Update app.py

app.py

@@ -5,6 +5,7 @@ import pytesseract
 from PIL import Image
 import requests
 from io import BytesIO
+import difflib
 
 # Load model
 reviewer = pipeline("text2text-generation", model="google/flan-t5-base")
@@ -27,22 +28,33 @@ def extract_text_from_url(url):
     else:
         return "❌ Blog Error: Could not fetch content from the URL."
 
-# Review line-by-line
+# Highlight differences using difflib
+def highlight_diffs(original, suggestion):
+    diff = difflib.ndiff(original.split(), suggestion.split())
+    result = []
+    for token in diff:
+        if token.startswith("- "):
+            result.append(f"~~{token[2:]}~~")
+        elif token.startswith("+ "):
+            result.append(f"**{token[2:]}**")
+        elif token.startswith("  "):
+            result.append(token[2:])
+    return " ".join(result)
+
+# Review lines with diffs
 def review_lines(text):
     lines = text.strip().split('\n')
     suggestions = []
     for line in lines:
         if line.strip() == "":
             continue
-        prompt = (
-            f"Review this line for grammar, tone, and offensive language. "
-            f"Show suggested corrections with changes **highlighted** in Markdown (e.g., ~~wrong~~ → **right**):\n\n{line}"
-        )
-        suggestion = reviewer(prompt, max_new_tokens=100)[0]['generated_text']
-        suggestions.append((line, suggestion.strip()))
+        prompt = f"Rewrite this to fix grammar, tone, and remove any offensive language:\n\n{line}"
+        suggestion = reviewer(prompt, max_new_tokens=100)[0]['generated_text'].strip()
+        highlighted = highlight_diffs(line.strip(), suggestion)
+        suggestions.append((line, highlighted, suggestion))
     return suggestions
 
-# Finalize text
+# Finalize accepted suggestions
 def finalize_text(originals, suggestions, decisions):
     final = []
     for orig, sugg, keep in zip(originals, suggestions, decisions):
@@ -65,33 +77,36 @@ with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as app:
         finalize_btn = gr.Button("✅ Finalize Clean Blog")
 
     review_section = gr.Column(visible=False)
-    review_boxes = []  # Will store tuples: (orig_box, sugg_box, accept_checkbox)
+    review_boxes = []  # Will store tuples: (original_box, highlighted_markdown_box, accept_checkbox, clean_suggestion)
 
     final_output = gr.Textbox(label="📦 Final Clean Blog", lines=10)
 
+    # Text extraction logic
     def extract_both(url, img_url):
         blog = extract_text_from_url(url)
         ocr = extract_text_from_image_url(img_url) if img_url else ""
         return blog + ("\n" + ocr if ocr else "")
 
+    # Review processing with diffs
     def process_review(text):
         results = review_lines(text)
         review_section.children.clear()
         review_boxes.clear()
 
-        for i, (orig, sugg) in enumerate(results):
+        for i, (orig, highlighted, clean_sugg) in enumerate(results):
             with review_section:
                 orig_box = gr.Textbox(value=orig, label=f"Original Line {i+1}", interactive=False)
-                sugg_box = gr.Markdown(value=sugg, label=f"Suggested Edit {i+1}")
+                markdown_sugg = gr.Markdown(value=highlighted, label=f"Suggested Edit {i+1}")
                 accept_box = gr.Checkbox(label="✅ Accept Suggestion", value=False)
-                review_boxes.append((orig_box, sugg_box, accept_box))
+                review_boxes.append((orig_box, markdown_sugg, accept_box, clean_sugg))
         return gr.update(visible=True)
 
+    # Finalization logic
     def collect_dynamic_decisions():
         originals_vals = [box[0].value for box in review_boxes]
-        suggestions_vals = [box[1].value for box in review_boxes]
+        clean_suggestions = [box[3] for box in review_boxes]
         accepts_vals = [box[2].value for box in review_boxes]
-        return finalize_text(originals_vals, suggestions_vals, accepts_vals)
+        return finalize_text(originals_vals, clean_suggestions, accepts_vals)
 
     # Wire actions
     extract_btn.click(fn=extract_both, inputs=[blog_url, image_url], outputs=combined_text)