Update app.py

app.py

@@ -247,7 +247,7 @@ def uploadNote(prompt,num_inference_steps,guidance_scale,timestamp):
 def captioning(img):
     prompts_array = [
        # "Adjectives describing this scene are:",
-        "The color scheme of this image is",
+      #  "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",
@@ -258,15 +258,15 @@ def captioning(img):
     ]
     output_prompt=[]
     cap_prompt = (
-            "Describe this image with a caption."
-    )    
+            "Describe this image with a caption to be used for image generation."
+    )
     inputsa = processor5(images=img, text=cap_prompt, return_tensors="pt").to('cuda')
     generated_ids = model5.generate(
         **inputsa,
         do_sample=False,
         num_beams=5,
-        max_length=512,
-        #min_length=1,
+        max_length=768,
+        min_length=64,
         top_p=0.9,
         repetition_penalty=1.5,
         length_penalty=1.0,
@@ -274,12 +274,12 @@ def captioning(img):
     )
     
     generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
+    generated_text = generated_text.replace(cap_prompt, "").strip() #Or could try .split(prompt, 1)[-1].strip()
     output_prompt.append(generated_text)
     print(generated_text)  
     # 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,
             do_sample=False,
@@ -290,21 +290,23 @@ def captioning(img):
             repetition_penalty=1.5,
             length_penalty=1.0,
             temperature=1,
-        ) 
-        
+        )
         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, max_length=768)
-    generated_text = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
-    response_text = generated_text.replace(generated_text, "").strip()
-    print(response_text)
+    inputf = processor5(
+       # images=img,
+        text=generated_text + 'So therefore',
+        return_tensors="pt"
+    ).to('cuda')
+    generated_ids = model5.generate(**inputf, max_length=256)
+    generated_texta = processor5.batch_decode(generated_ids, skip_special_tokens=True)[0].strip()
+    response_text = generated_texta.replace(generated_text, "").strip()
+    #print(response_text)
     output_prompt.append(response_text)
-    print(output_prompt)
+    #print(output_prompt)
+    output_prompt = " ".join(output_prompt)
     return output_prompt
     
 def flatten_and_stringify(data):
@@ -437,8 +439,7 @@ def generate_30(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
-        captions = caption.append(flatten_and_stringify(prompt))
-        captions = flatten_and_stringify(captions)
+        prompt = " ".join(prompt)
         captions = " ".join(captions)
         print(captions)
         print("-- not generating further caption --")
@@ -450,7 +451,7 @@ def generate_30(
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
         #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        new_prompt = prompt + ' ' + captions
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)
@@ -563,8 +564,7 @@ def generate_60(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
-        captions =caption.append(prompt)
-        captions = flatten_and_stringify(captions)
+        prompt = " ".join(prompt)
         captions = " ".join(captions)
         print(captions)
         print("-- not generating further caption --")
@@ -576,7 +576,7 @@ def generate_60(
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
         #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        new_prompt = prompt + ' ' + captions
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)
@@ -689,8 +689,7 @@ def generate_90(
         timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
         filename= f'rv_IPb_{timestamp}.png'
         print("-- using image file --")
-        captions =caption.append(prompt)
-        captions = flatten_and_stringify(captions)
+        prompt = " ".join(prompt)
         captions = " ".join(captions)
         print(captions)
         print("-- not generating further caption --")
@@ -702,7 +701,7 @@ def generate_90(
         torch.cuda.empty_cache()
         torch.cuda.reset_peak_memory_stats()
         #expanded = expand_prompt(captions)
-        new_prompt = captions #prompt+' '+expanded
+        new_prompt = prompt + ' ' + captions
         print("-- ------------ --")
         print("-- FINAL PROMPT --")
         print(new_prompt)