modifying requirements.txt

inference/cli_demo_camera_i2v_pcd.py

@@ -14,7 +14,7 @@ from diffusers import (
     AutoencoderKLCogVideoX
 )
 from diffusers.utils import export_to_video, load_video
-
+from torchvision.transforms import ToPILImage
 from controlnet_pipeline import ControlnetCogVideoXImageToVideoPCDPipeline
 from cogvideo_transformer import CustomCogVideoXTransformer3DModel
 from cogvideo_controlnet_pcd import CogVideoXControlnetPCD
@@ -369,8 +369,8 @@ def generate_video(
         # ++++++++++++++++++++++++++++++++++++++
         latents = video_generate_all # This is a latent
         # Fix the nested list structure
-        if isinstance(latents, list) and len(latents) == 1 and isinstance(latents[0], list):
-            latents = latents[0]  # ✅ Unwrap the inner list of PIL images
+        # if isinstance(latents, list) and len(latents) == 1 and isinstance(latents[0], list):
+        #     latents = latents[0]  # ✅ Unwrap the inner list of PIL images
 
         transform = T.ToTensor()  # Converts PIL image to torch.FloatTensor in [0,1]
 
@@ -403,7 +403,13 @@ def generate_video(
         if rife_status:
             latents = rife_inference_with_latents(frame_interpolation_model, latents)
 
+
+        # Convert latents back to PIL images after processing
+        latents = latents.clamp(0, 1)  # Clamp values to [0,1]
+        latents = [T.ToPILImage()(frame.cpu()) for frame in latents]
+
         video_generate_all = latents
+
         # ++++++++++++++++++++++++++++++++++++++
 
         video_generate = video_generate_all[0]