Joon’s API Special Lecture[2024.06.16. 14:00-18:00]

import streamlit as st from pytube import YouTube from moviepy.editor import * import google.generativeai as genai from datetime import datetime import time import os from PIL import Image from streamlit_player import st_player from googleapiclient.discovery import build from googleapiclient.errors import HttpError import pandas as pd import re import mock from pytube.cipher import get_throttling_function_code from pytube import YouTube st.set_page_config( page_title="나만의 YouTube 요약 서비스", page_icon="🎬", ) google_youtube_api_key = "AIzaSyAuKiyYNDPq_Q3Sf0FLw7onz6vIifsB7NQ" api_key_index = 0 def patched_throttling_plan(js: str): raw_code = get_throttling_function_code(js) transform_start = r"try{" plan_regex = re.compile(transform_start) match = plan_regex.search(raw_code) transform_plan_raw = js step_start = r"c\[(\d+)\]\(c\[(\d+)\](,c(\[(\d+)\]))?\)" step_regex = re.compile(step_start) matches = step_regex.findall(transform_plan_raw) transform_steps = [] for match in matches: if match[4] != '': transform_steps.append((match[0],match[1],match[4])) else: transform_steps.append((match[0],match[1])) return transform_steps def get_comments(youtube, video_id): comments = [] next_page_token = None while True: response = youtube.commentThreads().list( part="snippet", videoId=video_id, pageToken=next_page_token, maxResults=100, textFormat="plainText" ).execute() for item in response["items"]: comment = item["snippet"]["topLevelComment"]["snippet"] comments.append({ "authorDisplayName": comment["authorDisplayName"], "textDisplay": comment["textDisplay"], "likeCount": comment["likeCount"], "publishedAt": comment["publishedAt"] }) comment_id = item["snippet"]["topLevelComment"]["id"] reply_response = youtube.comments().list( part="snippet", parentId=comment_id, textFormat="plainText" ).execute() for reply in reply_response["items"]: reply_comment = reply["snippet"] comments.append({ "authorDisplayName": reply_comment["authorDisplayName"], "textDisplay": reply_comment["textDisplay"], "likeCount": reply_comment["likeCount"], "publishedAt": reply_comment["publishedAt"] }) next_page_token = response.get("nextPageToken") if not next_page_token: return pd.DataFrame(comments) def get_video_details(youtube, video_id): request = youtube.videos().list( part=["snippet","statistics","contentDetails"], id=video_id ) response = request.execute() for item in response["items"]: data = dict(contentDetails=item["snippet"]["description"]) if "likeCount" in item["statistics"]: data["like_count"] = item["statistics"]["likeCount"] if "commentCount" in item["statistics"]: data["comment_count"] = item["statistics"]["commentCount"] return data st.title("YouTube 영상 요약기") st.caption("유튜브 영상 URL을 입력하시면 무료로 분석해드립니다.") api_key = "AIzaSyAJnXJEkXtWUDDFJGUw8AzGMWspyOe21uk" genai.configure(api_key=api_key) generation_config = { "temperature": 0.4, "top_p": 0.95, "top_k": 0, } safety_settings = [ { "category": "HARM_CATEGORY_HARASSMENT", "threshold": "BLOCK_NONE" }, { "category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "BLOCK_NONE" }, { "category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "BLOCK_NONE" }, { "category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "BLOCK_NONE" }, ] model_selection = st.sidebar.radio("**사용할 모델을 선택하세요 :**",("gemini-1.5-pro", "gemini-1.5-flash"), captions = ("가격↑/성능↑/속도↓", "가격↓/성능↓/속도↑")) if model_selection == "gemini-1.5-pro": model_name = "gemini-1.5-pro" else: model_name = "gemini-1.5-flash" url = st.text_input("YouTube 영상 URL을 입력해주세요:") youtube = build("youtube", "v3", developerKey=google_youtube_api_key) output_lang = st.selectbox("출력 언어를 선택하세요:", ("한국어", "영어", "일본어", "중국어")) if url: try: if "youtube.com/shorts/" in url: video_id = url.split("shorts/")[1] url = f"https://www.youtube.com/watch?v={video_id}" elif "youtube.com/live/" in url: video_id = url.split("live/")[1] url = f"https://www.youtube.com/watch?v={video_id}" yt = YouTube(url) video_id = yt.vid_info['videoDetails']['videoId'] st.markdown(f"## {yt.title}") st_player(url) with st.expander("영상 정보"): detail = get_video_details(youtube, video_id) st.text_area(label="**영상 설명**", value=detail['contentDetails']) st.markdown(f"- #### 조회수 : {yt.views:,} 회") hours, remainder = divmod(yt.length, 3600) minutes, seconds = divmod(remainder, 60) if hours > 0: st.markdown(f"- #### 영상 길이 : {hours}시간 {minutes}분 {seconds}초") else: st.markdown(f"- #### 영상 길이 : {minutes}분 {seconds}초") publish_date = yt.publish_date.strftime("%Y년 %m월 %d일") st.markdown(f"- #### 게시 날짜 : {publish_date}") st.markdown(f"- #### 채널명 : [{yt.author}]({yt.channel_url})") if 'like_count' in detail: st.markdown(f"- #### 좋아요 수 : {detail['like_count']} 개") else: st.markdown("- #### 좋아요 수가 제공되지 않습니다.") if 'comment_count' in detail: st.markdown(f"- #### 댓글 수 : {detail['comment_count']} 개") else: st.markdown("- #### 댓글 수 : 댓글 사용이 중지되어 가져올 수 없습니다.") if yt.length > 3599: attachment_type = st.radio("**영상 분석 방식을 선택하세요 :**", ("음성만",), index=0) st.warning("1시간 이상의 영상은 현재 음성 분석만 가능합니다.") else: attachment_type = st.radio("**첨부 파일 형식을 선택하세요 :**", ("영상 + 음성", "영상만", "영성만"), index=0) analysis_type = st.radio("**영상 분석 방식을 선택하세요 :**", ("3줄 요약", "심화 요약", "타임스탬프 생성"), index=0) analyze_video = st.button("영상 분석") if analyze_video: output_lang_dict = {"한국어": "한국어로", "영어": "영어로", "일본어": "일본어로", "중국어": "중국어로"} with st.spinner("Loading..."): with mock.patch('pytube.cipher.get_throttling_plan', patched_throttling_plan): videos = yt.streams.filter(adaptive=True, file_extension='mp4', res="360p") if len(videos) == 0: videos = yt.streams.filter(adaptive=True, file_extension='mp4') video = videos[0] else: video = videos[0] video.download(filename='video.mp4') audios = yt.streams.filter(only_audio=True, abr="128kbps", file_extension='mp4') if len(audios) == 0: audios = yt.streams.filter(only_audio=True, file_extension='mp4') audio = audios[0] else: audio = audios[0] audio.download(filename='audio.mp4') audio_clip = AudioFileClip("audio.mp4") audio_clip.write_audiofile("audio.m4a", codec='aac') video_file = genai.upload_file(path="video.mp4") audio_file = genai.upload_file(path="audio.m4a") while video_file.state.name == "PROCESSING" or audio_file.state.name == "PROCESSING": time.sleep(5) video_file = genai.get_file(video_file.name) audio_file = genai.get_file(audio_file.name) if video_file.state.name == "FAILED" or audio_file.state.name == "FAILED": raise ValueError("파일 처리 실패") with st.spinner("영상 분석 중..."): if analysis_type == "3줄 요약": system_instruction = f""" 업로드한 영상을 한국어로 3줄로 불렛포인트 형식으로 요약해줘. 어미는 ~임, ~함과 같이 간결하게 할 것. 각 문장의 앞에는 이모지를 하나 넣을 것. """ elif analysis_type == "심화 요약": system_instruction = f""" 업로드한 영상을 한국어로 마크다운 형식으로 구체적으로 요약해줘. 어미는 ~임, ~함과 같이 간결하게 할 것. 각 문장의 앞에는 이모지를 하나 넣을 것. """ else: system_instruction = f""" 업로드한 영상을 다음과 같은 형식으로 타임스탬프를 생성해줘 : Timestamp ** : {{subtitle}} ** : {{subtitle}} ** : {{subtitle}} ... """ model = genai.GenerativeModel(model_name=model_name, generation_config=generation_config, system_instruction=system_instruction, safety_settings=safety_settings) if attachment_type == "영상 + 음성": response = model.generate_content([video_file, audio_file], request_options={"timeout": 600}) elif attachment_type == "영상만": response = model.generate_content(video_file, request_options={"timeout": 600}) else: response = model.generate_content(audio_file, request_options={"timeout": 600}) if 'result_text' not in st.session_state: st.session_state.result_text = "" result_text = st.empty() result_text.success(response.text) st.session_state.result_text = response.text with st.expander("📋 마크다운 복사"): st.code(st.session_state.result_text, language='markdown') genai.delete_file(video_file.name) genai.delete_file(audio_file.name) os.remove("video.mp4") os.remove("audio.mp4") os.remove("audio.m4a") except Exception as e: st.error(f"에러 발생: {str(e)}")