diff --git a/install.py b/install.py
index 8f2563a..836b382 100644
--- a/install.py
+++ b/install.py
@@ -20,4 +20,7 @@ if not launch.is_installed('llama-cpp-python'):
     if os.name == 'nt':
         launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.23+cu118-cp310-cp310-win_amd64.whl', 'llama-cpp-python')
     else:
-        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.23+cu118-cp310-cp310-manylinux_2_31_x86_64.whl', 'llama-cpp-python')
\ No newline at end of file
+        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.23+cu118-cp310-cp310-manylinux_2_31_x86_64.whl', 'llama-cpp-python')
+
+if not launch.is_installed('gpt-stream-json-parser'):
+    launch.run_pip('install git+https://github.com/furnqse/gpt-stream-json-parser.git', 'gpt-stream-json-parser')
\ No newline at end of file
diff --git a/scripts/chatgptapi.py b/scripts/chatgptapi.py
index 020e2a3..a831597 100644
--- a/scripts/chatgptapi.py
+++ b/scripts/chatgptapi.py
@@ -5,6 +5,7 @@ import os
 import openai
 import sys
 import json
+from gpt_stream_parser import force_parse_json
 
 class ChatGptApi:
     chatgpt_messages = []
@@ -29,6 +30,8 @@ class ChatGptApi:
         },
     }]
     model = 'gpt-3.5-turbo'
+    recieved_json = ''
+    recieved_message = ''
 
     def __init__(self, model=None, apikey=None):
         if model is not None:
@@ -81,20 +84,26 @@ class ChatGptApi:
         self.chatgpt_response = openai.ChatCompletion.create(
             model=self.model,
             messages=self.chatgpt_messages,
-            functions=self.chatgpt_functions
+            functions=self.chatgpt_functions,
+            stream=True,
         )
         ignore_result = False
-        result = str(self.chatgpt_response["choices"][0]["message"]["content"])
+        self.recieved_json = ''
+        self.recieved_message = ''
+        for chunk in self.chatgpt_response:
+            if 'function_call' in chunk.choices[0].delta and chunk.choices[0].delta.function_call is not None and 'arguments' in chunk.choices[0].delta.function_call:
+                self.recieved_json += chunk.choices[0].delta.function_call.arguments
+            else:
+                self.recieved_message += chunk.choices[0].delta.get('content', '')
+        result = self.recieved_message
         prompt = None
-        if "function_call" in self.chatgpt_response["choices"][0]["message"].keys():
-            function_call = self.chatgpt_response["choices"][0]["message"]["function_call"]
-            if function_call is not None and function_call["name"] == "txt2img":
-                func_args = json.loads(function_call["arguments"])
-                prompt = func_args["prompt"]
-                if "message" in func_args:
-                    result = func_args["message"]
-                else:
-                    ignore_result = True
+        if self.recieved_json != '':
+            func_args = json.loads(self.recieved_json)
+            prompt = func_args["prompt"]
+            if "message" in func_args:
+                result = func_args["message"]
+            else:
+                ignore_result = True
         self.chatgpt_response = None
         if prompt is None:
             self.chatgpt_messages.append({"role": "assistant", "content": result})
@@ -118,4 +127,13 @@ class ChatGptApi:
     def clear(self):
         self.chatgpt_messages = []
         self.chatgpt_response = None
-        self.log_file_name = None
\ No newline at end of file
+        self.log_file_name = None
+
+    def get_stream(self):
+        if self.recieved_json == '':
+            return self.recieved_message, None
+        func_args = force_parse_json(self.recieved_json)
+        if func_args is not None and "message" in func_args:
+            return func_args["message"], func_args["prompt"]
+        else:
+            return None, None
diff --git a/scripts/langchainapi.py b/scripts/langchainapi.py
index 3d3eac7..16951bc 100644
--- a/scripts/langchainapi.py
+++ b/scripts/langchainapi.py
@@ -243,4 +243,7 @@ If you understand, please reply to the following:<|end_of_turn|>
 
     def clear(self):
         self.memory.chat_memory.clear()
-        self.log_file_name = None
\ No newline at end of file
+        self.log_file_name = None
+
+    def get_stream(self):
+        return None, None
\ No newline at end of file
diff --git a/scripts/main.py b/scripts/main.py
index 2caf0ac..4e87e07 100644
--- a/scripts/main.py
+++ b/scripts/main.py
@@ -9,6 +9,7 @@ import uuid
 import copy
 import inspect
 import sys
+import time
 import gradio as gr
 from PIL import PngImagePlugin
 from modules.scripts import basedir
@@ -234,9 +235,12 @@ def on_ui_tabs():
         images[0].save(last_image_name, pnginfo=(metadata if use_metadata else None))
 
     def append_chat_history(chat_history, text_input_str, result, prompt):
-        global last_image_name, chat_history_images
+        global last_image_name, chat_history_images, txt2img_thread
         if prompt is not None and prompt != '':
-            chatgpt_txt2img(prompt)
+            if txt2img_thread is None:
+                chatgpt_txt2img(prompt)
+            else:
+                txt2img_thread.join()
             if result is None:
                 chat_history_images[len(chat_history)] = last_image_name
                 chat_history.append((text_input_str, (last_image_name, )))
@@ -248,12 +252,54 @@ def on_ui_tabs():
             chat_history.append((text_input_str, result))
         return chat_history
 
-    def chatgpt_generate(text_input_str: str, chat_history):
-        result, prompt = chat_gpt_api.send(text_input_str)
+    def recv_stream(chat_history):
+        global txt2img_thread, stop_recv_thread
+        txt2img_thread = None
+        while not stop_recv_thread:
+            message, prompt = chat_gpt_api.get_stream()
+            if prompt is not None and prompt != '' and txt2img_thread is None:
+                txt2img_thread = threading.Thread(target=chatgpt_txt2img, args=(prompt, ))
+                txt2img_thread.start()
+            if message is not None:
+                chat_history[-1][1] = message
+            time.sleep(0.01)
 
-        chat_history = append_chat_history(chat_history, text_input_str, result, prompt)
+    def chatgpt_generate(chat_history):
+        global stop_recv_thread, chatgpt_generate_result, chatgpt_generate_prompt
+        stop_recv_thread = False
 
-        return [last_image_name, info_html, comments_html, info_html.replace('<br>', '\n').replace('<p>', '').replace('</p>', '\n').replace('&lt;', '<').replace('&gt;', '>'), '', chat_history]
+        text_input_str = chat_history[-1][0]
+
+        def recv_thread_func():
+            recv_stream(chat_history)
+        thread = threading.Thread(target=recv_thread_func)
+        thread.start()
+
+        chatgpt_generate_result = None
+        chatgpt_generate_prompt = None
+        def send_thread_func():
+            global stop_recv_thread, chatgpt_generate_result, chatgpt_generate_prompt
+            chatgpt_generate_result, chatgpt_generate_prompt = chat_gpt_api.send(text_input_str)
+            stop_recv_thread = True
+        thread2 = threading.Thread(target=send_thread_func)
+        thread2.start()
+
+        prev_message = chat_history[-1][1]
+        while not stop_recv_thread:
+            if prev_message != chat_history[-1][1]:
+                prev_message = chat_history[-1][1]
+                yield chat_history
+            time.sleep(0.01)
+
+        thread.join()
+
+        chat_history = chat_history[:-1]
+        chat_history = append_chat_history(chat_history, text_input_str, chatgpt_generate_result, chatgpt_generate_prompt)
+
+        yield chat_history
+
+    def chatgpt_generate_finished():
+        return [last_image_name, info_html, comments_html, info_html.replace('<br>', '\n').replace('<p>', '').replace('</p>', '\n').replace('&lt;', '<').replace('&gt;', '>'), '']
 
     def chatgpt_remove_last(text_input_str: str, chat_history):
         if chat_history is None or len(chat_history) <= 0:
@@ -463,9 +509,21 @@ def on_ui_tabs():
                     txt2img_params_base = json.loads(txt2img_params_json)
                 btn_settings_reflect.click(fn=json_reflect, inputs=txt_json_settings)
         
-        btn_generate.click(fn=chatgpt_generate,
-            inputs=[text_input, chatbot],
-            outputs=[image_gr, info_html_gr, comments_html_gr, info_text_gr, text_input, chatbot])
+        btn_generate.click(
+                fn=lambda t, c: ['', c + [(t, None)]], 
+                inputs=[text_input, chatbot],
+                outputs=[text_input, chatbot],
+                #queue=False,
+            ).then(
+                fn=chatgpt_generate,
+                inputs=chatbot,
+                outputs=chatbot,
+                #queue=False,
+            ).then(
+                fn=chatgpt_generate_finished,
+                outputs=[image_gr, info_html_gr, comments_html_gr, info_text_gr, text_input],
+                #queue=False,
+            )
         btn_regenerate.click(fn=chatgpt_regenerate,
             inputs=chatbot,
             outputs=[image_gr, info_html_gr, comments_html_gr, info_text_gr, chatbot])