Upload rewrite_slide.py

rewrite_slide.py

@@ -1,4 +1,4 @@
-"""Fast rewrite as slides - no AI needed, extracts key points + images from article."""
+"""Fast rewrite as slides - AI rewrites each key point line-by-line."""
 from main import app
 from fastapi import Request
 from fastapi.responses import JSONResponse
@@ -30,6 +30,34 @@ except:
 def _clean(s): return re.sub(r'\s+', ' ', str(s or '')).strip()
 
 
+def _ai_rewrite_line(text, title=""):
+    """Rewrite a single line using AI - keep meaning, rephrase naturally."""
+    hf_token = os.environ.get("HF_TOKEN", "")
+    if not hf_token:
+        return None
+    try:
+        api_url = "https://api-inference.huggingface.co/models/google/flan-t5-large"
+        headers = {"Authorization": f"Bearer {hf_token}"}
+        prompt = f"paraphrase keeping meaning: {text}"
+        payload = {
+            "inputs": prompt,
+            "parameters": {"max_length": 256, "temperature": 0.7, "do_sample": True}
+        }
+        r = requests.post(api_url, headers=headers, json=payload, timeout=20)
+        if r.status_code == 200:
+            result = r.json()
+            if isinstance(result, list) and len(result) > 0:
+                rewritten = result[0].get("generated_text", "").strip()
+                # Validate: must be different but similar length, in Vietnamese
+                if rewritten and len(rewritten) > 20 and rewritten != text:
+                    # Make sure it's not English
+                    if re.search(r'[àáạảãâầấậẩẫăằắặẳẵèéẹẻẽêềếệểễìíịỉĩòóọỏõôồốộổỗơờớợởỡùúụủũưừứựửữỳýỵỷỹđ]', rewritten):
+                        return rewritten
+    except:
+        pass
+    return None
+
+
 def _scrape_article_full(url):
     """Scrape article: extract paragraphs + ALL images."""
     try:
@@ -84,7 +112,7 @@ def _scrape_article_full(url):
         return None
 
 
-def _extract_key_points(paragraphs, max_points=5):
+def _extract_key_points(paragraphs, max_points=6):
     """Extract key points: take first sentence of each significant paragraph."""
     points = []
     for p in paragraphs:
@@ -108,12 +136,12 @@ def _extract_key_points(paragraphs, max_points=5):
 @app.post("/api/rewrite_slide")
 async def api_rewrite_slide(request: Request):
     """
-    Fast rewrite as SLIDES:
-    - Extract key points from article (1 sentence each, full and complete)
-    - Pair each point with an image from the article
+    Fast rewrite as SLIDES with AI:
+    - Extract key points from article (1 sentence each)
+    - AI rewrites EACH line individually (keep meaning, rephrase)
+    - Pair each rewritten point with an image from the article
     - Return as slides array for frontend to display
     - Save to Tường AI
-    NO AI NEEDED - instant response.
     """
     body = await request.json()
     url = _clean(body.get("url", ""))
@@ -140,16 +168,21 @@ async def api_rewrite_slide(request: Request):
     if not points:
         return JSONResponse({"error": "Không tìm được ý chính"}, status_code=422)
     
-    # Build slides: pair each point with an image
+    # Build slides: AI rewrite each line individually, pair with image
     images = data.get('images', [])
     slides = []
     for i, point in enumerate(points):
         img = images[i] if i < len(images) else (images[-1] if images else '')
-        # Proxy dantri images
-        if img and 'cdnphoto.dantri' in img:
+        # Proxy images that need it
+        if img and ('cdnphoto.dantri' in img or 'xemlaibongda' in img):
             img = '/api/proxy/img?url=' + quote(img, safe='')
+        
+        # AI rewrite this line only (keep meaning, rephrase naturally)
+        rewritten = _ai_rewrite_line(point, data.get('title', '')) or point
+        
         slides.append({
-            'text': point,
+            'text': rewritten,
+            'original': point,
             'image': img,
             'index': i + 1
         })