Update app.py

app.py

@@ -1,11 +1,25 @@
 import gradio as gr
 from newspaper import Article
 from transformers import pipeline
+import pytesseract
+from PIL import Image
+import requests
+from io import BytesIO
 
-# Load small LLM
+# Load model
 reviewer = pipeline("text2text-generation", model="google/flan-t5-base")
 
-# Step 1: Extract blog content
+# OCR
+def extract_text_from_image_url(img_url):
+    try:
+        response = requests.get(img_url)
+        img = Image.open(BytesIO(response.content))
+        text = pytesseract.image_to_string(img)
+        return text
+    except Exception as e:
+        return f"❌ OCR Error: {e}"
+
+# Extract blog
 def extract_text_from_url(url):
     try:
         article = Article(url)
@@ -13,37 +27,80 @@ def extract_text_from_url(url):
         article.parse()
         return article.text
     except Exception as e:
-        return f"Error: {e}"
-
-# Step 2: Review content
-def review_blog(text):
-    prompt = f"""Please review the following blog content. Fix grammar, crude/offensive language, or policy-violating parts. Suggest replacements:
+        return f"❌ Blog Error: {e}"
 
-{text}"""
-    result = reviewer(prompt, max_new_tokens=300)[0]['generated_text']
-    return result
+# Review line-by-line
+def review_lines(text):
+    lines = text.strip().split('\n')
+    suggestions = []
+    for line in lines:
+        if line.strip() == "":
+            continue
+        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']
+        suggestions.append((line, suggestion.strip()))
+    return suggestions
 
-# Step 3: Finalize after approval
-def finalize_text(original_text, reviewed_suggestions):
-    # For simplicity, we'll just return reviewed version for now
-    # Later you can implement side-by-side diff + selective replacement
-    return reviewed_suggestions
+# Finalize by combining accepted lines
+def finalize_text(originals, suggestions, decisions):
+    final = []
+    for orig, sugg, keep in zip(originals, suggestions, decisions):
+        final.append(sugg if keep else orig)
+    return "\n".join(final)
 
 # Gradio UI
-with gr.Blocks() as app:
-    gr.Markdown("## 🧠 BlogChecker AI")
-    
-    url = gr.Textbox(label="Enter Blog URL")
-    blog_content = gr.Textbox(label="Extracted Blog Text", lines=10)
-    reviewed = gr.Textbox(label="Suggested Revisions", lines=10)
-    final_output = gr.Textbox(label="Final Clean Blog", lines=10)
-
-    extract_btn = gr.Button("1️⃣ Extract Blog Text")
-    review_btn = gr.Button("2️⃣ Review Content")
-    approve_btn = gr.Button("3️⃣ Approve and Finalize")
-
-    extract_btn.click(fn=extract_text_from_url, inputs=url, outputs=blog_content)
-    review_btn.click(fn=review_blog, inputs=blog_content, outputs=reviewed)
-    approve_btn.click(fn=finalize_text, inputs=[blog_content, reviewed], outputs=final_output)
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as app:
+    gr.Markdown("## ✨ BlogChecker AI\nSmart AI reviewer for blog content, with interactive approval and OCR image support.")
+
+    with gr.Row():
+        blog_url = gr.Textbox(label="📎 Blog URL")
+        image_url = gr.Textbox(label="🖼️ Image URL (optional)")
+        extract_btn = gr.Button("🔍 Extract")
+
+    combined_text = gr.Textbox(label="📝 Combined Blog + OCR Text", lines=10)
+
+    with gr.Row():
+        review_btn = gr.Button("🧠 Review Content")
+        finalize_btn = gr.Button("✅ Finalize Clean Blog")
+
+    line_outputs = []
+    decisions = []
+    originals = []
+    suggestions = []
+
+    with gr.Column() as dynamic_review_section:
+        for i in range(5):  # support 5 lines for demo; can be dynamic later
+            orig = gr.Textbox(label=f"Original Line {i+1}", interactive=False)
+            sugg = gr.Textbox(label=f"Suggested Line {i+1}", interactive=False)
+            accept = gr.Checkbox(label="✅ Accept Suggestion")
+            originals.append(orig)
+            suggestions.append(sugg)
+            decisions.append(accept)
+            line_outputs.append((orig, sugg, accept))
+
+    final_output = gr.Textbox(label="📦 Final Clean Blog", lines=10)
+
+    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 "")
+
+    def process_review(text):
+        results = review_lines(text)
+        outputs = []
+        for i in range(len(line_outputs)):
+            if i < len(results):
+                outputs.extend([results[i][0], results[i][1], False])  # orig, suggestion, unselected
+            else:
+                outputs.extend(["", "", False])  # clear unused slots
+        return outputs
+
+    def collect_decisions(origs, suggs, accepts):
+        return finalize_text(origs, suggs, accepts)
+
+    # Wire actions
+    extract_btn.click(fn=extract_both, inputs=[blog_url, image_url], outputs=combined_text)
+    review_btn.click(fn=process_review, inputs=combined_text, outputs=[el for group in line_outputs for el in group])
+    finalize_btn.click(fn=collect_decisions, inputs=[originals, suggestions, decisions], outputs=final_output)
 
 app.launch()