Update pipeline.py

pipeline.py

@@ -999,20 +999,21 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     if do_classifier_free_guidance:
                         noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                         noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
-
-                    # # remove the appended frames from the end of the current_context_latents
-                    # if wrap_count > 0:
-                    #     # remove the ending frames from current_context_latents
-                    #     current_context_latents = current_context_latents[:, :, :-wrap_count, :, :]
-                    #     # remove the ending frames from noise_pred
-                    #     noise_pred = noise_pred[:, :, :-wrap_count, :, :]
-                    #     # print the shape of the current_context_latents and noise_pred
-                    #     print(f"current_context_latents shape: {current_context_latents.shape}")
-                    #     print(f"noise_pred shape: {noise_pred.shape}")
-                    
+                        
                     # 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
                     
+
+                    # remove the appended frames from the end of the current_context_latents
+                    if wrap_count > 0:
+                        # remove the ending frames from current_context_latents
+                        current_context_latents = current_context_latents[:, :, :-wrap_count, :, :]
+                        # remove the ending frames from noise_pred
+                        noise_pred = noise_pred[:, :, :-wrap_count, :, :]
+                        # print the shape of the current_context_latents and noise_pred
+                        print(f"current_context_latents shape: {current_context_latents.shape}")
+                        print(f"noise_pred shape: {noise_pred.shape}")
+                    
                     # # 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