Update app.py

app.py

@@ -156,13 +156,13 @@ def infer(
     print('-- filtered prompt --')
     print(enhanced_prompt)
     if latent_file:  # Check if a latent file is provided
-        initial_latents = pipeline.prepare_latents(
+        initial_latents = pipe.prepare_latents(
             batch_size=1,
-            num_channels_latents=pipeline.unet.in_channels,
-            height=pipeline.unet.sample_size[0],
-            width=pipeline.unet.sample_size[1],
-            dtype=pipeline.unet.dtype,
-            device=pipeline.device,
+            num_channels_latents=pipe.unet.in_channels,
+            height=pipe.unet.sample_size[0],
+            width=pipe.unet.sample_size[1],
+            dtype=pipe.unet.dtype,
+            device=pipe.device,
             generator=generator,
         )
         sd_image_a = torch.load(latent_file.name)  # Load the latent
@@ -203,13 +203,13 @@ def infer(
     # Encode the generated image into latents
     with torch.no_grad():
         generated_latents = vae.encode(generated_image_tensor.to(torch.bfloat16)).latent_dist.sample().mul_(0.18215)
-        initial_latents = pipeline.prepare_latents(
+        initial_latents = pipe.prepare_latents(
             batch_size=1,
-            num_channels_latents=pipeline.unet.in_channels,
-            height=pipeline.unet.sample_size[0],
-            width=pipeline.unet.sample_size[1],
-            dtype=pipeline.unet.dtype,
-            device=pipeline.device,
+            num_channels_latents=pipe.unet.in_channels,
+            height=pipe.unet.sample_size[0],
+            width=pipe.unet.sample_size[1],
+            dtype=pipe.unet.dtype,
+            device=pipe.device,
             generator=generator,
             )
         initial_latents += generated_latents