Update pipeline.py

pipeline.py

@@ -541,7 +541,7 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
         latents = latents * self.scheduler.init_noise_sigma
         return latents
     
-    def prepare_latents_same_start(self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None, context_size=16, blend_frames=4):
+    def prepare_latents_same_start(self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None, context_size=16, overlap=4, strength=0.5):
         shape = (
             batch_size,
             num_channels_latents,
@@ -560,12 +560,13 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
         else:
             latents = latents.to(device)
             
-        # make every (context_size-blend_frames) frames have the same noise
-        loop_size = context_size - blend_frames
+        # make every (context_size-overlap) frames have the same noise
+        loop_size = context_size - overlap
         loop_count = num_frames // loop_size
         for i in range(loop_count):
             # repeat the first frames noise for i*loop_size frame
-            latents[:, :, i*loop_size, :, :] = latents[:, :, 0, :, :]
+            # lerp the first frames noise
+            latents[:, :, i*loop_size:(i*loop_size)+overlap, :, :] = torch.lerp(latents[:, :, i*loop_size:(i*loop_size)+overlap, :, :], latents[:, :, 0:overlap, :, :], strength)
             
         # scale the initial noise by the standard deviation required by the scheduler
         latents = latents * self.scheduler.init_noise_sigma