Update pipelines/pipeline_seesr.py

pipelines/pipeline_seesr.py

@@ -1025,7 +1025,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
 
         if use_KDS:
             # 1) update batch_size to account for the new particles
-            print(f'beforehand {latents.shape}:{prompt_embeds.shape}')
             batch_size = batch_size * num_particles
         
             # 2) now repeat latents/images/prompts
@@ -1061,19 +1060,14 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
                     if guess_mode and do_classifier_free_guidance:
                         # Infer ControlNet only for the conditional batch.
                         controlnet_latent_model_input = latents
-                        controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
-                        print("well unexpected")
-                        
+                        controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]                   
                     else:
                         controlnet_latent_model_input = latent_model_input
                         controlnet_prompt_embeds = prompt_embeds
-                        print("a possiblity")
 
                     if h*w<=tile_size*tile_size: # tiled latent input
                         down_block_res_samples, mid_block_res_sample = [None]*10, None
 
-                        print(f"controlnet 1 started with {controlnet_latent_model_input.shape}:{ram_encoder_hidden_states.shape}")
-                        
                         down_block_res_samples, mid_block_res_sample = self.controlnet(
                             controlnet_latent_model_input,
                             t,
@@ -1094,7 +1088,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
                             mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
 
                         # predict the noise residual
-                        print(f"unet started with {latent_model_input.shape}:{prompt_embeds.shape}")
                         noise_pred = self.unet(
                             latent_model_input,
                             t,
@@ -1157,7 +1150,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
                                     cond_list_t = torch.cat(cond_list, dim=0)
                                     img_list_t = torch.cat(img_list, dim=0)
                                     #print(input_list_t.shape, cond_list_t.shape, img_list_t.shape, fg_mask_list_t.shape)
-                                    print(f"controlnet 2 started with {cond_list_t.shape}:{controlnet_prompt_embeds.shape}")
 
                                     down_block_res_samples, mid_block_res_sample = self.controlnet(
                                         cond_list_t,
@@ -1178,7 +1170,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
                                         mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
 
                                     # predict the noise residual
-                                    print(f"unet started with {input_list_t.shape}:{prompt_embeds.shape}")
                                     model_out = self.unet(
                                         input_list_t,
                                         t,
@@ -1233,8 +1224,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
 
                 if use_KDS:
 
-                    print("using KDS")
-
                     # 2) Compute x₀ prediction
                     beta_t   = 1 - self.scheduler.alphas_cumprod[t]
                     alpha_t  = self.scheduler.alphas_cumprod[t].sqrt()
@@ -1275,8 +1264,6 @@ class StableDiffusionControlNetPipeline(DiffusionPipeline, TextualInversionLoade
                     ).detach().requires_grad_(True)
                 else:
 
-                    print("KDS Free")
-                    
                     # compute the previous noisy sample x_t -> x_t-1
                     latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]