Update app.py

app.py

@@ -28,7 +28,7 @@ hf_token = os.environ.get('HF_TOKEN')
 device = "cuda"
 
 #snapshot_download("runwayml/stable-diffusion-v1-5", cache_dir="checkpoints")
-sd_path = "runwayml/stable-diffusion-v1-5"
+sd_path = "checkpoints/stable-diffusion-v1-5"
 inter_path = "checkpoints/flownet.pkl"
 controlnet_dict = {
     "pose": "checkpoints/sd-controlnet-openpose",
@@ -89,8 +89,8 @@ def infer(prompt, video_path, condition, video_length, is_long_video):
     tokenizer = CLIPTokenizer.from_pretrained(sd_path, subfolder="tokenizer")
     text_encoder = CLIPTextModel.from_pretrained(sd_path, subfolder="text_encoder").to(dtype=torch.float16)
     vae = AutoencoderKL.from_pretrained(sd_path, subfolder="vae").to(dtype=torch.float16)
-    unet = UNet3DConditionModel.from_pretrained(sd_path, subfolder="unet").to(dtype=torch.float16)
-    controlnet = ControlNetModel3D.from_pretrained(controlnet_dict[condition]).to(dtype=torch.float16)
+    unet = UNet3DConditionModel.from_pretrained_2d(sd_path, subfolder="unet").to(dtype=torch.float16)
+    controlnet = ControlNetModel3D.from_pretrained_2d(controlnet_dict[condition]).to(dtype=torch.float16)
     interpolater = IFNet(ckpt_path=inter_path).to(dtype=torch.float16)
     scheduler=DDIMScheduler.from_pretrained(sd_path, subfolder="scheduler")
 
@@ -150,7 +150,7 @@ with gr.Blocks() as demo:
         video_path = gr.Video(source="upload", type="filepath")
         condition = gr.Textbox(label="Condition", value="depth")
         video_length = gr.Slider(label="video length", minimum=1, maximum=15, step=1, value=2)
-        seed = gr.Number(label="seed", valie=42)
+        seed = gr.Number(label="seed", value=42)
         submit_btn = gr.Button("Submit")
         video_res = gr.Video(label="result")