Update pipeline.py

pipeline.py

@@ -1206,6 +1206,18 @@ class AnimateDiffPipeline(DiffusionPipeline, TextualInversionLoaderMixin, IPAdap
                     # select the relevent context from the latents
                     current_context_latents = latents[:, :, current_context_indexes, :, :]
                     
+                    if self.controlnet != None:
+                        # if we are using multiple controlnets, select the context window for each controlnet
+                        if isinstance(controlnet, MultiControlNetModel):
+                            for c in range(len(controlnet.nets)):
+                                current_context_conditioning_frames[c] = conditioning_frames[c][current_context_indexes, :, :, :]
+                        else:
+                            # select the relevent context from the conditioning frames of shape (frame_number, channel, height, width)
+                            current_context_conditioning_frames = conditioning_frames[current_context_indexes, :, :, :]
+                    else:
+                        current_context_conditioning_frames = None
+                    
+                    
                     # expand the latents if we are doing classifier free guidance
                     latent_model_input = torch.cat([current_context_latents] * 2) if do_classifier_free_guidance else current_context_latents
                     latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)