Update app.py

app.py

@@ -96,20 +96,21 @@ def resize_image_to_bucket(image: Union[Image.Image, np.ndarray], bucket_reso: T
     return image
 
 
-def generate_video(pipe, prompt: str, frame1_path: str, frame2_path: str, guidance_scale: float, num_frames: int, num_inference_steps: int) -> bytes:
+def generate_video(pipe, prompt: str, frame1: Image.Image, frame2: Image.Image, guidance_scale: float, num_frames: int, num_inference_steps: int) -> bytes:
     # Load and preprocess frames
-    cond_frame1 = Image.open(frame1_path)
-    cond_frame2 = Image.open(frame2_path)
+    print(f"Frame 1 Path: {frame1_path}, Type: {type(frame1_path)}")
+    print(f"Frame 2 Path: {frame2_path}, Type: {type(frame2_path)}")
     
+    # Load and preprocess frames
+    cond_frame1 = frame1
+    cond_frame2 = frame2
     height, width = 720, 1280
     cond_frame1 = resize_image_to_bucket(cond_frame1, bucket_reso=(width, height))
     cond_frame2 = resize_image_to_bucket(cond_frame2, bucket_reso=(width, height))
-    
     cond_video = np.zeros(shape=(num_frames, height, width, 3))
     cond_video[0], cond_video[-1] = np.array(cond_frame1), np.array(cond_frame2)
     cond_video = torch.from_numpy(cond_video.copy()).permute(0, 3, 1, 2)
     cond_video = torch.stack([video_transforms(x) for x in cond_video], dim=0).unsqueeze(0)
-    
     with torch.no_grad():
         image_or_video = cond_video.to(device="cuda", dtype=pipe.dtype)
         image_or_video = image_or_video.permute(0, 2, 1, 3, 4).contiguous()  # [B, F, C, H, W] -> [B, C, F, H, W]
@@ -117,7 +118,6 @@ def generate_video(pipe, prompt: str, frame1_path: str, frame2_path: str, guidan
         cond_latents = cond_latents * pipe.vae.config.scaling_factor
         cond_latents = cond_latents.to(dtype=pipe.dtype)
         assert not torch.any(torch.isnan(cond_latents))
-    
     # Generate video
     video = call_pipe(
         pipe,
@@ -130,17 +130,13 @@ def generate_video(pipe, prompt: str, frame1_path: str, frame2_path: str, guidan
         guidance_scale=guidance_scale,
         generator=torch.Generator(device="cuda").manual_seed(0),
     ).frames[0]
-    
     # Export to video
     video_path = "output.mp4"
     export_to_video(video, video_path, fps=24)
-    
     with open(video_path, "rb") as video_file:
         video_bytes = video_file.read()
-    
     return video_bytes
 
-
 @torch.inference_mode()
 def call_pipe(
     pipe,
@@ -303,8 +299,8 @@ def main():
     # Define the interface inputs
     inputs = [
         gr.Textbox(label="Prompt", value="a woman"),
-        gr.Image(label="Frame 1", type="filepath"),
-        gr.Image(label="Frame 2", type="filepath"),
+        gr.Image(label="Frame 1", type="pil"),
+        gr.Image(label="Frame 2", type="pil"),
         # gr.Textbox(label="Frame 1 URL", value="https://i-bacon.bunkr.ru/11b45aa7-630b-4189-996f-a6b37a697786.png"),
         # gr.Textbox(label="Frame 2 URL", value="https://i-bacon.bunkr.ru/2382224f-120e-482d-a75d-f1a1bf13038c.png"),
         gr.Slider(minimum=0.1, maximum=20, step=0.1, label="Guidance Scale", value=6.0),