Update app.py

app.py

@@ -10,8 +10,8 @@ from diffusers.utils import export_to_video, load_video
 from torchvision import transforms
 import random
 from controlnet_aux import CannyDetector
-from image_gen_aux import DepthPreprocessor
-import mediapipe as mp
+# from image_gen_aux import DepthPreprocessor
+# import mediapipe as mp
 from PIL import Image
 import cv2
 
@@ -28,12 +28,11 @@ pipe_upsample.to(device)
 pipeline.vae.enable_tiling()
 
 canny_processor = CannyDetector()
-depth_processor = DepthPreprocessor.from_pretrained("LiheYoung/depth-anything-large-hf")
 
 # Initialize MediaPipe pose estimation
-mp_drawing = mp.solutions.drawing_utils
-mp_drawing_styles = mp.solutions.drawing_styles
-mp_pose = mp.solutions.pose
+# mp_drawing = mp.solutions.drawing_utils
+# mp_drawing_styles = mp.solutions.drawing_styles
+# mp_pose = mp.solutions.pose
 
 CONTROL_LORAS = {
     "canny": {
@@ -53,6 +52,14 @@ CONTROL_LORAS = {
     }
 }
 
+# load canny lora
+pipeline.load_lora_weights(
+            CONTROL_LORAS["canny"]["repo"],
+            weight_name=CONTROL_LORAS["canny"]["weight_name"],
+            adapter_name=CONTROL_LORAS["canny"]["adapter_name"]
+        )
+pipeline.set_adapters([lora_config["adapter_name"]], adapter_weights=[1.0])
+
 @spaces.GPU()
 def read_video(video) -> torch.Tensor:
     """
@@ -116,16 +123,6 @@ def process_video_for_canny(video):
      
     return canny_video
 
-@spaces.GPU()
-def process_video_for_depth(video):
-    """
-    Process video for depth control.
-    """
-    print("Processing video for depth control...")
-    depth_video = []
-    for frame in video: 
-        depth_video.append(depth_processor(frame)[0].convert("RGB"))
-    return depth_video
 
 @spaces.GPU()
 def process_video_for_pose(video):
@@ -230,7 +227,7 @@ def generate_video(
         progress(0.05, desc="Loading control LoRA...")
         
         # Load the appropriate control LoRA and update state
-        updated_lora_state = load_control_lora(control_type, current_lora_state)
+        # updated_lora_state = load_control_lora(control_type, current_lora_state)
         
         # Loads video into a list of pil images 
         video = load_video(reference_video)
@@ -323,7 +320,7 @@ def generate_video(
 with gr.Blocks() as demo:
     gr.Markdown(
         """
-        # LTX Video Control
+        # LTX Video Control Canny
         """
     )
     
@@ -350,6 +347,7 @@ with gr.Blocks() as demo:
                 label="Control Type",
                 choices=["canny", "depth", "pose"],
                 value="canny",
+                visible=False,
                 info="Choose the type of control guidance for video generation"
             )