ACE-Step 1.5

### TL;DR

ACE-Step 1.5 is an open-source music generation model designed to produce high-quality songs efficiently on local hardware. It integrates diffusion-based generation with a Deep Compression AutoEncoder and a lightweight linear transformer, enabling rapid synthesis of up to 4-minute tracks in approximately 20 seconds on an A100 GPU.

Key Insights & Metrics

Key Features

• Fast music generation: Synthesizes up to 4-minute songs in about 20 seconds on an A100 GPU.
• High-quality output: Achieves superior musical coherence and lyric alignment across melody, harmony, and rhythm metrics.
• Local operation: Runs efficiently on consumer hardware, with generation times around 10 seconds on an RTX 3090.
• User fine-tuning: Supports LoRA fine-tuning, allowing users to train the model with their own music for personalized outputs.
• Versatile capabilities: Offers full song generation, song covers, repainting, stem addition, generative stem splitting, and song completion.

→ Related Releases

Discussion