Spaces:
Runtime error
Runtime error
Dy
commited on
Commit
•
2ac6b63
1
Parent(s):
f18bf59
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,314 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from bs4 import BeautifulSoup
|
3 |
+
import gradio as gr
|
4 |
+
from langchain import OpenAI, ConversationChain, LLMChain, PromptTemplate
|
5 |
+
from langchain.memory import ConversationBufferWindowMemory
|
6 |
+
import openai
|
7 |
+
import requests
|
8 |
+
from langchain.chat_models import ChatOpenAI
|
9 |
+
import ast
|
10 |
+
import imgkit
|
11 |
+
import pdfkit
|
12 |
+
import imgkit
|
13 |
+
import re
|
14 |
+
import glob
|
15 |
+
|
16 |
+
import openai
|
17 |
+
|
18 |
+
|
19 |
+
OPENAI_API_KEY="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"
|
20 |
+
|
21 |
+
|
22 |
+
dict_list_format = '''[{'header': 'slide1_title',
|
23 |
+
'paragraphs': ['bullet_point1',
|
24 |
+
'bullet_point2',
|
25 |
+
'bullet_point3',]},
|
26 |
+
'header': 'slide2_title',
|
27 |
+
'paragraphs': ['bullet_point1',
|
28 |
+
'bullet_point2',
|
29 |
+
'bullet_point3',
|
30 |
+
...]},
|
31 |
+
'header': 'slide3_title',
|
32 |
+
'paragraphs': ['bullet_point1',
|
33 |
+
'bullet_point2',
|
34 |
+
'bullet_point3',
|
35 |
+
...]},
|
36 |
+
'header': 'slide4_title',
|
37 |
+
'paragraphs': ['bullet_point1',
|
38 |
+
'bullet_point2',
|
39 |
+
'bullet_point3',
|
40 |
+
...]},
|
41 |
+
'header': 'slide5_title',
|
42 |
+
'paragraphs': ['bullet_point1',
|
43 |
+
'bullet_point2',
|
44 |
+
'bullet_point3',
|
45 |
+
...]}]
|
46 |
+
'''
|
47 |
+
|
48 |
+
import google.cloud.texttospeech as tts
|
49 |
+
from google.oauth2 import service_account
|
50 |
+
credentials = service_account.Credentials.from_service_account_file("tts_google.json")
|
51 |
+
|
52 |
+
def text_to_wav(voice_name: str, text: str, file_name: str):
|
53 |
+
language_code = "-".join(voice_name.split("-")[:2])
|
54 |
+
text_input = tts.SynthesisInput(text=text)
|
55 |
+
voice_params = tts.VoiceSelectionParams(
|
56 |
+
language_code=language_code, name=voice_name
|
57 |
+
)
|
58 |
+
audio_config = tts.AudioConfig(audio_encoding=tts.AudioEncoding.LINEAR16)
|
59 |
+
|
60 |
+
client = tts.TextToSpeechClient(credentials=credentials)
|
61 |
+
response = client.synthesize_speech(
|
62 |
+
input=text_input,
|
63 |
+
voice=voice_params,
|
64 |
+
audio_config=audio_config,
|
65 |
+
)
|
66 |
+
|
67 |
+
filename = f"{file_name}"
|
68 |
+
with open(filename, "wb") as out:
|
69 |
+
out.write(response.audio_content)
|
70 |
+
print(f'Generated speech saved to "{filename}"')
|
71 |
+
|
72 |
+
|
73 |
+
|
74 |
+
def prompt_to_video(video_prompt):
|
75 |
+
|
76 |
+
template = '''
|
77 |
+
{history}
|
78 |
+
{human_input}
|
79 |
+
'''
|
80 |
+
prompt = PromptTemplate(
|
81 |
+
input_variables=["history", "human_input"],
|
82 |
+
template=template
|
83 |
+
)
|
84 |
+
|
85 |
+
chatgpt_chain = LLMChain(
|
86 |
+
llm=ChatOpenAI(model="gpt-4", temperature=0.5,openai_api_key=OPENAI_API_KEY),
|
87 |
+
prompt=prompt,
|
88 |
+
verbose=True,
|
89 |
+
memory=ConversationBufferWindowMemory(k=10),
|
90 |
+
)
|
91 |
+
|
92 |
+
prompt_input1 = f'''
|
93 |
+
You are a world expert oracle that knows everything.
|
94 |
+
You are also an excellent teacher that explains everything succintly and simply like towards a kid.
|
95 |
+
You are also an expert slide maker and think everything step by step.
|
96 |
+
You are tasked to create 5 slides today.
|
97 |
+
|
98 |
+
Here is the topic:
|
99 |
+
{video_prompt}
|
100 |
+
|
101 |
+
Here is the output python list format:
|
102 |
+
{dict_list_format}
|
103 |
+
|
104 |
+
The slides should be created in a python list format.
|
105 |
+
The list consists of python dictionary objects in the list.
|
106 |
+
Each dictionary object contains the header and paragraphs as keys.
|
107 |
+
Do not name the slide as "Slide 1" or any number. Insert header as header string.
|
108 |
+
The header is the title of the slide and the paragraph should be a list of string object.
|
109 |
+
Return the output in a python list format.
|
110 |
+
Make sure there is only 5 objects in the python list.
|
111 |
+
Do not declare a new variable, output the python list object only.
|
112 |
+
Do not say "Here's your". Directly output the python list object only.
|
113 |
+
Make sure there is nothing before or after the python list object. ONLY output the python list object.
|
114 |
+
'''
|
115 |
+
slide_str_list = []
|
116 |
+
while len(slide_str_list) != 5:
|
117 |
+
slide_dict=chatgpt_chain.predict(human_input=prompt_input1)
|
118 |
+
try:
|
119 |
+
slide_str_list = ast.literal_eval(slide_dict)
|
120 |
+
except:
|
121 |
+
print("Already formatted.")
|
122 |
+
|
123 |
+
|
124 |
+
print("this is the slides:", slide_str_list)
|
125 |
+
print("length is:", len(slide_str_list))
|
126 |
+
|
127 |
+
html_out_list = []
|
128 |
+
for i in slide_str_list:
|
129 |
+
|
130 |
+
template = '''
|
131 |
+
{history}
|
132 |
+
{human_input}
|
133 |
+
'''
|
134 |
+
prompt = PromptTemplate(
|
135 |
+
input_variables=["history", "human_input"],
|
136 |
+
template=template
|
137 |
+
)
|
138 |
+
|
139 |
+
chatgpt_chain = LLMChain(
|
140 |
+
llm=ChatOpenAI(model="gpt-3.5-turbo", temperature=0.5,openai_api_key=OPENAI_API_KEY),
|
141 |
+
prompt=prompt,
|
142 |
+
verbose=True,
|
143 |
+
memory=ConversationBufferWindowMemory(k=10),
|
144 |
+
)
|
145 |
+
prompt_input2 = f'''
|
146 |
+
You are a world expert oracle that knows everything.
|
147 |
+
You are also an excellent teacher that explains everything succintly and simply like towards a kid.
|
148 |
+
You are also an expert slide maker and thinks about everything step by step.
|
149 |
+
You are tasked to convert a python dictionary into a formatted HTML code.
|
150 |
+
The dictionary object consist of the header and paragraph key.
|
151 |
+
The paragraph key is a list of strings.
|
152 |
+
Here is the dictionary object:
|
153 |
+
{i}
|
154 |
+
The slide should be created in a HTML format with the correct format of 16:9 aspect ratio.
|
155 |
+
The wording of the slides should be formatted appropriately with the header and paragraph.
|
156 |
+
The paragraph in the slides should be formatted in bullet points and each bullet point should be 1.5 line spacing apart.
|
157 |
+
Header and paragraph should be aligned in an aesthetically pleasing way.
|
158 |
+
Return the output as a nicely formatted HTML string.
|
159 |
+
Font should be Roboto.
|
160 |
+
Do not say "Here's your" or "Sure". Directly output the HTML string only.
|
161 |
+
Make sure there is nothing before or after the HTML string. ONLY output the HTML string.
|
162 |
+
Do not explain what is the HTML code about.
|
163 |
+
Do not declare a new variable, output the HTML string only.
|
164 |
+
'''
|
165 |
+
html_out_list.append(chatgpt_chain.predict(human_input=prompt_input2))
|
166 |
+
|
167 |
+
os.makedirs("slide", exist_ok=True)
|
168 |
+
|
169 |
+
num = 1
|
170 |
+
for html_string in html_out_list:
|
171 |
+
print(html_string)
|
172 |
+
with open(f"slide/slide_{num}.html", "w") as file:
|
173 |
+
file.write(html_string)
|
174 |
+
num = num + 1
|
175 |
+
|
176 |
+
extract_path = 'slide'
|
177 |
+
|
178 |
+
# Create the directory to extract to if it doesn't exist
|
179 |
+
os.makedirs(extract_path, exist_ok=True)
|
180 |
+
|
181 |
+
# Configuration for imgkit
|
182 |
+
config = imgkit.config(wkhtmltoimage='C:\\Program Files\\wkhtmltopdf\\bin\\wkhtmltoimage.exe')
|
183 |
+
|
184 |
+
# The path to store the images
|
185 |
+
image_path = os.path.join(extract_path, 'images')
|
186 |
+
os.makedirs(image_path, exist_ok=True)
|
187 |
+
|
188 |
+
# Get the list of HTML files
|
189 |
+
html_files = sorted([f for f in os.listdir(extract_path) if f.endswith('.html')])
|
190 |
+
|
191 |
+
# Dictionary to store the file names and their corresponding images
|
192 |
+
file_images = {}
|
193 |
+
|
194 |
+
# Loop through the HTML files and convert them to images
|
195 |
+
for html_file in html_files:
|
196 |
+
# Full path of the HTML file
|
197 |
+
full_path = os.path.join(extract_path, html_file)
|
198 |
+
|
199 |
+
# Image file name
|
200 |
+
image_file = re.sub('.html$', '.jpg', html_file)
|
201 |
+
|
202 |
+
# Full path of the image file
|
203 |
+
full_image_path = os.path.join(image_path, image_file)
|
204 |
+
|
205 |
+
# Convert the HTML to an image
|
206 |
+
imgkit.from_file(full_path, full_image_path, config=config)
|
207 |
+
|
208 |
+
# Store the image file name
|
209 |
+
file_images[html_file] = image_file
|
210 |
+
|
211 |
+
print(file_images)
|
212 |
+
|
213 |
+
template = '''
|
214 |
+
{history}
|
215 |
+
{human_input}
|
216 |
+
'''
|
217 |
+
prompt = PromptTemplate(
|
218 |
+
input_variables=["history", "human_input"],
|
219 |
+
template=template
|
220 |
+
)
|
221 |
+
|
222 |
+
chatgpt_chain = LLMChain(
|
223 |
+
llm=ChatOpenAI(model="gpt-4", temperature=0.5,openai_api_key=OPENAI_API_KEY),
|
224 |
+
prompt=prompt,
|
225 |
+
verbose=True,
|
226 |
+
memory=ConversationBufferWindowMemory(k=10),
|
227 |
+
)
|
228 |
+
|
229 |
+
prompt_input3 = f'''
|
230 |
+
You are a world expert oracle that knows everything.
|
231 |
+
You are also an excellent teacher that explains everything succintly and simply like towards a kid.
|
232 |
+
You are an expert orator and presenter.
|
233 |
+
You are tasked to create a voiceover for 5 slides.
|
234 |
+
The slides are formatted in a python list of dictionary objects.
|
235 |
+
Each dictionary object is a slide.
|
236 |
+
{slide_str_list}
|
237 |
+
|
238 |
+
Input: Python list of dictionary objects
|
239 |
+
Output: Python list of string objects
|
240 |
+
|
241 |
+
The output list consists of string objects.
|
242 |
+
The voiceover text purpose is a speech presentation of the slide.
|
243 |
+
The voiceover text should be about the content of each slide but at the same time add additional information to make the presentation funny and engaging.
|
244 |
+
Each string is a voiceover text of each slide of the python dictionary.
|
245 |
+
Each voiceover string object should be around 80 words.
|
246 |
+
|
247 |
+
Make sure there is only 5 objects in the python list.
|
248 |
+
Do not declare a new variable, output the python list object only.
|
249 |
+
Make sure there is nothing before or after the python list object. ONLY output the python list object.
|
250 |
+
Return the output in a python list format.
|
251 |
+
Do not say "Here's your" or "Sure". Directly output python list of dictionary object only.
|
252 |
+
Do not declare a new variable, output the python list of dictionary object only.
|
253 |
+
'''
|
254 |
+
|
255 |
+
voiceover_list = []
|
256 |
+
|
257 |
+
voiceover_list=chatgpt_chain.predict(human_input=prompt_input3)
|
258 |
+
|
259 |
+
try:
|
260 |
+
voiceover_list = ast.literal_eval(voiceover_list)
|
261 |
+
except:
|
262 |
+
print("Already formatted.")
|
263 |
+
|
264 |
+
|
265 |
+
num = 1
|
266 |
+
for i in voiceover_list:
|
267 |
+
file_name = "slide/slide" + f"_{num}" + ".wav"
|
268 |
+
text_to_wav("en-US-Neural2-F",i, file_name)
|
269 |
+
print(file_name)
|
270 |
+
num = num + 1
|
271 |
+
|
272 |
+
|
273 |
+
# Get list of .jpg and .wav files from the correct directories
|
274 |
+
jpg_files = sorted(glob.glob("slide/images/*.jpg"))
|
275 |
+
wav_files = sorted(glob.glob("slide/*.wav"))
|
276 |
+
|
277 |
+
jpg_files, wav_files
|
278 |
+
|
279 |
+
from moviepy.editor import ImageSequenceClip, AudioFileClip, concatenate_videoclips
|
280 |
+
|
281 |
+
# Create a list to store the clips
|
282 |
+
clips = []
|
283 |
+
|
284 |
+
# Loop through each jpg and wav file
|
285 |
+
for jpg_file, wav_file in zip(jpg_files, wav_files):
|
286 |
+
# Load the audio file and get its duration
|
287 |
+
audio = AudioFileClip(wav_file)
|
288 |
+
duration = audio.duration
|
289 |
+
print(duration)
|
290 |
+
|
291 |
+
# Calculate the frame rate as the inverse of the duration
|
292 |
+
fps = 1 / duration if duration != 0 else 1
|
293 |
+
|
294 |
+
# Create a video clip from the image and set its duration and fps to match the audio
|
295 |
+
clip = ImageSequenceClip([jpg_file], durations=[duration], fps=fps)
|
296 |
+
|
297 |
+
# Set the audio of the clip to the wav file
|
298 |
+
clip = clip.set_audio(audio)
|
299 |
+
|
300 |
+
# Add the clip to the list of clips
|
301 |
+
clips.append(clip)
|
302 |
+
|
303 |
+
# Concatenate all clips into a single video
|
304 |
+
video = concatenate_videoclips(clips)
|
305 |
+
|
306 |
+
video_path = "slide/output.mp4"
|
307 |
+
|
308 |
+
# Write the video to a file
|
309 |
+
video.write_videofile(video_path)
|
310 |
+
|
311 |
+
|
312 |
+
return video_path
|
313 |
+
iface = gr.Interface(fn=prompt_to_video, inputs="text", outputs=["file"])
|
314 |
+
iface.launch()
|