Update Era_s20_updt.py

Era_s20_updt.py

@@ -1,6 +1,5 @@
-#!pip install -q --upgrade  transformers==4.25.1 diffusers ftfy accelerate
-#import transformers as t
-#assert t.__version__=='4.25.1', "Transformers version should be as specified"
+# import transformers as t
+# assert t.__version__=='4.25.1', "Transformers version should be as specified"
 
 
 import torch
@@ -25,7 +24,7 @@ import torch.nn.functional as F
 
 torch.manual_seed(1)
 
-#if not (Path.home()/'.cache/huggingface'/'token').exists(): notebook_login()
+if not (Path.home()/'.cache/huggingface'/'token').exists(): notebook_login()
 
 # Supress some unnecessary warnings when loading the CLIPTextModel
 logging.set_verbosity_error()
@@ -88,7 +87,7 @@ vae = vae.to(torch_device)
 text_encoder = text_encoder.to(torch_device)
 unet = unet.to(torch_device);
 
-embeds_folder = Path('stable_diffusion_experiment/paintings_embed')
+embeds_folder = Path('C:/Users/shivs/Downloads/paintings_embed')
 file_names = [path.name for path in embeds_folder.glob('*') if path.is_file()]
 print(file_names)
 
@@ -193,7 +192,7 @@ def get_output_embeds(input_embeddings):
 def generate_with_embs_custom(text_embeddings,seed):
     height = 512                        # default height of Stable Diffusion
     width = 512                         # default width of Stable Diffusion
-    num_inference_steps = 30            # Number of denoising steps
+    num_inference_steps = 10            # Number of denoising steps
     guidance_scale = 7.5                # Scale for classifier-free guidance
     generator = torch.manual_seed(seed)   # Seed generator to create the inital latent noise
     batch_size = 1
@@ -242,7 +241,7 @@ def generate_with_embs_custom(text_embeddings,seed):
 # ref_latent = pil_to_latent(ref_image)
 
 ## Guidance through Custom Loss Function
-def custom_loss(latent):
+def custom_loss(latent,ref_latent):
     error = F.mse_loss(0.5*latent,0.8*ref_latent)
     return error
 
@@ -279,11 +278,11 @@ class Styles_paintings():
     def generate_styles_with_custom_loss(self, image):
         height = 512                        # default height of Stable Diffusion
         width = 512                         # default width of Stable Diffusion
-        num_inference_steps = 50  #@param           # Number of denoising steps
+        num_inference_steps = 20  #@param           # Number of denoising steps
         guidance_scale = 8 #@param               # Scale for classifier-free guidance
         batch_size = 1
         custom_loss_scale = 200 #@param
-        #print('image shape there is',image.size)
+        print('image shape there is',image.size)
         self.output_styles_with_custom_loss = []
         #ref_image = Image.open('C:/Users/shivs/Downloads/ig.jpg').resize((512,512))
         ref_latent = pil_to_latent(image)            
@@ -344,7 +343,7 @@ class Styles_paintings():
                     #denoised_images = vae.decode((1 / 0.18215) * latents_x0).sample / 2 + 0.5 # range (0, 1)
 
                     # Calculate loss
-                    loss = custom_loss(latents_x0) * custom_loss_scale
+                    loss = custom_loss(latents_x0,ref_latent) * custom_loss_scale
                     #loss = blue_loss(denoised_images) * blue_loss_scale
 
                     # Occasionally print it out
@@ -362,14 +361,13 @@ class Styles_paintings():
 
             self.output_styles_with_custom_loss.append(latents_to_pil(latents)[0])
 
-def generate_final_image(im1,in_prompt):
+def generate_final_image(im1,in_prompt="an oil painting of an baby girl with flowers in a park"):
     paintings = Styles_paintings(in_prompt)
     paintings.generate_styles()
     r_image = im1.resize((512,512))
     print('image shape is',r_image.size)
     paintings.generate_styles_with_custom_loss(r_image)
     
-    print(len(paintings.output_styles))
-    print(len(paintings.output_styles_with_custom_loss))
+    #print(len(paintings.output_styles))
 
-    return [paintings.output_styles[0]], [paintings.output_styles[1]],[paintings.output_styles[2]],[paintings.output_styles[3]],[paintings.output_styles[4]],[paintings.output_styles_with_custom_loss[0]],[paintings.output_styles_with_custom_loss[1]],[paintings.output_styles_with_custom_loss[2]],[paintings.output_styles_with_custom_loss[3]],[paintings.output_styles_with_custom_loss[4]]
+    return [paintings.output_styles[0]], [paintings.output_styles[1]],[paintings.output_styles[2]],[paintings.output_styles[3]],[paintings.output_styles[4]], [paintings.output_styles_with_custom_loss[0]],[paintings.output_styles_with_custom_loss[1]],[paintings.output_styles_with_custom_loss[2]],[paintings.output_styles_with_custom_loss[3]],[paintings.output_styles_with_custom_loss[4]]