Fix VibeVoice load path cleanup bug & update README file structure

This commit is contained in:
Hawk Lee
2025-12-29 15:45:16 +08:00
parent e8de3283b3
commit d76d2712e6
2 changed files with 8 additions and 5 deletions
+2 -1
View File
@@ -443,7 +443,6 @@ git clone https://github.com/havvk/ComfyUI_AIIA.git
ComfyUI/models/vibevoice/microsoft/VibeVoice-1.5B/
├── model-00001-of-00003.safetensors ... (模型权重)
├── config.json
├── modeling_vibevoice_*.py (Python代码)
└── [Tokenizer Files] (必须包含以下 Qwen 文件!)
├── tokenizer.json
├── tokenizer_config.json
@@ -451,6 +450,8 @@ git clone https://github.com/havvk/ComfyUI_AIIA.git
└── merges.txt
```
**💡 说明**: 插件已内置并修复了所有 VibeVoice 的 Python 核心代码 (`vibevoice_core`)。你**不需要**也不建议在模型目录中保留 `modeling_vibevoice_*.py` 等 Python 脚本,以避免潜在的冲突。
**⚠️ 重要提示**: VibeVoice 依赖 **Qwen2.5-1.5B** 的分词器。如果你下载的模型包里没有上面列出的 tokenizer 文件,请手动从 [Qwen/Qwen2.5-1.5B](https://huggingface.co/Qwen/Qwen2.5-1.5B/tree/main) 下载这 4 个文件并放入模型目录。
**手动下载命令**:
+6 -4
View File
@@ -245,13 +245,15 @@ class AIIA_VibeVoice_Loader:
trust_remote_code=False
)
if sys_path_added:
sys.path.remove(load_path)
# Cleanup not needed as we didn't add load_path to sys.path
# if sys_path_added:
# sys.path.remove(core_path) # We keep core_path for now to ensure sub-modules resolve
pass
except Exception as e:
print(f"[AIIA] Manual import/registration failed: {e}")
if 'sys_path_added' in locals() and sys_path_added:
sys.path.remove(load_path)
# if 'sys_path_added' in locals() and sys_path_added:
# sys.path.remove(load_path)
raise e
# 7. Load Tokenizer & Processor