Update app.py

app.py

@@ -863,7 +863,7 @@ async def generate_subtitle(
         if context:
             context_instruction = f"\n6.  **USER INSTRUCTIONS**: {context}\n"
 
-        prompt = f"""Task: Translate and Correct the SRT subtitles to Brazilian Portuguese (PT-BR).
+        prompt = f"""Translate and Correct the SRT subtitles to Brazilian Portuguese (PT-BR).
 
 SOURCE MATERIAL: I have attached the video/audio file. Use it as the PRIMARY source of truth for context, speaker identification, and meaning.
 INPUT SRT: I have provided a draft SRT generated by an automated transcription service.
@@ -871,8 +871,14 @@ INPUT SRT: I have provided a draft SRT generated by an automated transcription s
 CRITICAL INSTRUCTIONS:
 1.  **CORRECTION**: The input SRT is a draft. It may have wrong words or misinterpretations. **You MUST correct the text based on the actual audio.**
 2.  **TIMESTAMPS**: The timestamps in the input SRT are generally correct. **Keep them as much as possible**, only adjust if they are clearly wrong / desynchronized.
-3.  **FORMATTING**: Do NOT change the line breaks or the number of lines. The visual structure (2 lines max, balanced) is already optimized.
-4.  **CONTEXT**: Translate naturally for a Brazilian audience.
+3.  **FORMATTING**: 
+    - Keep the visual structure (2 lines max). 
+    - **DIALOGUES**: If two different people speak in the same block, separate them with a hyphen "- " at the start of each line.
+        *   Example Input: "I did it. So did I."
+        *   Example Output:
+            - Eu fiz isso.
+            - Eu também.
+4.  **CONTEXT**: Translate naturally for a Brazilian audience. Prioritize clarity and meaning over literal translation. Make it readable and elegant.
 5.  **PUNCTUATION**: Adjust punctuation to fit the context best.
 6.  **OUTPUT**: Return ONLY the corrected and translated SRT content. No extra text.{context_instruction}