Feat: Add VoxCPM 1.5 Node Skeleton

2026-01-04 15:34:42 +08:00
parent 6cfa1274e2
commit d0d652446b
2 changed files with 95 additions and 0 deletions
@@ -112,6 +112,9 @@ else:
    # 17. 处理 aiia_vibevoice_preset_maker.py
    _load_nodes_from_module(".aiia_vibevoice_preset_maker", "aiia_vibevoice_preset_maker")

+    # 18. 处理 aiia_voxcpm_nodes.py (新增 VoxCPM 1.5)
+    _load_nodes_from_module(".aiia_voxcpm_nodes", "aiia_voxcpm_nodes")
+
    # 告诉 ComfyUI 这个节点包有一个包含网页资源的 'js' 目录
    WEB_DIRECTORY = "js"

@@ -0,0 +1,92 @@
+import os
+import sys
+import torch
+import torchaudio
+import numpy as np
+import folder_paths
+from tqdm import tqdm
+
+class AIIA_VoxCPM_Loader:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "model_name": (["VoxCPM-1.5-800M"], {"default": "VoxCPM-1.5-800M"}),
+                "precision": (["fp16", "fp32"], {"default": "fp16"}),
+            }
+        }
+
+    RETURN_TYPES = ("VOXCPM_MODEL",)
+    RETURN_NAMES = ("voxcpm_model",)
+    FUNCTION = "load_model"
+    CATEGORY = "AIIA/VoxCPM"
+
+    def load_model(self, model_name, precision):
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        dtype = torch.float16 if precision == "fp16" else torch.float32
+        
+        # Define paths
+        hf_repo_id = "openbmb/VoxCPM-1.5"
+        base_path = os.path.dirname(os.path.abspath(__file__))
+        models_dir = os.path.join(base_path, "models", "voxcpm")
+        model_path = os.path.join(models_dir, "VoxCPM-1.5")
+        
+        # Download if not exists
+        if not os.path.exists(model_path):
+            print(f"[AIIA] Downloading VoxCPM 1.5 to {model_path}...")
+            try:
+                from huggingface_hub import snapshot_download
+                snapshot_download(repo_id=hf_repo_id, local_dir=model_path, local_dir_use_symlinks=False)
+            except ImportError:
+                raise ImportError("Please install huggingface_hub to download models automatically.")
+            except Exception as e:
+                raise RuntimeError(f"Failed to download model: {e}")
+
+        print(f"[AIIA] Loading VoxCPM from {model_path}...")
+        
+        # TODO: Actual Model Loading Logic
+        # We need to determine if we use 'transformers' or a custom 'voxcpm' package.
+        # For now, we'll placeholder the object.
+        model = {"path": model_path, "device": device, "dtype": dtype}
+        
+        return (model,)
+
+class AIIA_VoxCPM_TTS:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "voxcpm_model": ("VOXCPM_MODEL",),
+                "text": ("STRING", {"multiline": True, "default": "Hello, world."}),
+                "reference_audio": ("AUDIO",), # VoxCPM is Zero-Shot, so this is required for cloning
+                "speed": ("FLOAT", {"default": 1.0, "min": 0.5, "max": 2.0, "step": 0.1}),
+                "top_p": ("FLOAT", {"default": 0.8, "min": 0.1, "max": 1.0, "step": 0.05}),
+                "temperature": ("FLOAT", {"default": 0.8, "min": 0.1, "max": 2.0, "step": 0.05}),
+            }
+        }
+
+    RETURN_TYPES = ("AUDIO",)
+    RETURN_NAMES = ("audio",)
+    FUNCTION = "generate"
+    CATEGORY = "AIIA/VoxCPM"
+
+    def generate(self, voxcpm_model, text, reference_audio, speed, top_p, temperature):
+        # Placeholder generation logic
+        print(f"[AIIA] Generating VoxCPM TTS for text: {text[:20]}...")
+        
+        # Mock output for now
+        sr = 44100
+        duration_sec = 2
+        waveform = torch.zeros((1, int(sr * duration_sec)))
+        
+        return ({"waveform": waveform, "sample_rate": sr},)
+
+NODE_CLASS_MAPPINGS = {
+    "AIIA_VoxCPM_Loader": AIIA_VoxCPM_Loader,
+    "AIIA_VoxCPM_TTS": AIIA_VoxCPM_TTS
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "AIIA_VoxCPM_Loader": "🎤 VoxCPM Loader",
+    "AIIA_VoxCPM_TTS": "🗣️ VoxCPM 1.5 TTS"
+}