Use official ACE-Step model downloader for automatic model downloads

README.md

@@ -13,7 +13,9 @@ python_version: 3.11
 
 # ACE-Step 1.5 Custom Edition
 
-A comprehensive music generation system built on ACE-Step 1.5, featuring:
+A comprehensive music generation system built on ACE-Step 1.5, featuring three specialized interfaces.
+
+**Models will download automatically on first run (~7GB)** from the official ACE-Step repository.
 
 ## 🌟 Features
 
@@ -42,6 +44,36 @@ Complete training interface for custom models:
 
 ## 🚀 Quick Start
 
+1. **First Launch**: Models download automatically from `ACE-Step/Ace-Step1.5`
+2. **Standard Generation**: Use Tab 1 for traditional text-to-music
+3. **Timeline Creation**: Use Tab 2 to build longer songs with consistent style
+4. **Custom Training**: Use Tab 3 to create specialized models
+
+## 📚 About
+Full-featured standard ACE-Step 1.5 GUI with all original capabilities including:
+- Text-to-music generation with style control
+- Variation generation
+- Section repainting
+- Lyric editing
+
+### 2. Custom Timeline Workflow
+Advanced timeline-based generation system:
+- Generate 32-second clips with seamless blending
+- Adjustable context length (0-120 seconds) for style consistency
+- Master timeline with visual representation
+- Extend, inpaint, and remix capabilities
+- Automatic crossfading between clips
+
+### 3. LoRA Training Studio
+Complete training interface for custom models:
+- Upload and preprocess audio files
+- Configure training parameters
+- Train specialized models for voices, instruments, or styles
+- Download and reuse trained models
+- Continue training from existing LoRAs
+
+## 🚀 Quick Start
+
 1. **Standard Generation**: Use Tab 1 for traditional text-to-music
 2. **Timeline Creation**: Use Tab 2 to build longer songs with consistent style
 3. **Custom Training**: Use Tab 3 to create specialized models

README_HF.md

@@ -13,7 +13,9 @@ python_version: 3.11
 
 # ACE-Step 1.5 Custom Edition
 
-A comprehensive music generation system built on ACE-Step 1.5, featuring:
+A comprehensive music generation system built on ACE-Step 1.5, featuring three specialized interfaces.
+
+**Models will download automatically on first run (~7GB)** from the official ACE-Step repository.
 
 ## 🌟 Features
 
@@ -42,6 +44,36 @@ Complete training interface for custom models:
 
 ## 🚀 Quick Start
 
+1. **First Launch**: Models download automatically from `ACE-Step/Ace-Step1.5`
+2. **Standard Generation**: Use Tab 1 for traditional text-to-music
+3. **Timeline Creation**: Use Tab 2 to build longer songs with consistent style
+4. **Custom Training**: Use Tab 3 to create specialized models
+
+## 📚 About
+Full-featured standard ACE-Step 1.5 GUI with all original capabilities including:
+- Text-to-music generation with style control
+- Variation generation
+- Section repainting
+- Lyric editing
+
+### 2. Custom Timeline Workflow
+Advanced timeline-based generation system:
+- Generate 32-second clips with seamless blending
+- Adjustable context length (0-120 seconds) for style consistency
+- Master timeline with visual representation
+- Extend, inpaint, and remix capabilities
+- Automatic crossfading between clips
+
+### 3. LoRA Training Studio
+Complete training interface for custom models:
+- Upload and preprocess audio files
+- Configure training parameters
+- Train specialized models for voices, instruments, or styles
+- Download and reuse trained models
+- Continue training from existing LoRAs
+
+## 🚀 Quick Start
+
 1. **Standard Generation**: Use Tab 1 for traditional text-to-music
 2. **Timeline Creation**: Use Tab 2 to build longer songs with consistent style
 3. **Custom Training**: Use Tab 3 to create specialized models

app.py

@@ -348,6 +348,8 @@ def create_ui():
         # 🎵 ACE-Step 1.5 Custom Edition
         
         **Three powerful interfaces for music generation and training**
+        
+        Models will download automatically on first use (~7GB from HuggingFace)
         """)
         
         with gr.Tabs():

src/ace_step_engine.py

@@ -8,7 +8,6 @@ from pathlib import Path
 import logging
 from typing import Optional, Dict, Any, Tuple
 import os
-from huggingface_hub import snapshot_download
 
 logger = logging.getLogger(__name__)
 
@@ -17,6 +16,7 @@ try:
     from acestep.handler import AceStepHandler
     from acestep.llm_inference import LLMHandler
     from acestep.inference import GenerationParams, GenerationConfig, generate_music
+    from acestep.model_downloader import ensure_main_model, get_checkpoints_dir, check_main_model_exists
     ACE_STEP_AVAILABLE = True
 except ImportError as e:
     logger.warning(f"ACE-Step 1.5 modules not available: {e}")
@@ -64,57 +64,35 @@ class ACEStepEngine:
 
     def _download_checkpoints(self):
         """Download model checkpoints from HuggingFace if not present."""
-        checkpoint_dir = self.config.get("checkpoint_dir", "./checkpoints")
+        checkpoints_dir = get_checkpoints_dir(self.config.get("checkpoint_dir"))
         
-        if os.path.exists(checkpoint_dir) and os.listdir(checkpoint_dir):
-            logger.info(f"Checkpoints already exist at {checkpoint_dir}")
+        # Check if main model already exists
+        if check_main_model_exists(checkpoints_dir):
+            logger.info(f"✓ ACE-Step 1.5 models already exist at {checkpoints_dir}")
             return
             
-        logger.info("Downloading ACE-Step 1.5 model checkpoints...")
+        logger.info("Downloading ACE-Step 1.5 models from HuggingFace...")
+        logger.info("This may take several minutes (models are ~7GB total)...")
         
         try:
-            # Download DiT model
-            dit_model = self.config.get("dit_model_path", "acestep-v15-turbo")
-            snapshot_download(
-                repo_id=f"ACE-Step/{dit_model}",
-                local_dir=os.path.join(checkpoint_dir, dit_model),
-                repo_type="model"
+            # Use the built-in model downloader
+            success, message = ensure_main_model(
+                checkpoints_dir=str(checkpoints_dir),
+                prefer_source="huggingface"  # Use HuggingFace for Spaces
             )
-            logger.info(f" Downloaded DiT model: {dit_model}")
             
-            # Download VAE
-            snapshot_download(
-                repo_id="ACE-Step/vae",
-                local_dir=os.path.join(checkpoint_dir, "vae"),
-                repo_type="model"
-            )
-            logger.info(" Downloaded VAE")
-            
-            # Download text encoder
-            snapshot_download(
-                repo_id="Qwen/Qwen3-Embedding-0.6B",
-                local_dir=os.path.join(checkpoint_dir, "Qwen3-Embedding-0.6B"),
-                repo_type="model"
-            )
-            logger.info(" Downloaded text encoder")
+            if not success:
+                raise RuntimeError(f"Failed to download models: {message}")
             
-            # Download LLM
-            lm_model = self.config.get("lm_model_path", "acestep-5Hz-lm-1.7B")
-            snapshot_download(
-                repo_id=f"ACE-Step/{lm_model}",
-                local_dir=os.path.join(checkpoint_dir, lm_model),
-                repo_type="model"
-            )
-            logger.info(f" Downloaded LLM: {lm_model}")
-            
-            logger.info(f" All checkpoints downloaded to {checkpoint_dir}")
+            logger.info(f"✓ {message}")
+            logger.info("✓ All ACE-Step 1.5 models downloaded successfully")
             
         except Exception as e:
             logger.error(f"Failed to download checkpoints: {e}")
             raise
 
     def _load_models(self):
-        """Initialize ACE-Step 1.5 models."""
+        """Initialize s_dir = get_checkpoints_dir(self.config.get("checkpoint_dir")
         try:
             if not ACE_STEP_AVAILABLE:
                 raise RuntimeError("ACE-Step 1.5 not available")
@@ -126,7 +104,7 @@ class ACEStepEngine:
             # Get project root
             project_root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
             
-            logger.info(f"Initializing DiT handler with model: {dit_model_path}")
+            logger.info(f"Instr(checkpoints_dir / dit_model_path) handler with model: {dit_model_path}")
             
             # Initialize DiT handler (handles main diffusion model, VAE, text encoder)
             status_dit, success_dit = self.dit_handler.initialize_service(
@@ -142,7 +120,7 @@ class ACEStepEngine:
                 raise RuntimeError(f"Failed to initialize DiT: {status_dit}")
             
             logger.info(f" DiT initialized: {status_dit}")
-            
+            str(checkpoints_dir)
             # Initialize LLM handler (handles 5Hz Language Model)
             logger.info(f"Initializing LLM handler with model: {lm_model_path}")