Upload app_v2_entry.py with huggingface_hub

app_v2_entry.py

@@ -1671,54 +1671,92 @@ async def api_personal_post_preview(request: Request):
         # Try to parse the ---SLIDE--- format
         pattern = r'---SLIDE\s*(\d+)---\s*\n(.*?)(?=---SLIDE|\Z)'
         matches = re.findall(pattern, ai_text, re.DOTALL)
+        
         if matches:
             for idx, (num, content) in enumerate(matches):
                 # Normalize: ensure complete sentences
                 text = _ensure_sentence_complete(content)
-                if len(text) > 30:
+                if len(text) > 40:
                     img = source_images[idx] if idx < len(source_images) else ""
                     slides.append({"text": text, "image": img, "index": idx + 1})
         
-        # Fallback: split by paragraphs
+        # If we have parsed slides, ensure minimum 3
         if len(slides) < 3:
-            paragraphs = [p.strip() for p in re.split(r'\n\n+', ai_text) if p.strip()]
-            slides = []
-            para_count = 0
-            for p in paragraphs:
-                p = p.strip()
-                if p.startswith('#') or p.startswith('---') or p.startswith('*Nguồn'):
+            # Use parsed slides as base, fill remaining from AI text
+            used_indices = set()
+            for s in slides:
+                used_indices.add(s['index'] - 1)
+            
+            # Split remaining AI text into more slides
+            sentences = re.split(r'(?<=[.!?])\s+', ai_text)
+            current_chunk = ""
+            next_idx = len(slides)
+            
+            for sent in sentences:
+                sent = _ensure_sentence_complete(sent)
+                if len(sent) < 20:
                     continue
-                # Combine multiple short paras into one slide, normalize
-                p_normalized = _ensure_sentence_complete(p)
-                if len(p_normalized) > 50:
-                    img = source_images[para_count] if para_count < len(source_images) else ""
-                    slides.append({"text": p_normalized, "image": img, "index": para_count + 1})
-                    para_count += 1
-                    if para_count >= 6:
+                
+                # Skip if this sentence is already in parsed slides
+                found = False
+                for slide in slides:
+                    if sent[:50] in slide['text']:
+                        found = True
                         break
+                
+                if found:
+                    continue
+                    
+                if current_chunk and len(current_chunk + " " + sent) <= 380:
+                    current_chunk += " " + sent
+                else:
+                    if len(current_chunk) > 50:
+                        img = source_images[next_idx] if next_idx < len(source_images) else ""
+                        slides.append({"text": current_chunk, "image": img, "index": next_idx + 1})
+                    current_chunk = sent
+                    next_idx += 1
+            
+            # Add final chunk
+            if len(current_chunk) > 50 and next_idx < 6:
+                img = source_images[next_idx] if next_idx < len(source_images) else ""
+                slides.append({"text": current_chunk, "image": img, "index": next_idx + 1})
     
+    # Ultimate fallback: create slides from opinion + source
     if len(slides) < 2:
-        # Fallback: create from opinion
-        slides = [{"text": f"Quan điểm: {opinion[:300]}", "image": source_images[0] if source_images else "", "index": 1}]
+        slides = []
+        # Slide 1: opinion
+        if opinion and len(opinion) > 20:
+            slides.append({"text": opinion[:450], "image": source_images[0] if source_images else "", "index": 1})
+        
+        # Slide 2-6: from AI text or sources
         if ai_text:
-            # Split by sentences for better chunks
             sentences = re.split(r'(?<=[.!?])\s+', ai_text)
-            current_chunk = ""
-            for sent in sentences:
-                sent = _ensure_sentence_complete(sent)
-                if len(sent) > 10 and current_chunk:
-                    if len(current_chunk + sent) < 400:
-                        current_chunk += " " + sent
-                    else:
-                        if len(current_chunk) > 50:
-                            idx = len(slides)
-                            img = source_images[idx] if idx < len(source_images) else ""
-                            slides.append({"text": current_chunk, "image": img, "index": idx + 1})
-                        current_chunk = sent
-            if len(current_chunk) > 50 and len(slides) < 6:
-                idx = len(slides)
-                img = source_images[idx] if idx < len(source_images) else ""
-                slides.append({"text": current_chunk, "image": img, "index": idx + 1})
+            for i, sent in enumerate(sentences[:5]):
+                text = _ensure_sentence_complete(_clean(sent))
+                if len(text) > 60:
+                    if len(slides) < 6:
+                        img = source_images[len(slides)] if len(slides) < len(source_images) else ""
+                        slides.append({"text": text, "image": img, "index": len(slides) + 1})
+        
+        # Fill remaining with key points from sources
+        src_idx = len(slides)
+        while len(slides) < 4 and src_idx < len(source_details):
+            paragraphs = source_details[src_idx].get("paragraphs", [])
+            for p in paragraphs[:2]:
+                if len(p) > 60 and len(slides) < 6:
+                    img = source_images[len(slides)] if len(slides) < len(source_images) else ""
+                    slides.append({"text": _ensure_sentence_complete(p[:400]), "image": img, "index": len(slides) + 1})
+            src_idx += 1
+        
+        # Final fallback: ensure at least 2-3 slides
+        while len(slides) < 3:
+            idx = len(slides)
+            if idx == 0 and opinion:
+                slides.append({"text": opinion[:400], "image": "", "index": 1})
+            elif ai_text:
+                slides.append({"text": ai_text[idx*300:(idx+1)*300], "image": "", "index": idx + 1})
+            else:
+                slides.append({"text": f"Nguồn tham khảo {idx + 1}", "image": "", "index": idx + 1})
     
     preview = {
         "title": title,
@@ -1726,7 +1764,7 @@ async def api_personal_post_preview(request: Request):
         "opinion": opinion,
         "images": source_images[:10],
         "sources": source_details[:5],
-        "slides": slides
+        "slides": slides[:6]  # Max 6 slides
     }
     
     return JSONResponse({"preview": preview})