diff --git a/src/gradio_runner.py b/src/gradio_runner.py
index ef8491d6d..1c49c36ed 100644
--- a/src/gradio_runner.py
+++ b/src/gradio_runner.py
@@ -132,6 +132,16 @@ def ask_block(kwargs, instruction_label, visible_upload, file_types, mic_sources
                     # info=None,
                     elem_id='prompt-form',
                     container=True,
+                    visible=False,
+                )
+                instruction_mm = gr.MultimodalTextbox(
+                    lines=kwargs['input_lines'],
+                    label=label_instruction,
+                    info=instruction_label,
+                    # info=None,
+                    elem_id='prompt-form',
+                    container=True,
+                    submit_btn=False,
                 )
                 mw0 = 20
                 mic_button = gr.Button(
@@ -263,7 +273,7 @@ def clear_audio_state():
                     .then(fn=lambda: None, **submit_kwargs)
                 stop_text.change(fn=clear_audio_state, outputs=audio_state) \
                     .then(fn=lambda: None, **stop_kwargs)
-    return attach_button, add_button, submit_buttons, instruction, submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn
+    return attach_button, add_button, submit_buttons, instruction, instruction_mm, submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn
 
 
 def go_gradio(**kwargs):
@@ -1093,17 +1103,25 @@ def get_df_langchain_mode_paths(selection_docs_state1, db1s, dbs1=None):
                     text_viewable_doc_count = gr.Textbox(lines=2, label=None, visible=False)
 
                 with gr.Accordion("Image/Video Query", open=False, visible=have_vision_models):
-                    image_file = gr.Image(value=kwargs['image_file'] if kwargs['image_file'] and any(
+                    image_file = gr.Gallery(value=kwargs['image_file'] if kwargs['image_file'] and any(
                         kwargs['image_file'].endswith(y) for y in IMAGE_EXTENSIONS) else None,
                                           label='Upload',
                                           show_label=False,
                                           type='filepath',
                                           elem_id="warning", elem_classes="feedback",
+                                          visible=False,
                                           )
                     video_file = gr.Video(value=None,
                                           label='Upload',
                                           show_label=False,
                                           elem_id="warning", elem_classes="feedback",
+                                          visible=False,
+                                          )
+                    audio_file = gr.Audio(value=None,
+                                          label='Upload',
+                                          show_label=False,
+                                          elem_id="warning", elem_classes="feedback",
+                                          visible=False,
                                           )
 
             col_tabs = gr.Column(elem_id="col-tabs", scale=10)
@@ -1156,7 +1174,8 @@ def get_df_langchain_mode_paths(selection_docs_state1, db1s, dbs1=None):
                         col_chat = gr.Column(visible=kwargs['chat'])
                         with col_chat:
                             if kwargs['visible_ask_anything_high']:
-                                attach_button, add_button, submit_buttons, instruction, submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn = \
+                                attach_button, add_button, submit_buttons, instruction, instruction_mm, \
+                                submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn = \
                                     ask_block(kwargs, instruction_label, visible_upload, file_types, mic_sources_kwargs,
                                               mic_kwargs, noqueue_kwargs2, submit_kwargs, stop_kwargs)
                             visible_model_choice = bool(kwargs['model_lock']) and \
@@ -1191,7 +1210,8 @@ def get_df_langchain_mode_paths(selection_docs_state1, db1s, dbs1=None):
                                                                                     **kwargs)
 
                             if not kwargs['visible_ask_anything_high']:
-                                attach_button, add_button, submit_buttons, instruction, submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn = \
+                                attach_button, add_button, submit_buttons, instruction, instruction_mm, \
+                                submit, retry_btn, undo, clear_chat_btn, save_chat_btn, stop_btn = \
                                     ask_block(kwargs, instruction_label, visible_upload, file_types, mic_sources_kwargs,
                                               mic_kwargs, noqueue_kwargs2, submit_kwargs, stop_kwargs)
                             with gr.Row():
@@ -2476,6 +2496,30 @@ def set_loaders(max_quality1,
                            inputs=max_quality,
                            outputs=[image_audio_loaders, pdf_loaders, url_loaders])
 
+        def fun_mm(instruction_mm1, instruction1, image_file1, video_file1, audio_file1):
+            if 'text' in instruction_mm1:
+                instruction1 = instruction_mm1['text']
+            if 'url' in instruction_mm1:
+                instruction1 = instruction_mm1['url']
+            if 'files' in instruction_mm1:
+                image_file2 = [x for x in instruction_mm1['files'] if any(x.endswith(y) for y in IMAGE_EXTENSIONS)]
+                if image_file2:
+                    image_file1 = image_file2
+                video_file2 = [x for x in instruction_mm1['files'] if any(x.endswith(y) for y in VIDEO_EXTENSIONS)]
+                if video_file2:
+                    video_file1 = video_file2
+                if isinstance(video_file1, list):
+                    if video_file1:
+                        video_file1 = video_file1[0]
+                    else:
+                        video_file1 = None
+            return instruction1, image_file1, video_file1, audio_file1
+
+        instruction_mm.submit(fn=fun_mm,
+                              inputs=[instruction_mm, instruction, image_file, video_file, audio_file],
+                              outputs=[instruction, image_file, video_file, audio_file]). \
+                              then(lambda: gr.MultimodalTextbox(interactive=True), None, [instruction_mm])
+
         # Add to UserData or custom user db
         update_db_func = functools.partial(update_user_db_gr,
                                            dbs=dbs,
diff --git a/src/version.py b/src/version.py
index c3919a9dd..5744f4cc5 100644
--- a/src/version.py
+++ b/src/version.py
@@ -1 +1 @@
-__version__ = "e1b7e1f5e5a1382b59713ed023298e96bf9df0c0"
+__version__ = "273bc363ef61006b47940650cd6cf7096648a24c"