vocal-to-BGM editing
• 🎯 Intrinsic reinforcement learning aligns without external reward models
• 🏢 Co-led by ACE Studio and StepFun
• 💬 Accepts optional lyrics with structured song sections

ACE Studio

ACE Studio is an AI lab focused on music generation, developing models that produce audio content from text-based inputs. The organization has carved out a niche in the emerging field of AI-composed music, building dedicated architectures for high-quality…

ACE-Step 1.5 is an open-source music generation model that turns text descriptions—optionally paired with lyrics—into complete songs with melody, harmony, rhythm, instrumentation, and vocals. Released in 2026, its authors describe it as a music foundation model co-led by ACE Studio and StepFun, aiming to bring high-quality generation to a single accessible framework.

Architecturally, version 1.5 advances the original ACE-Step by introducing a hybrid design. A language model acts as an "omni-capable planner," using chain-of-thought reasoning to draft song blueprints, metadata, and lyrics that then guide a Diffusion Transformer, rather than relying on a diffusion generator alone. This planner-plus-generator structure is the central change the paper presents over the earlier single-stage approach.

Beyond text-to-music, ACE-Step 1.5 unifies editing tasks within one framework—cover generation, repainting, and vocal-to-background-music conversion—alongside controllable song generation. For alignment, the authors describe an intrinsic reinforcement learning method derived from the model's own comprehension tasks, which they say avoids biases introduced by external reward models and improves multilingual lyric compliance.

As an open release positioned for local use, ACE-Step 1.5 is distributed with documentation and code, and the model is framed by its authors as pushing the boundaries of open-source music generation. Because primary and peer-reviewed details here come from the project's technical paper, specific deployment figures and language counts circulated elsewhere are best confirmed against that source.