Update app.py

app.py

@@ -39,7 +39,21 @@ def ocr_on_image(image):
     prompt2 =(  "Extract all visible text from the image, including both handwritten and printed content."
                 "Do not translate the text — preserve the original language exactly as it appears."
                 "Return only the extracted text, with no explanation, no formatting, and no additions." )
-    messages = [{"role": "user", "content": [{"type": "text", "text": prompt2}, {"type": "image"}]}]
+    prompt3 = (
+    "Output ONLY the raw text as it appears in the image, nothing else."
+    "You have an image containing both handwritten and printed text in French and/or English, and alsos punctuation and underscores.\n"
+    "Your task: transcribe EXACTLY all visible text, preserving all characters, accents, punctuation, spacing, and line breaks.\n"
+    "Include tables and forms clearly if present.\n"
+    "Do NOT add any explanations, comments, summaries, or extra text.\n"
+    "Check the output first to not duplicate results."
+    "Preserve the original reading order, including line breaks and the natural layout of tables or forms. Output the text exactly as it appears visually, maintaining the structure."
+    "Don't indicate blank space."
+    "Don't separate handwritten and printex text."
+    "DO NOT confuse between '.' a point and '|' a boder"
+    "Extract only the raw text with and do not add any comment"
+    "Extract only the data available"
+    )
+    messages = [{"role": "user", "content": [{"type": "text", "text": prompt3}, {"type": "image"}]}]
     texts = processor.apply_chat_template(messages, add_generation_prompt=True)
     inputs = processor(text=texts, images=[image], return_tensors="pt").to(device)
     outputs = model.generate(**inputs, max_new_tokens=250)