Update app.py

app.py

@@ -228,11 +228,16 @@ def extract_fields_from_image(image_path, progress=None):
         progress(0.25, desc="Preparing prompt...")
 
     # Prompt engineering: Carefully crafted prompt for multilingual form extraction
-    # prompt = "<image_soft_token> Extract all the form field labels from this image. Return ONLY a JSON object where keys are field names and values are 'text'. For Hindi/Devanagari forms, preserve the original script. Do NOT include any introductory or concluding text."
-    # print(f"[Prompt] Using engineered prompt: {prompt[:80]}...")
-
-    prompt = "Extract all form field labels from this image <image>. Return ONLY a JSON object where keys are field names and values are 'text'. For Hindi/Devanagari forms, preserve the original script. Do NOT include any introductory or concluding text."
-    messages = [{"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": prompt}]}]
+    prompt = (
+    "<image_soft_token> Extract ALL possible form field labels from this image. "
+    "Return ONLY a JSON object where each key is a field name and each value is 'text'. "
+    "For Hindi/Devanagari forms, preserve the original script. "
+    "Do NOT include any introductory or concluding text. "
+    "If a field is partially visible, still include it. "
+    "Sample output: {\"ग्राम\": \"text\", \"उपकेन्द्र\": \"text\", \"आयु\": \"text\", \"लिंग\": \"text\", ...}"
+    )
+    #prompt = "<image_soft_token> Extract all the form field labels from this image. Return ONLY a JSON object where keys are field names and values are 'text'. For Hindi/Devanagari forms, preserve the original script. Do NOT include any introductory or concluding text."
+    print(f"[Prompt] Using engineered prompt: {prompt[:80]}...")
 
 
     if progress: