Update pipeline.py

pipeline.py

@@ -1006,17 +1006,22 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                         noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                         noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
-                    # compute the previous noisy sample x_t -> x_t-1
-                    current_context_latents = self.scheduler.step(noise_pred, t, current_context_latents, **extra_step_kwargs).prev_sample
-                    
-                    # if context_start + context_size > num_frames: remove the appended frames from the end of the current_context_latents
+                    # remove the appended frames from the end of the current_context_latents
                     if wrap_count > 0:
-                        # add the ending frames from current_context_latents to the start of the latent_sum
-                        latent_sum[:, :, 0:wrap_count, :, :] += current_context_latents[:, :, -wrap_count:, :, :]
-                        # increase the counter for the ending frames
-                        latent_counter[0:wrap_count] += 1
                         # remove the ending frames from current_context_latents
                         current_context_latents = current_context_latents[:, :, :-wrap_count, :, :]
+                    
+                    # compute the previous noisy sample x_t -> x_t-1
+                    current_context_latents = self.scheduler.step(noise_pred, t, current_context_latents, **extra_step_kwargs).prev_sample
+                    
+                    # # if context_start + context_size > num_frames: remove the appended frames from the end of the current_context_latents
+                    # if wrap_count > 0:
+                    #     # add the ending frames from current_context_latents to the start of the latent_sum
+                    #     latent_sum[:, :, 0:wrap_count, :, :] += current_context_latents[:, :, -wrap_count:, :, :]
+                    #     # increase the counter for the ending frames
+                    #     latent_counter[0:wrap_count] += 1
+                    #     # remove the ending frames from current_context_latents
+                    #     current_context_latents = current_context_latents[:, :, :-wrap_count, :, :]
 
                     #add the context current_context_latents back to the latent sum starting from the current context start
                     latent_sum[:, :, current_context_start : current_context_start + context_size, :, :] += current_context_latents