Update app.py

app.py

@@ -239,13 +239,13 @@ def uploadNote(prompt,num_inference_steps,guidance_scale,timestamp):
 @torch.no_grad()
 def captioning(img):
     prompts_array = [
-        "Adjectives describing this scene are:",
-        "The color scheme of this image is",
-        "This scene could be described in detail as",
-        "The characters in this scene are",
-        "The larger details in this scene include",
-        "The smaller details in this scene include",
-        "The feeling this scene seems like",
+      #  "Adjectives describing this scene are:",
+      #  "The color scheme of this image is",
+      #  "This scene could be described in detail as",
+      #  "The characters in this scene are",
+      #  "The larger details in this scene include",
+      #  "The smaller details in this scene include",
+      #  "The feeling this scene seems like",
         "The setting of this scene must be located", 
     # Add more prompts here
     ]
@@ -259,18 +259,18 @@ def captioning(img):
     # Loop through prompts array:
     for prompt in prompts_array:
         inputs = processor5(images=img, text=prompt, return_tensors="pt").to('cuda')
-        generated_ids = model5.generate(**inputs, min_length=42, max_length=64) # Adjust max_length if needed
+        generated_ids = model5.generate(**inputs, min_length=32, max_length=96) # Adjust max_length if needed
         generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
         response_text = generated_text.replace(prompt, "").strip() #Or could try .split(prompt, 1)[-1].strip()
         output_prompt.append(response_text)
         print(f"{response_text}\n") # Print only the response text
-    # Continue conversation:
-   # inputf = processor5(images=img, text=generated_text + ' So therefore, ', return_tensors="pt").to('cuda')
-   # generated_ids = model5.generate(**inputf, min_length=24, max_length=42)
-  #  generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
-   # response_text = generated_text.replace(generated_text, "").strip()  # Remove the previous text plus 'So therefore'
-  # print(response_text)
-    #output_prompt.append(response_text)
+     Continue conversation:
+    inputf = processor5(images=img, text=generated_text + ' So therefore, ', return_tensors="pt").to('cuda')
+    generated_ids = model5.generate(**inputf, min_length=32, max_length=96)
+    generated_texta = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
+    response_text = generated_texta.replace(generated_text, "").strip()  # Remove the previous text plus 'So therefore'
+    print(response_text)
+    output_prompt.append(response_text)
     print(output_prompt)
     return output_prompt
     
@@ -300,7 +300,7 @@ def expand_prompt(prompt):
         outputs = model.generate(
             input_ids=input_ids,
             attention_mask=attention_mask,
-            max_new_tokens=128,
+            max_new_tokens=1024,
             temperature=0.2,
             top_p=0.9,
             do_sample=True,
@@ -376,6 +376,9 @@ def generate_30(
         caption_2=[]
         #caption.append(captioner(sd_image_a))
         caption.append(captioner2(sd_image_a))
+        cap = captioner2(sd_image_b)
+        caption.append(cap)
+        print(cap)
         #caption.append(captioner_3(sd_image_a))
         caption_2.append(captioning(sd_image_a))
         if latent_file_2 is not None:  # Check if a latent file is provided
@@ -383,7 +386,9 @@ def generate_30(
             #sd_image_b.resize((height,width), Image.LANCZOS)
             sd_image_b.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_b))
-            caption.append(captioner2(sd_image_b))
+            cap = captioner2(sd_image_b)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_b))
             caption_2.append(captioning(sd_image_b))
         else:
@@ -393,7 +398,9 @@ def generate_30(
             #sd_image_c.resize((height,width), Image.LANCZOS)
             sd_image_c.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_c))
-            caption.append(captioner2(sd_image_c))
+            cap = captioner2(sd_image_c)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_c))
             caption_2.append(captioning(sd_image_c))
         else:
@@ -403,7 +410,9 @@ def generate_30(
             #sd_image_d.resize((height,width), Image.LANCZOS)
             sd_image_d.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_d))
-            caption.append(captioner2(sd_image_d))
+            cap = captioner2(sd_image_d)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_d))
             caption_2.append(captioning(sd_image_d))
         else:
@@ -413,7 +422,9 @@ def generate_30(
             #sd_image_e.resize((height,width), Image.LANCZOS)
             sd_image_e.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_e))
-            caption.append(captioner2(sd_image_e))
+            cap = captioner2(sd_image_e)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_e))
             caption_2.append(captioning(sd_image_e))
         else:
@@ -442,11 +453,10 @@ def generate_30(
         print(new_prompt)
         print("-- FINAL PROMPT --")
         print("-- ------------ --")
-        
-        #global model
-        #global txt_tokenizer
-        #del model
-        #del txt_tokenizer
+        global model
+        global txt_tokenizer
+        del model
+        del txt_tokenizer
         gc.collect()
         torch.cuda.empty_cache()
         global text_encoder_1
@@ -529,6 +539,9 @@ def generate_60(
         caption_2=[]
         #caption.append(captioner(sd_image_a))
         caption.append(captioner2(sd_image_a))
+        cap = captioner2(sd_image_b)
+        caption.append(cap)
+        print(cap)
         #caption.append(captioner_3(sd_image_a))
         caption_2.append(captioning(sd_image_a))
         if latent_file_2 is not None:  # Check if a latent file is provided
@@ -536,7 +549,9 @@ def generate_60(
             #sd_image_b.resize((height,width), Image.LANCZOS)
             sd_image_b.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_b))
-            caption.append(captioner2(sd_image_b))
+            cap = captioner2(sd_image_b)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_b))
             caption_2.append(captioning(sd_image_b))
         else:
@@ -546,7 +561,9 @@ def generate_60(
             #sd_image_c.resize((height,width), Image.LANCZOS)
             sd_image_c.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_c))
-            caption.append(captioner2(sd_image_c))
+            cap = captioner2(sd_image_c)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_c))
             caption_2.append(captioning(sd_image_c))
         else:
@@ -556,7 +573,9 @@ def generate_60(
             #sd_image_d.resize((height,width), Image.LANCZOS)
             sd_image_d.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_d))
-            caption.append(captioner2(sd_image_d))
+            cap = captioner2(sd_image_d)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_d))
             caption_2.append(captioning(sd_image_d))
         else:
@@ -566,7 +585,9 @@ def generate_60(
             #sd_image_e.resize((height,width), Image.LANCZOS)
             sd_image_e.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_e))
-            caption.append(captioner2(sd_image_e))
+            cap = captioner2(sd_image_e)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_e))
             caption_2.append(captioning(sd_image_e))
         else:
@@ -595,11 +616,10 @@ def generate_60(
         print(new_prompt)
         print("-- FINAL PROMPT --")
         print("-- ------------ --")
-        
-        #global model
-        #global txt_tokenizer
-        #del model
-        #del txt_tokenizer
+        global model
+        global txt_tokenizer
+        del model
+        del txt_tokenizer
         gc.collect()
         torch.cuda.empty_cache()
         global text_encoder_1
@@ -682,6 +702,9 @@ def generate_90(
         caption_2=[]
         #caption.append(captioner(sd_image_a))
         caption.append(captioner2(sd_image_a))
+        cap = captioner2(sd_image_b)
+        caption.append(cap)
+        print(cap)
         #caption.append(captioner_3(sd_image_a))
         caption_2.append(captioning(sd_image_a))
         if latent_file_2 is not None:  # Check if a latent file is provided
@@ -689,7 +712,9 @@ def generate_90(
             #sd_image_b.resize((height,width), Image.LANCZOS)
             sd_image_b.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_b))
-            caption.append(captioner2(sd_image_b))
+            cap = captioner2(sd_image_b)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_b))
             caption_2.append(captioning(sd_image_b))
         else:
@@ -699,7 +724,9 @@ def generate_90(
             #sd_image_c.resize((height,width), Image.LANCZOS)
             sd_image_c.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_c))
-            caption.append(captioner2(sd_image_c))
+            cap = captioner2(sd_image_c)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_c))
             caption_2.append(captioning(sd_image_c))
         else:
@@ -709,7 +736,9 @@ def generate_90(
             #sd_image_d.resize((height,width), Image.LANCZOS)
             sd_image_d.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_d))
-            caption.append(captioner2(sd_image_d))
+            cap = captioner2(sd_image_d)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_d))
             caption_2.append(captioning(sd_image_d))
         else:
@@ -719,7 +748,9 @@ def generate_90(
             #sd_image_e.resize((height,width), Image.LANCZOS)
             sd_image_e.resize((768,768), Image.LANCZOS)
             #caption.append(captioner(sd_image_e))
-            caption.append(captioner2(sd_image_e))
+            cap = captioner2(sd_image_e)
+            caption.append(cap)
+            print(cap)
             #caption.append(captioner_3(sd_image_e))
             caption_2.append(captioning(sd_image_e))
         else:
@@ -748,11 +779,10 @@ def generate_90(
         print(new_prompt)
         print("-- FINAL PROMPT --")
         print("-- ------------ --")
-        
-        #global model
-        #global txt_tokenizer
-        #del model
-        #del txt_tokenizer
+        global model
+        global txt_tokenizer
+        del model
+        del txt_tokenizer
         gc.collect()
         torch.cuda.empty_cache()
         global text_encoder_1