Update pipeline.py

pipeline.py

@@ -519,7 +519,7 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
 
     # Copied from diffusers.pipelines.text_to_video_synthesis.pipeline_text_to_video_synth.TextToVideoSDPipeline.prepare_latents
     def prepare_latents(
-        self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
+        self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None, smooth_weight=0.5
     ):
         shape = (
             batch_size,
@@ -573,10 +573,8 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     weight = (smooth_steps - abs(s)) / smooth_steps
                     blended_latent += latents[:, :, frame_index] * weight
                 latents[:, :, i] = blended_latent / (2 * smooth_steps)
-            # just for testing make all frames have the same latent noise as frame 1
-            single_latent = randn_tensor((batch_size, num_channels_latents, 1, height // self.vae_scale_factor, width // self.vae_scale_factor), generator=generator, device=device, dtype=dtype)
-            latents = single_latent.repeat(1, 1, num_frames, 1, 1)
-
+                
+            latents = torch.lerp(randn_tensor(shape, generator=generator, device=device, dtype=dtype),latents, smooth_weight)
         else:
             latents = latents.to(device)
 
@@ -783,7 +781,7 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
         latents = latents.to(device)
         return latents, init_latents
 
-
+ 
     @torch.no_grad()
     # @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(