diff --git a/modules/chat.py b/modules/chat.py
index 881f7330..2db72f36 100644
--- a/modules/chat.py
+++ b/modules/chat.py
@@ -578,7 +578,6 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
     visible_text = None
     stopping_strings = get_stopping_strings(state)
     is_stream = state['stream']
-
     # Prepare the input
     if not (regenerate or _continue):
         visible_text = html.escape(text)
@@ -598,8 +597,6 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
         # Apply extensions
         text, visible_text = apply_extensions('chat_input', text, visible_text, state)
         text = apply_extensions('input', text, state, is_chat=True)
-
-        # Current row index
         output['internal'].append([text, ''])
         output['visible'].append([visible_text, ''])
         # Add metadata with timestamp
@@ -644,16 +641,14 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
                     'internal': output['internal'],
                     'metadata': output['metadata']
                 }
-
     # Generate the prompt
     kwargs = {
         '_continue': _continue,
         'history': output if _continue else {
             k: (v[:-1] if k in ['internal', 'visible'] else v)
             for k, v in output.items()
-        }
     }
-
+    }
     prompt = apply_extensions('custom_generate_chat_prompt', text, state, **kwargs)
     if prompt is None:
         prompt = generate_chat_prompt(text, state, **kwargs)
@@ -665,27 +660,53 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
     # Generate
     reply = None
     for j, reply in enumerate(generate_reply(prompt, state, stopping_strings=stopping_strings, is_chat=True, for_ui=for_ui)):
+        # Handle start_with text (add to both internal and visible)
+        if state.get('start_with', '') and j == 0 and not _continue and not regenerate:
+            start_text = state['start_with']
+            # Add to internal (English) version
+            reply = start_text + " " + reply
+            # Add to visible (translated) version
+            translated_start = apply_extensions('output', start_text, state, is_chat=True)
+            state['start_with'] = ''  # Clear it after using
 
         # Extract the reply
         if state['mode'] in ['chat', 'chat-instruct']:
             visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
         else:
             visible_reply = reply
-
         visible_reply = html.escape(visible_reply)
-
         if shared.stop_everything:
             output['visible'][-1][1] = apply_extensions('output', output['visible'][-1][1], state, is_chat=True)
             yield output
             return
 
         if _continue:
-            output['internal'][-1] = [text, last_reply[0] + reply]
-            output['visible'][-1] = [visible_text, last_reply[1] + visible_reply]
-        elif not (j == 0 and visible_reply.strip() == ''):
-            output['internal'][-1] = [text, reply.lstrip(' ')]
-            output['visible'][-1] = [visible_text, visible_reply.lstrip(' ')]
+            # Separate already existing content from new content
+            original_internal = output['internal'][-1][1]
+            original_visible = output['visible'][-1][1]
 
+            # Get only the new generated part
+            new_content = reply[len(original_internal):] if reply.startswith(original_internal) else reply
+            new_content = new_content.lstrip()
+
+            # Translate only the new part
+            translated_new = apply_extensions('output', new_content, state, is_chat=True)
+
+            # Update both internal and visible versions
+            updated_internal = original_internal + " " + new_content
+            updated_visible = original_visible + " " + translated_new
+
+            output['internal'][-1] = [text, updated_internal]
+            output['visible'][-1] = [visible_text, updated_visible]
+
+            if is_stream:
+                yield output
+        elif not (j == 0 and visible_reply.strip() == ''):
+            # For normal generation, translate the whole reply
+            translated_reply = apply_extensions('output', visible_reply.lstrip(' '), state, is_chat=True)
+            output['internal'][-1] = [text, reply.lstrip(' ')]
+            output['visible'][-1] = [visible_text, translated_reply]
+            
         # Keep version metadata in sync during streaming (for regeneration)
         if regenerate:
             row_idx = len(output['internal']) - 1
@@ -695,11 +716,8 @@ def chatbot_wrapper(text, state, regenerate=False, _continue=False, loading_mess
                 'content': output['internal'][row_idx][1],
                 'visible_content': output['visible'][row_idx][1]
             })
-
-        if is_stream:
-            yield output
-
-    output['visible'][-1][1] = apply_extensions('output', output['visible'][-1][1], state, is_chat=True)
+            if is_stream:
+                yield output
 
     # Final sync for version metadata (in case streaming was disabled)
     if regenerate: