Update app.py

app.py

@@ -15,20 +15,21 @@ num_frames, width, height = 49, 832, 480
 gpu_id = 0
 device = f'cuda:{gpu_id}' if torch.cuda.is_available() else 'cpu'
 
-# from modelscope import snapshot_download
-# model_dir = snapshot_download(
-#     model_id = 'AI-ModelScope/RMBG-2.0',
-#     local_dir = 'ckpt/',
-# )
-
-from huggingface_hub import snapshot_download, hf_hub_download
-hf_hub_download(
-    repo_id="alibaba-pai/Wan2.1-Fun-1.3B-Control",
-    filename="Wan2.1_VAE.pth",
-    local_dir="ckpt/Wan2.1-Fun-1.3B-Control/",
-    local_dir_use_symlinks=False,
-    resume_download=True,
+from modelscope import snapshot_download
+model_dir = snapshot_download(
+    model_id = 'AI-ModelScope/RMBG-2.0',
+    local_dir = 'ckpt/',
 )
+print(model_dir)
+
+# from huggingface_hub import snapshot_download, hf_hub_download
+# hf_hub_download(
+#     repo_id="alibaba-pai/Wan2.1-Fun-1.3B-Control",
+#     filename="Wan2.1_VAE.pth",
+#     local_dir="ckpt/Wan2.1-Fun-1.3B-Control/",
+#     local_dir_use_symlinks=False,
+#     resume_download=True,
+# )
 
 # snapshot_download( # 下载整个仓库; 下briaai/RMBG-2.0需要token
 #     repo_id="alibaba-pai/Wan2.1-Fun-1.3B-Control",
@@ -46,10 +47,10 @@ hf_hub_download(
 #     resume_download=True,
 # )
 
-# rmbg_model = AutoModelForImageSegmentation.from_pretrained('ckpt/RMBG-2.0', trust_remote_code=True) # ckpt/RMBG-2.0
-# torch.set_float32_matmul_precision(['high', 'highest'][0])
-# rmbg_model.to(device)
-# rmbg_model.eval()
+rmbg_model = AutoModelForImageSegmentation.from_pretrained('ckpt/RMBG-2.0', trust_remote_code=True) # ckpt/RMBG-2.0
+torch.set_float32_matmul_precision(['high', 'highest'][0])
+rmbg_model.to(device)
+rmbg_model.eval()
 
 
 model_manager = ModelManager(device="cpu") # 1.3b: device=cpu: uses 6G VRAM, device=device: uses 16G VRAM; about 1-2 min per video