.
================================================
FILE: README.md
================================================
# 半佛视频风格生成器
视频地址: [看我如何克隆半佛仙人，批量产出视频！](https://www.bilibili.com/video/BV12v411u7kw)
## 项目背景
半佛在2020年凭借众多沙雕表情包视频 + 魔性的文案迅速出圈。正如上一句所说，一个优秀的视频需要两部分组成，优秀的文案以及优秀的视频。那么半佛的成功是否可以复制呢？我认为在一定程度上是可以复制的。那么具体应该如何复制呢？我们尝试分析一下：
### 优秀的文案
半佛仙人的文案普遍比较nice，直接分析视频的文案似乎有点儿难度。好在半佛仙人还有微信公众号，直接学习微信公众号的写作风格是一个不错的选择，但是需要我们付出比较多的时间成本，并且学习效果不够直观。有没有什么更省事更直观效果更nice的方法呢？有。得益于GPT2模型强大的文本生成能力，我们也可以借助该模型训练一个半佛仙人风格文章生成器。
### 优秀的视频
直观上来说视频的生成似乎要简单一些，事实上确实如此。半佛仙人的视频大部分是由相关性不是很强的表情包堆砌起来，我们不可能有大量精力来寻找下载一个个导入视频制作软件并设置字幕。有没有什么自动化的方法？有。搞一个图片搜索接口，不妨我们自定义一个视频制作软件。
**理论存在，实践开始。**
## 思路：
我们再来详细复习一下思路，理清一下我们要做的每一件事情。
### 文案生成
直接生成大篇幅文案似乎不太现实，因为就目前来看，GPT2模型逻辑性不够强，上下文能力也有待提高，雪上加霜的是我们的电脑性能也不太行，只能采用GPT2蒸馏过后的模型，效果进一步下降，因此我们可以做一个文案提示器。当我们撰写文案的时候，轻松一按，即可生成相应的下文提示，这个思路是没有任何问题的。
1. 爬取半佛仙人微信公众号历史文章
2. 此处我们思考了一下框架的选择。不是很想在本地训练，目前体验以及部署最快的方案应该属于百度的AI Studio了。正好百度paddlenlp提供了中文GPT2预训练模型，因此我们就利用AI studio对GPT2Chinese预训练模型进行微调。
3. 数据集预处理。既然要微调自然需要预处理，由于只爬取到800+文章，为了扩大数据量，我们将滑动窗口设置的小一些。
4. 直接开整训练，数据不是很多的情况下，并不需要训练太久。百度每天8个小时免费GPU，训练几天也就ok了。
5. 写一个小小的窗口。
### 视频生成
1. 输入断句后的文案
2. 将文案根据短句分割，每句作为一条字幕
3. 根据字幕搜索表情包并选择设置
4. 利用语音合成手段合成配音
5. 重复3、4步骤，直到所有的字幕均完成表情包设定、字幕设定、配音设定
6. 合成视频并加入背景音乐
7. 导出成品
## 安装
1. 安装python3.8.3
2. 安装requirements.txt依赖
3. 运行`python main.py`
4. 运行`edit.py`请先按照百度paddlepaddle安装教程安装相应的库
## 使用
### 文案生成
1. 输入文件名或拖入文件
2. 输入文案
3. 按F1或者点击提示获取提示
4. 导出文案
5. 需要注意的是，导出文案需手动逐句分割
### 视频生成
1. 确定文件名，点击设置按钮可以新建工程。
2. 可以拖入txt格式的文案或者直接拖入软件内，或者直接增加文案。
3. 点击上一句以及下一句文案，预览区可预览字幕，搜索区可搜索相关图片/表情包。
4. 预览字幕时，若需要将一句话分为两页，则应用\n分开。
5. 所有字幕均设置完毕可进行生成视频。
6. 软件每5秒保存一次工程文件。崩溃时可重新拖放加载工程文件。工程文件位置：`material/{ProgectName}/*.bfs`
7. 工程所用到的素材默认会存放至`material/{ProgectName}/audio/`以及`material/{ProgectName}/img/`
8. 可将工程文件拖放入软件加载上一次的工程内容。需要注意的是加载工程要求素材文件夹必须存在！
9. 可将txt格式的文案拖入软件！
10. 可将表情包图片/gif文件拖入软件，用于设置当前字幕的表情包！
## TODO
- [x] 优化gif显示
- [x] 表情包尺寸的统一问题
- [x] 编辑保存防止异常退出工作丢失
- [x] 支持修改文案
- [x] 文件名自定义
- [x] 使用的图片与未使用的图片缓存分开，依据文件名归档
- [x] 使用过的配音缓存分类归档
- [x] 支持字幕不设置表情包
- [x] 修复只修改字幕最后成品无效果的bug
- [x] 修改单词拼写错误的小问题
- [x] 支持单条字幕分割成多页, 需要分割请在单条字幕分割的地方加上\n
- [ ] 自定义背景音乐
- [x] 跳转到n条文案
- [x] 生成视频的过程中主页面卡死问题
- [x] 生成视频显示进度条
- [ ] 下拉到底继续加载图片
- [x] 文案生成
## Reference
[一键生成半佛仙人视频，表情包之王你也可以！](https://www.bilibili.com/video/BV1oz411e7Jk)
[MoviePy](https://zulko.github.io/moviepy/)
[斗图啦](https://www.doutula.com/article/list/)
[百度图片](https://image.baidu.com/)
[百度在线语音合成](https://cloud.baidu.com/product/speech/tts_online)
[百度飞桨AI studio](https://aistudio.baidu.com/aistudio/index)
[百度飞桨NLP](https://paddlenlp.readthedocs.io/zh/latest/)
[deepai](https://deepai.org/)
## License
[GNU General Public License v3.0](LICENSE)
================================================
FILE: banfoTextSpider.py
================================================
import json
import os.path
import re
from collections.abc import Iterable
import requests
from config import BANFOMSGURL, BANFOMSGHEADERS
def getAllMsgUrl(start: int = 0) -> [Iterable, None]:
"""
获取半佛仙人公众号文章.
:param start: 起始索引
:return: None
"""
offset = start
# 请将下面两个参数换成自己的!抓包公众号历史消息即可.
while True:
url = BANFOMSGURL.format(offset)
headers = BANFOMSGHEADERS
ret = requests.get(url, headers=headers)
if ret.status_code != 200:
return None
data = json.loads(ret.text)
if 'general_msg_list' not in data:
return None
can_msg_continue = data['can_msg_continue']
data = json.loads(data['general_msg_list'])
data = data['list']
for msg in data:
if 'app_msg_ext_info' not in msg:
continue
msgUrl = msg['app_msg_ext_info']['content_url']
if len(msgUrl) >= 1:
yield msgUrl
if can_msg_continue == 1:
offset += 10
else:
return None
def getText(msgUrl: str) -> str:
"""
获取公众号单条文章的文本记录
:param msgUrl:
:return:
"""
data = requests.get(msgUrl)
if data.status_code != 200:
return ''
pattern = '([^(([^<]*?)|([^<]*?)
'
matchs = re.findall(pattern, data.text)
ret = ''
for match in matchs:
ret += match[0].strip() + match[1].strip() + match[2].strip()
return ret.replace('半佛仙人', '').replace('看一看入口已关闭', '')
if __name__ == '__main__':
for index, msgUrl in enumerate(getAllMsgUrl(start=0)):
if index < 557:
if index < 217 or index > 226:
continue
ret = getText(msgUrl)
if ret == '':
continue
print(index, msgUrl)
print(ret)
with open(os.path.join('banfoText', '{}.txt'.format(index)), 'w+', encoding='utf-8') as f:
f.write(ret)
print('done!')
================================================
FILE: conf.py
================================================
BackgroundMusic = 'backgroundMusic.mp3'
DoutulaButton = 0
BaiduButton = 1
MODELNAME = 'gpt-cpm-small-cn-distill'
================================================
FILE: edit.py
================================================
# -*- coding: utf-8 -*-
# Form implementation generated from reading ui file 'edit.ui'
#
# Created by: PyQt5 UI code generator 5.15.4
#
# WARNING: Any manual changes made to this file will be lost when pyuic5 is
# run again. Do not edit this file unless you know what you are doing.
import os
import sys
import time
from PyQt5 import QtCore, QtWidgets
from PyQt5.QtCore import QThread, pyqtSignal, QObject
from system_hotkey import SystemHotkey
from generation import getPredictText
from utils import getTips
class getTipsThread(QThread):
"""
获取提示线程类
"""
signal = pyqtSignal(str)
def __init__(self, mText: str) -> None:
"""
初始化获取提示线程类
:param mText: 关键词句
"""
super().__init__()
self.text = mText
def __del__(self):
self.wait()
def run(self):
"""
文本生成
:return: None
"""
'''
# 使用deepai的文本生成服务
text = getTips(self.text)
self.signal.emit(text)
'''
text = getPredictText(self.text.replace('\n', ''), length=500)
self.signal.emit(text)
class Ui_MainWindow(QObject):
hotkeySign = pyqtSignal()
def __init__(self):
super().__init__()
self.subThread = None
self.fileName = None
def setupUi(self, MainWindow):
MainWindow.resize(862, 579)
self.mainWindow = MainWindow
self.centralwidget = QtWidgets.QWidget(MainWindow)
# 文案输入&编辑框
self.writingEdit = QtWidgets.QPlainTextEdit(self.centralwidget)
self.writingEdit.setGeometry(QtCore.QRect(10, 10, 431, 501))
# 给文案提示加个框框
self.groupBox = QtWidgets.QGroupBox(self.centralwidget)
self.groupBox.setGeometry(QtCore.QRect(450, 10, 401, 501))
self.groupBox.setCheckable(False)
self.groupBox.setTitle('提示')
# 文案提示框框
self.tipsEdit = QtWidgets.QPlainTextEdit(self.groupBox)
self.tipsEdit.setGeometry(QtCore.QRect(10, 20, 381, 471))
# 文件名
self.FileNameLabel = QtWidgets.QLabel(self.centralwidget)
self.FileNameLabel.setGeometry(QtCore.QRect(10, 520, 101, 31))
self.FileNameLabel.setText('文件路径:')
self.fileNameEdit = QtWidgets.QLineEdit(self.centralwidget)
self.fileNameEdit.setGeometry(QtCore.QRect(90, 520, 351, 31))
# 三个按钮水平分布
self.horizontalLayoutWidget = QtWidgets.QWidget(self.centralwidget)
self.horizontalLayoutWidget.setGeometry(QtCore.QRect(450, 514, 401, 41))
self.horizontalLayout = QtWidgets.QHBoxLayout(self.horizontalLayoutWidget)
self.horizontalLayout.setContentsMargins(0, 0, 0, 0)
self.horizontalLayout.setObjectName('horizontalLayout')
# 打开
self.openButton = QtWidgets.QPushButton(self.horizontalLayoutWidget)
self.openButton.setText('打开')
self.openButton.clicked.connect(self.openFile)
self.horizontalLayout.addWidget(self.openButton)
# 保存
self.saveButton = QtWidgets.QPushButton(self.horizontalLayoutWidget)
self.saveButton.setText('保存')
self.saveButton.clicked.connect(self.saveFile)
self.horizontalLayout.addWidget(self.saveButton)
# 提示
self.tipsButton = QtWidgets.QPushButton(self.horizontalLayoutWidget)
self.tipsButton.setText('提示')
self.tipsButton.clicked.connect(self.tips)
self.horizontalLayout.addWidget(self.tipsButton)
# 热键
self.hotkeySign.connect(self.tips)
self.F1Hotkey = SystemHotkey()
self.F1Hotkey.register(['f1'], callback=lambda x: self.hotkeyEvent())
MainWindow.setCentralWidget(self.centralwidget)
QtCore.QMetaObject.connectSlotsByName(MainWindow)
def hotkeyEvent(self):
self.hotkeySign.emit()
def msgBox(self, msg: str, hasQuery: bool = False) -> bool:
"""
:param msg: 消息
:param hasQuery: 是否带询问
:return: 如果hasQuery为True,返回值为用户是否点击了确定
"""
if not hasQuery:
QtWidgets.QMessageBox.information(self.mainWindow, '提示', msg, QtWidgets.QMessageBox.Ok,
QtWidgets.QMessageBox.Ok)
return True
reply = QtWidgets.QMessageBox.question(self.mainWindow, '提示', msg,
QtWidgets.QMessageBox.Yes | QtWidgets.QMessageBox.Cancel,
QtWidgets.QMessageBox.Cancel)
return reply == QtWidgets.QMessageBox.Yes
def openFile(self) -> None:
"""
打开一个文件
:return: None
"""
fileName, _ = QtWidgets.QFileDialog.getOpenFileName(self.mainWindow, '选择文案', os.path.join(os.getcwd(), 'writing'), 'Text Files (*.txt)')
if not os.path.exists(fileName):
self.msgBox('未选择文件!')
return
with open(fileName, 'r+', encoding='UTF-8') as f:
data = f.read()
self.writingEdit.appendPlainText(data)
self.fileName = fileName
self.fileNameEdit.setText(fileName)
def saveFile(self) -> None:
"""
保存文件
:return: None
"""
fileName = self.fileNameEdit.text()
if os.path.sep not in fileName:
if fileName == '':
self.msgBox('请先设置文件名!')
return
if len(fileName) <= 4:
self.msgBox('文件名非法!')
return
fileName = os.path.join('writing', fileName)
if not os.path.exists('writing'):
os.mkdir('writing')
self.fileName = fileName
with open(fileName, 'w+', encoding='UTF-8') as f:
f.write(self.writingEdit.toPlainText())
self.msgBox('文件已保存在{}'.format(fileName))
def tips(self) -> None:
text = self.writingEdit.toPlainText()
if text == '':
return
self.tipsEdit.setPlainText('')
if self.subThread is not None:
self.subThread.terminate()
while self.subThread.isRunning() and not self.subThread.isFinished():
time.sleep(0.1)
self.subThread = getTipsThread(self.writingEdit.toPlainText())
self.subThread.signal.connect(self.setTips)
self.subThread.start()
def setTips(self, text: str) -> None:
"""
设置提示
:param text: 提示文本
:return: None
"""
self.tipsEdit.setPlainText(text)
if __name__ == "__main__":
app = QtWidgets.QApplication(sys.argv)
MainWindow = QtWidgets.QMainWindow()
ui = Ui_MainWindow()
ui.setupUi(MainWindow)
MainWindow.show()
sys.exit(app.exec_())
================================================
FILE: generation.py
================================================
import paddle
import paddlenlp
from conf import MODELNAME
paddle.set_device('gpu')
gptModel = paddlenlp.transformers.GPTModel.from_pretrained('models')
gptModel = paddlenlp.transformers.GPTForPretraining(gptModel)
gptModel.eval()
tokenizer = paddlenlp.transformers.GPTChineseTokenizer.from_pretrained(MODELNAME)
def getPredictText(text: str, length: int = 200) -> str:
"""
生成半佛风格文本
:param text: 前面部分的文本
:param length: 生成文本长度
:return: 生成的文本
"""
encodedText = tokenizer(text=text, return_token_type_ids=False)
inputIds = paddle.to_tensor(encodedText['input_ids'], dtype='int64').unsqueeze(0)
ids, _ = gptModel.generate(input_ids=inputIds, max_length=length, min_length=32, decode_strategy='sampling')
ids = ids[0].numpy().tolist()
# 使用tokenizer将生成的id转为文本
generatedText = tokenizer.convert_ids_to_string(ids)
return generatedText
getPredictText('开始预测模型会先初始化一下, 抵消掉这个时间')
================================================
FILE: main.py
================================================
import logging
import math
import os.path
import pickle
import random
import shutil
import threading
import time
from PyQt5 import QtWidgets
from PyQt5.QtCore import QThread, pyqtSignal
from PyQt5.QtGui import QDropEvent, QDragEnterEvent, QStandardItem
import mainWindow
from PyQt5.QtWidgets import QApplication, QDialog
from moviepy.editor import *
from conf import BackgroundMusic, BaiduButton, DoutulaButton
from utils import getUuid, getBaiDuAudio, getBaiduImgPath, getDoutulaImgPath, resizeImg, convertToRGB
class genVideoThread(QThread):
"""
生成视频线程类
"""
signal = pyqtSignal(str)
def __init__(self, sections: list, materialName: str, fileName: str) -> None:
"""
初始化几个参数
:param sections: 字幕以及图片信息
:param materialName: 素材路径
:param fileName: 保存视频名称
"""
super().__init__()
self.sections = sections
self.materialName = materialName
self.fileName = fileName
def __del__(self):
self.wait()
def run(self):
screensize = (800, 600)
videoClips = []
for i, section in enumerate(self.sections):
imgPath, text = section[0], section[1]
if len(text) >= 1:
mark = text[-1]
if mark == '$':
text = text[:-1]
text = text.split(r'\\')
text = '\\'.join([x.replace(r'\n', '\n') for x in text])
text = text.strip().split('\n')
if imgPath is None:
imgPath = 'background.png'
# gif用到,用于标记当前字幕对应的gif从哪儿开始
index = 0
print(text, imgPath)
if imgPath.endswith('.gif'):
# 首先计算一下当前所有语音时间长度
clip = VideoFileClip(imgPath)
clip = clip.loop()
else:
convertToRGB(imgPath)
clip = ImageClip(imgPath)
# 设置一下图片/gif大小
if mark != '$':
width, height = clip.size
width, height = resizeImg(width, height)
clip = clip.resize((width, height))
# 考虑到每张表情包可能对应多句字幕
for txt in text:
# 合成语音
txtAudio = getBaiDuAudio(txt, os.path.join(self.materialName, 'audio'))
if len(txt) < 12:
fontsize = 50
else:
fontsize = 40
txtClip = TextClip(txt, color='white', font='STKaiti', kerning=5, fontsize=fontsize, align='South')
if txtAudio is None:
logging.error('get the audio of {} failed!'.format(txt))
continue
txtAudio = AudioFileClip(txtAudio)
# 表情包视频与字幕融合
cvc = CompositeVideoClip([clip.set_position(('center', 'center')).subclip(index, txtAudio.duration),
txtClip.set_position(('center', 0.85), relative=True)],
size=screensize)
cvc = cvc.subclip(0, txtAudio.duration)
index += txtAudio.duration
# 添加配音
cvc = cvc.set_audio(txtAudio)
videoClips.append(cvc)
self.signal.emit('进度: {}%'.format(math.ceil((i * 80) / len(self.sections))))
finalClip = concatenate_videoclips(videoClips)
# 获取原视频声音
audio = finalClip.audio
# 整体背景音乐
audioClip = AudioFileClip(BackgroundMusic)
if audioClip.duration > finalClip.duration:
audioClip = audioClip.subclip(0, audio.duration)
elif audioClip.duration < finalClip.duration:
audioClip = afx.audio_loop(audioClip, duration=audio.duration)
audioClip = afx.volumex(audioClip, factor=0.35)
self.signal.emit('进度: {}%'.format(math.ceil(i * 80 / len(self.sections)) + random.randint(5, 15)))
# 声音结合起来
audio = CompositeAudioClip([audio, audioClip])
finalClip = finalClip.set_audio(audio)
fileName = os.path.join('out', self.fileName)
self.signal.emit('进度: {}'.format('处理完成! 正在写出文件...'))
finalClip.write_videofile(fileName, fps=25, codec='mpeg4')
self.signal.emit(fileName)
class addImgThread(QThread):
"""
获取网络表情包线程类
"""
signal = pyqtSignal(str)
def __init__(self, mText: str, mButton: int) -> None:
"""
初始化获取网络表情包线程类
:param mText: 搜索关键词
:param mButton: 采用哪个搜索引擎
"""
super().__init__()
self.text = mText
self.button = mButton
def __del__(self):
self.wait()
def run(self):
if self.button == BaiduButton:
for path in getBaiduImgPath(self.text):
self.signal.emit(path)
if self.button == DoutulaButton:
for path in getDoutulaImgPath(self.text):
self.signal.emit(path)
class MainDialog(QDialog):
def __init__(self, parent=None):
super(QDialog, self).__init__(parent)
# 当前指向的句子
self.nowPos = None
# 获取表情包线程句柄
self.subThread = None
# 已经设定好的句子以及表情包,三元组(图片路径, 文案, 时间戳),时间戳一旦生成不再修改,主要用于区分文案并生成对应的图片路径
self.sections = []
# 接收拖放对象
self.setAcceptDrops(True)
self.fileName = None
self.materialName = None
# lock of save and read the bfs file
self.lock = threading.Lock()
self.ui = mainWindow.Ui_MainWindow()
self.ui.setupUi(self)
# start save thread
self.saveThread = threading.Thread(target=self.save)
self.saveThread.start()
def save(self) -> None:
"""
每隔5秒保存一次工程信息
:return: None
"""
time.sleep(5)
# 窗口存在则一直保存
while self.ui.windowIsVisible():
self.lock.acquire()
if len(self.sections) <= 0:
self.lock.release()
time.sleep(5)
continue
data = dict()
data['nowPos'] = self.nowPos
data['sections'] = self.sections
data['fileName'] = self.fileName
data['materialName'] = self.materialName
fileName = os.path.join(self.materialName, self.fileName[:self.fileName.rfind('.')] + '.bfs')
with open(fileName, 'wb') as f:
pickle.dump(data, f)
self.lock.release()
time.sleep(5)
def dragEnterEvent(self, event: QDragEnterEvent) -> None:
"""
拖放事件
:param event: QDragEnterEvent
:return: None
"""
if event.mimeData().hasText():
event.accept()
else:
event.ignore()
def dropEvent(self, event: QDropEvent) -> None:
"""
拖放事件,主要处理.bfs工程文件, .txt文案文件, .gif/.png等图片文件
:param event: 拖放对象事件
:return: None
"""
filePathList = event.mimeData().text()
filePath = filePathList.split('\n')[0].replace('file:///', '', 1)
# 说明是加载的工程文件
if filePath.endswith('.bfs'):
self.loadBfs(filePath)
return
# 说明加载的是文案文件
if filePath.endswith('.txt'):
if self.fileName is None:
self.ui.msgBox('请先设置工程目录!')
return
self.loadText(True, filePath)
return
# 其他情况应该是加载的图片文件,先判断是不是gif/图片
if not filePath.endswith('.gif'):
# 通过加载文件来判断是否为图片,不是则返回
try:
ImageClip(filePath)
except:
return
self.loadPic(filePath)
def loadBfs(self, filePath: str) -> None:
"""
加载工程文件
:param filePath: 工程文件路径
:return: None
"""
if len(self.sections) > 0:
if not self.ui.msgBox('导入工程文件将清空当前工作内容,可能导致部分内容丢失,是否继续?', True):
return
# 需要在material目录下有对应的文件夹,没有的禁止载入
fileName = os.path.basename(filePath).replace('.bfs', '')
if not os.path.exists(os.path.join('material', fileName)):
self.ui.msgBox('未找到对应的素材文件夹!')
return
with open(filePath, 'rb') as f:
data = pickle.load(f)
self.lock.acquire()
self.nowPos = data['nowPos']
self.sections = data['sections']
# 检查资源文件是否都存在
for section in self.sections:
if section[0] is not None and not os.path.exists(section[0]):
self.nowPos = None
self.sections = []
self.ui.msgBox('对应的素材缺失!')
self.lock.release()
return
self.fileName = data['fileName']
self.ui.setFileName(self.fileName)
self.materialName = data['materialName']
# 将sections的内容填充到表格
self.ui.delAllRow()
for section in self.sections:
self.ui.addRow(section[1])
if self.nowPos is not None:
self.ui.setSubTileText(self.sections[self.nowPos][1])
self.ui.setSearchText(self.sections[self.nowPos][1])
imgPath = self.sections[self.nowPos][0]
if imgPath is None:
self.ui.delVideoImg()
else:
self.ui.changeVideoImg(imgPath)
else:
self.ui.delVideoImg()
self.lock.release()
def loadPic(self, imgPath: str) -> None:
"""
加载拖放进来的图片文件
:param imgPath: 图片文件路径
:return: None
"""
if self.nowPos is None:
# 没有工程内容则忽略该次拖入文件
self.ui.msgBox('请先设置字幕内容!')
return
self.ui.changeVideoImg(imgPath)
imgBaseName = os.path.basename(imgPath)
suffix = imgBaseName[imgBaseName.rfind('.') + 1:]
uuid = self.sections[self.nowPos][2]
newPath = '{}.{}'.format(uuid, suffix)
newPath = os.path.join(os.path.join(self.materialName, 'img'), newPath)
shutil.copyfile(imgPath, newPath)
# 保存图片与字幕信息
self.sections[self.nowPos] = [newPath, self.ui.getSubtitle(), uuid]
self.ui.setRowText(self.nowPos, self.ui.getSubtitle())
def setFilename(self) -> None:
"""
设置文件名
:return: None
"""
# 加锁,禁止保存或者载入工程文件
self.lock.acquire()
fileName = self.ui.getFileName()
materialName = os.path.join('material', fileName[:fileName.rfind('.')])
# 当前工作区没内容,说明是新建的工程,新建的工程的名字不能和之前重复
if len(self.sections) <= 0 and os.path.exists(materialName):
self.ui.msgBox('文件名已存在,请更换名字或加载之前缓存!')
self.lock.release()
return
self.fileName = fileName
# 文件名并未更改,可能是未修改或者只修改了后缀名,都可以忽略
if self.materialName == materialName:
self.ui.msgBox('设置工程文件夹成功!')
self.lock.release()
return
# 当前工作区无内容,说明是新建的工程,需要新建文件夹
if len(self.sections) <= 0:
os.makedirs(os.path.join(materialName, 'audio'))
os.makedirs(os.path.join(materialName, 'img'))
self.ui.msgBox('新建工程文件夹成功!')
else:
# 当前工作区有内容,说明工程已经存在,需要对所有数据进行重命名
self.changeFileName(self.materialName, materialName)
self.ui.msgBox('重命名工程文件夹成功!')
self.materialName = materialName
self.lock.release()
def changeFileName(self, oldMaterialName: str, materialName: str) -> None:
"""
修改了文件名需要对文件夹等全部进行修改
:param oldMaterialName: 旧的素材文件夹名
:param materialName: 新的素材文件夹名
:return:
"""
# 对sections里面包含的图片信息的地址进行修改
for section in self.sections:
if section[0] is None:
continue
imgBaseName = os.path.basename(section[0])
newPath = os.path.join(os.path.join(materialName, 'img'), imgBaseName)
section[0] = newPath
# 更改文件夹名字
os.rename(oldMaterialName, materialName)
def setSubtitleInfo(self) -> None:
"""
设置上一句/下一句对应的字幕以及图片信息
:return: None
"""
self.ui.setSubTileText(self.sections[self.nowPos][1])
self.ui.setSearchText(self.sections[self.nowPos][1])
# 如果已经设置好了表情包,显示出来;否则就清空
imgPath = self.sections[self.nowPos][0]
if imgPath is None:
self.ui.delVideoImg()
else:
self.ui.changeVideoImg(imgPath)
def last(self) -> None:
"""
上一句,加载上一句字幕与图片
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('当前工作区暂无内容!')
return
# 计算当前应该到达的光标
if self.nowPos is None:
self.nowPos = 1
else:
# 保存下一句的字幕信息
self.sections[self.nowPos][1] = self.ui.getSubtitle()
self.ui.setRowText(self.nowPos, self.ui.getSubtitle())
self.nowPos = self.nowPos - 1 if self.nowPos > 0 else len(self.sections) - 1
self.setSubtitleInfo()
def next(self) -> None:
"""
下一句,加载下一句字幕与图片
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('当前工作区暂无内容!')
return
if self.nowPos is None:
self.nowPos = -1
else:
# 保存上一句的字幕信息
self.sections[self.nowPos][1] = self.ui.getSubtitle()
self.ui.setRowText(self.nowPos, self.ui.getSubtitle())
self.nowPos = self.nowPos + 1 if self.nowPos < len(self.sections) - 1 else 0
self.setSubtitleInfo()
def changeThePicText(self, text: str) -> None:
"""
视频字幕实时更改
:param text: 字幕信息
:return: None
"""
if len(self.sections) <= 0:
return
self.ui.setVideoText(text)
self.sections[self.nowPos][1] = text
self.ui.setRowText(self.nowPos, text)
def previewImg(self, path: str) -> None:
"""
将网络表情包加载预览以供选择
:param path: 网络表情包路径
:return: None
"""
self.ui.addImg(path)
def search(self, button: int) -> None:
"""
搜索表情包,button代表了不同的搜索引擎
:param button: 来自哪个按钮,代表了不同的搜索引擎
:return: None
"""
if len(self.sections) <= 0 or self.ui.getSearchText() == '':
self.ui.msgBox('工作区暂无内容或未输入搜索文字!')
return
if self.subThread is not None:
self.subThread.terminate()
while self.subThread.isRunning() and not self.subThread.isFinished():
time.sleep(0.1)
# 清空当前的所有表情包图片
self.ui.delImg()
self.subThread = addImgThread(self.ui.getSearchText(), button)
self.subThread.signal.connect(self.previewImg)
self.subThread.start()
def imgClicked(self, index: int) -> None:
"""
表情包点击回调函数,将选好的表情包加载到视频预览区
:param index: 选好的表情包索引
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('工作区无内容!')
return
if self.nowPos is None:
self.ui.msgBox('暂未选择文案与字幕!')
return
imgPath = self.ui.getImgPathByIndex(index)
self.ui.changeVideoImg(imgPath)
# 将表情包复制到material目录
imgBaseName = os.path.basename(imgPath)
suffix = imgBaseName[imgBaseName.rfind('.') + 1:]
uuid = self.sections[self.nowPos][2]
newPath = '{}.{}'.format(uuid, suffix)
newPath = os.path.join(os.path.join(self.materialName, 'img'), newPath)
shutil.copyfile(imgPath, newPath)
# 保存图片与字幕信息
self.sections[self.nowPos] = [newPath, self.ui.getSubtitle(), uuid]
self.ui.setRowText(self.nowPos, self.ui.getSubtitle())
def genVideo(self) -> None:
"""
开始生成视频
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('工作区无内容!')
return
self.setDisabled(True)
self.gvt = genVideoThread(self.sections, self.materialName, self.fileName)
self.gvt.signal.connect(self.genVideoFinished)
self.gvt.start()
def genVideoFinished(self, msg: str) -> None:
"""
视频生成完成回调函数
:param msg: 生成视频传回的信息。主要有两种，第一是进度，第二是输出的视频的路径
:return: None
"""
if ': ' in msg:
self.ui.setGenVideoText(msg)
return
self.setDisabled(False)
self.ui.setGenVideoText('生成视频')
self.ui.msgBox('生成完毕!位置:{}'.format(msg))
def loadText(self, drag=False, fileName: str=None) -> None:
"""
文件浏览器回调函数/同时支持拖放导入文案解析
:param drag: 是否是拖放导入的
:param fileName: 拖放进来的文件名
:return: None
"""
if self.fileName is None:
self.ui.msgBox('请先设置工程目录!')
return
if len(self.sections) > 0 and not self.ui.msgBox('当前导入会覆盖工作区内容,不可撤销!是否继续?', True):
return
self.ui.delAllRow()
if not drag:
fileName, _ = QtWidgets.QFileDialog.getOpenFileName(self, '选择文案', os.getcwd(), 'Text Files (*.txt)')
if not os.path.exists(fileName):
self.ui.msgBox('未选择文件!')
return
with open(fileName, 'r', encoding='utf-8') as f:
data = f.read()
self.sections = list()
for text in data.split('\n'):
if text == '':
continue
text = text.strip()
self.ui.addRow(text)
self.sections.append([None, text, getUuid()])
self.ui.msgBox('导入完成!')
def addFrontText(self) -> None:
"""
在当前选中的表格单元前面增加一行空白行
:return: None
"""
if len(self.sections) <= 0:
if self.fileName is not None:
index = 0
else:
self.ui.msgBox('请设置文件名后再添加!')
return
else:
index = self.ui.getCurrentSelected()
if index == -1:
self.ui.msgBox('未选中表格!')
return
if self.nowPos is not None and self.nowPos >= index:
self.nowPos += 1
self.ui.insertRow(index)
self.sections.insert(index, [None, '', getUuid()])
def addBehindText(self) -> None:
"""
在当前选中的表格单元后面增加一行空白行
:return: None
"""
if len(self.sections) <= 0:
if self.fileName is not None:
index = -1
else:
self.ui.msgBox('请设置文件名后再添加!')
return
else:
index = self.ui.getCurrentSelected()
if index == -1:
self.ui.msgBox('未选中表格!')
return
if self.nowPos is not None and self.nowPos > index:
self.nowPos += 1
self.ui.insertRow(index + 1)
self.sections.insert(index + 1, [None, '', getUuid()])
def delText(self) -> None:
"""
删除该行
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('请先创建工程或输入文案!')
return
index = self.ui.getCurrentSelected()
if index == -1:
self.ui.msgBox('未选中表格!')
return
# 如果删除了当前预览位置的文案
if self.nowPos is not None and self.nowPos == index:
# 加载下一句文案
self.next()
# 如果在当前位置之前删除了一个文案,那么当前位置-1
if self.nowPos is not None and self.nowPos >= index:
self.nowPos -= 1
self.ui.delRow(index)
del self.sections[index]
def exportText(self) -> None:
"""
导出文案内容
:return: None
"""
if len(self.sections) <= 0:
self.ui.msgBox('当前工作区没有内容!')
return
texts = ''
for section in self.sections:
texts += section[1] + '\n'
path = os.path.join(self.materialName, 'work.txt')
with open(path, 'w+') as f:
f.write(texts)
self.ui.msgBox('导出文案成功!位置:{}'.format(path))
def tableItemChange(self, item: QStandardItem) -> None:
"""
文案内容被修改时,同步到sections
:param item: 被修改的表格item
:return:
"""
if len(self.sections) > item.row():
self.sections[item.row()][1] = item.text()
# 如果当前单句字幕正好是修改的单元格部分,则修改的内容同步修改到单句字幕编辑处. 需要排除由于单句字幕修改造成的表格修改的情况
if self.nowPos is not None and self.nowPos == item.row() and self.ui.getSubtitle() != item.text():
self.ui.setSubTileText(item.text())
self.ui.setSearchText(item.text())
def jumpToIndex(self) -> None:
"""
跳转到指定的文案
:return: None
"""
if self.nowPos is not None and self.ui.getCurrentSelected() != -1:
# 保存上一句的字幕信息
self.sections[self.nowPos][1] = self.ui.getSubtitle()
self.ui.setRowText(self.nowPos, self.ui.getSubtitle())
self.nowPos = self.ui.getCurrentSelected()
self.setSubtitleInfo()
if __name__ == '__main__':
if os.path.exists('tmp'):
shutil.rmtree('tmp')
os.mkdir('tmp')
if not os.path.exists('out'):
os.mkdir('out')
myapp = QApplication(sys.argv)
myDlg = MainDialog()
myDlg.show()
sys.exit(myapp.exec_())
================================================
FILE: mainWindow.py
================================================
# -*- coding: utf-8 -*-
# Form implementation generated from reading ui file 'mainWindow.ui'
#
# Created by: PyQt5 UI code generator 5.15.4
#
# WARNING: Any manual changes made to this file will be lost when pyuic5 is
# run again. Do not edit this file unless you know what you are doing.
import math
import os.path
import shutil
import time
from PyQt5 import QtCore, QtGui, QtWidgets
from PyQt5.QtGui import QMovie, QStandardItemModel, QStandardItem
from moviepy.video.VideoClip import ImageClip
from moviepy.video.io.VideoFileClip import VideoFileClip
from conf import DoutulaButton, BaiduButton
from utils import resizeImg
class clickedButton(QtWidgets.QPushButton):
"""
搜索表情包按钮事件,主要可以区分哪个按钮
"""
clicked = QtCore.pyqtSignal(int)
def __init__(self, button, parent=None):
super(clickedButton, self).__init__(parent)
self.button = button
def mouseReleaseEvent(self, QMouseEvent):
self.clicked.emit(self.button)
class clickedLabel(QtWidgets.QLabel):
"""
标签类,主要增加了点击事件,用于图片缓存区,用户点击了图片后可知道点击了哪张图
"""
clicked = QtCore.pyqtSignal(int)
def __init__(self, index: int, parent=None):
"""
:param index: 图片索引
:param parent:
"""
super(clickedLabel, self).__init__(parent)
self.index = index
def mouseReleaseEvent(self, QMouseEvent):
self.clicked.emit(self.index)
class Ui_MainWindow(object):
def __init__(self):
# 所有缓存的图片
self.img = []
# 当前视频预览区gif
self.gif = None
# 当前视频预览区图片
self.videoImg = None
def setupUi(self, MainWindow):
# 初始设置窗口信息
MainWindow.setObjectName('MainWindow')
MainWindow.resize(1890, 702)
MainWindow.setLayoutDirection(QtCore.Qt.LeftToRight)
MainWindow.setWindowTitle('半佛风格视频生成')
self.mainWindow = MainWindow
self.centralwidget = QtWidgets.QWidget(MainWindow)
# 输入文件名标签
self.fileLabel = QtWidgets.QLabel(self.centralwidget)
self.fileLabel.setGeometry(QtCore.QRect(10, 23, 121, 16))
self.fileLabel.setText('导出视频名称: ')
# 文件名
self.filenName = QtWidgets.QLineEdit(self.centralwidget)
self.filenName.setGeometry(QtCore.QRect(140, 16, 361, 31))
self.filenName.setText('{}.mp4'.format(int(time.time())))
# 输入文件名确定按钮
self.filenameButton = QtWidgets.QPushButton(self.centralwidget)
self.filenameButton.setGeometry(QtCore.QRect(501, 16, 60, 31))
self.filenameButton.setText('设置')
self.filenameButton.clicked.connect(MainWindow.setFilename)
# 给输入框加个分组box
self.groupBox = QtWidgets.QGroupBox(self.centralwidget)
self.groupBox.setGeometry(QtCore.QRect(0, 53, 561, 618))
self.groupBox.setTitle('在此输入/编辑文案')
# 文案以表格的形式展示
self.model = QStandardItemModel(0, 0)
self.model.itemChanged.connect(MainWindow.tableItemChange)
# 设置水平方向四个头标签文本内容
self.model.setHorizontalHeaderLabels(['文案&字幕'])
self.row = 0
self.tableView = QtWidgets.QTableView(self.groupBox)
self.tableView.setGeometry(QtCore.QRect(10, 20, 541, 542))
self.tableView.setShowGrid(True)
self.tableView.setModel(self.model)
self.tableView.horizontalHeader().setStretchLastSection(True)
self.tableView.horizontalHeader().setSectionResizeMode(QtWidgets.QHeaderView.Stretch)
self.tableView.clicked.connect(MainWindow.jumpToIndex)
# 打开文案/前增一句/后增一句/删除一句/修改一句/导出文案按钮布局
self.tableButtonWidget = QtWidgets.QWidget(self.groupBox)
self.tableButtonWidget.setGeometry(QtCore.QRect(10, 561, 541, 55))
# 水平分布
self.hbox = QtWidgets.QHBoxLayout()
self.hbox.setGeometry(QtCore.QRect())
self.hbox.setContentsMargins(0, 0, 0, 0)
# 打开文案
self.openText = QtWidgets.QPushButton()
self.openText.setText('打开')
self.openText.clicked.connect(MainWindow.loadText)
self.hbox.addWidget(self.openText)
# 在前面增加一句
self.addFrontText = QtWidgets.QPushButton()
self.addFrontText.setText('前增一句')
self.addFrontText.clicked.connect(MainWindow.addFrontText)
self.hbox.addWidget(self.addFrontText)
# 在后面增加一句
self.addBehindText = QtWidgets.QPushButton()
self.addBehindText.setText('后增一句')
self.addBehindText.clicked.connect(MainWindow.addBehindText)
self.hbox.addWidget(self.addBehindText)
# 删除该句
self.delText = QtWidgets.QPushButton()
self.delText.setText('删除该句')
self.delText.clicked.connect(MainWindow.delText)
self.hbox.addWidget(self.delText)
# 导出文案
self.exportText = QtWidgets.QPushButton()
self.exportText.setText('导出文案')
self.exportText.clicked.connect(MainWindow.exportText)
self.hbox.addWidget(self.exportText)
self.tableButtonWidget.setLayout(self.hbox)
# 单句字幕
self.singleText = QtWidgets.QLineEdit(self.centralwidget)
self.singleText.setGeometry(QtCore.QRect(660, 620, 611, 50))
self.singleText.textChanged.connect(MainWindow.changeThePicText)
# 搜索框文本
self.searchText = QtWidgets.QLineEdit(self.centralwidget)
self.searchText.setGeometry(QtCore.QRect(1360, 17, 421, 41))
# 斗图啦/百度搜索按钮
self.searchDou = clickedButton(DoutulaButton, self.centralwidget)
self.searchDou.setGeometry(QtCore.QRect(1785, 17, 45, 41))
self.searchDou.setText('斗图')
self.searchDou.clicked.connect(MainWindow.search)
self.searchBai = clickedButton(BaiduButton, self.centralwidget)
self.searchBai.setGeometry(QtCore.QRect(1834, 17, 45, 41))
self.searchBai.setText('百度')
self.searchBai.clicked.connect(MainWindow.search)
# 视频背景图
self.videoBackgroud = QtWidgets.QLabel(self.centralwidget)
self.videoBackgroud.setGeometry(QtCore.QRect(560, 16, 800, 600))
self.videoBackgroud.setPixmap(QtGui.QPixmap('background.png'))
self.videoBackgroud.setScaledContents(True)
# 视频字幕
self.addSubtitleLayout()
# 上一句按钮
self.last = QtWidgets.QPushButton(self.centralwidget)
self.last.setGeometry(QtCore.QRect(560, 620, 101, 51))
self.last.setText('上一句')
self.last.clicked.connect(MainWindow.last)
# 下一句按钮
self.next = QtWidgets.QPushButton(self.centralwidget)
self.next.setGeometry(QtCore.QRect(1270, 620, 91, 51))
self.next.setText('下一句')
self.next.clicked.connect(MainWindow.next)
# 生成视频按钮
self.genVideo = QtWidgets.QPushButton(self.centralwidget)
self.genVideo.setGeometry(QtCore.QRect(1360, 620, 521, 51))
self.genVideo.setText('生成视频')
self.genVideo.clicked.connect(MainWindow.genVideo)
# 表情包图片位置布局
self.widget = QtWidgets.QWidget(self.centralwidget)
self.widget.setGeometry(QtCore.QRect(1349, 48, 540, 579))
self.topFiller = QtWidgets.QWidget()
self.scroll = QtWidgets.QScrollArea()
self.scroll.setWidget(self.topFiller)
self.vbox = QtWidgets.QVBoxLayout()
self.vbox.addWidget(self.scroll)
self.widget.setLayout(self.vbox)
QtCore.QMetaObject.connectSlotsByName(MainWindow)
def addSubtitleLayout(self) -> None:
"""
视频字幕设置布局
:return:
"""
# 视频字幕
self.videoText = QtWidgets.QLabel(self.centralwidget)
self.videoText.setGeometry(QtCore.QRect(560, 526, 801, 51))
# 视频字幕字体颜色
palette = QtGui.QPalette()
brush = QtGui.QBrush(QtGui.QColor(255, 255, 255))
brush.setStyle(QtCore.Qt.SolidPattern)
palette.setBrush(QtGui.QPalette.Active, QtGui.QPalette.WindowText, brush)
brush = QtGui.QBrush(QtGui.QColor(255, 255, 255))
brush.setStyle(QtCore.Qt.SolidPattern)
palette.setBrush(QtGui.QPalette.Inactive, QtGui.QPalette.WindowText, brush)
brush = QtGui.QBrush(QtGui.QColor(120, 120, 120))
brush.setStyle(QtCore.Qt.SolidPattern)
palette.setBrush(QtGui.QPalette.Disabled, QtGui.QPalette.WindowText, brush)
self.videoText.setPalette(palette)
# 视频字幕字体设置
font = QtGui.QFont()
font.setFamily('华文楷体')
font.setPointSize(20)
self.videoText.setFont(font)
self.videoText.setTextFormat(QtCore.Qt.PlainText)
self.videoText.setScaledContents(False)
self.videoText.setAlignment(QtCore.Qt.AlignCenter)
self.videoText.setWordWrap(False)
def delVideoImg(self) -> None:
"""
清空视频预览区图片
:return: None
"""
if self.videoImg is not None:
self.videoImg.deleteLater()
self.videoImg = None
if self.gif is not None:
self.gif.deleteLater()
self.gif = None
def getResizedOfVideoImg(self, path) -> (int, int):
"""
获取路径图片重新调整过的大小
:param path: 图片路径
:return: (width, height)
"""
try:
clip = VideoFileClip(path)
except:
clip = ImageClip(path)
width, height = clip.size
return resizeImg(width, height)
def changeVideoImg(self, path: str) -> None:
"""
加载图片到视频预览区
:param path: 图片路径
:return: None
"""
# 如果图片不是在tmp目录保存,则转到tmp目录再加载,不然重命名工程会有问题
if not path.startswith('tmp' + os.path.sep):
newPath = os.path.join('tmp', os.path.basename(path))
if not os.path.exists(newPath):
shutil.copyfile(path, newPath)
path = newPath
self.delVideoImg()
self.videoImg = QtWidgets.QLabel(self.centralwidget)
wight, height = self.getResizedOfVideoImg(path)
self.videoImg.setGeometry(QtCore.QRect(int(960 - wight / 2), int(316 - height / 2), wight, height))
self.videoImg.setScaledContents(True)
self.gif = QMovie(path)
self.videoImg.setMovie(self.gif)
self.videoImg.setAlignment(QtCore.Qt.AlignCenter)
self.gif.start()
self.videoImg.show()
self.centralwidget.show()
def addImg(self, path: str) -> None:
"""
给图片缓存区加一张图片
:param path: 图片路径
:return: None
"""
# 计算图片框位置
row = math.ceil((len(self.img) + 1) / 3) - 1
col = len(self.img) % 3
img_label = clickedLabel(len(self.img), self.topFiller)
img_label.setGeometry(QtCore.QRect(1370, 60, 151, 151))
gif = QMovie(path)
img_label.setMovie(gif)
img_label.setScaledContents(True)
gif.start()
img_label.move(col * (151 + 10) + 10, row * (151 + 10) + 10)
self.img.append((img_label, gif, path))
self.topFiller.setMinimumSize(490, (row + 1) * (151 + 10))
self.scroll.setWidget(self.topFiller)
img_label.clicked.connect(self.mainWindow.imgClicked)
img_label.show()
self.topFiller.show()
self.widget.show()
def delImg(self) -> None:
"""
清空从网络获取的所有的表情包
:return: None
"""
for imgLabel, gif, _ in self.img:
imgLabel.deleteLater()
gif.deleteLater()
self.img = []
self.widget.show()
def getImgPathByIndex(self, index: int) -> str:
"""
通过表情包缓存区的索引获取表情包的实际路径
:param index: 表情包缓存区的索引
:return: 表情包的实际位置
"""
return self.img[index][2]
def addRow(self, text: str) -> None:
"""
向表中添加一行新的数据
:param text:
:return:
"""
item = QStandardItem(text)
self.model.setItem(self.row, 0, item)
self.row += 1
def delRow(self, row: int) -> None:
"""
删除一行表中的数据
:param row: 删除索引
:return:
"""
self.model.removeRow(row)
def delAllRow(self) -> None:
"""
删除所有行
:return:
"""
for index in range(self.model.rowCount())[::-1]:
self.delRow(index)
def msgBox(self, msg: str, hasQuery: bool = False) -> bool:
"""
:param msg: 消息
:param hasQuery: 是否带询问
:return: 如果hasQuery为True,返回值为用户是否点击了确定
"""
if not hasQuery:
QtWidgets.QMessageBox.information(self.mainWindow, '提示', msg, QtWidgets.QMessageBox.Ok,
QtWidgets.QMessageBox.Ok)
return True
reply = QtWidgets.QMessageBox.question(self.mainWindow, '提示', msg,
QtWidgets.QMessageBox.Yes | QtWidgets.QMessageBox.Cancel,
QtWidgets.QMessageBox.Cancel)
return reply == QtWidgets.QMessageBox.Yes
def getCurrentSelected(self) -> int:
"""
获取当前选中的表格单元索引
:return: 前选中的表格单元索引
"""
selectedIndex = self.tableView.selectedIndexes()
if len(selectedIndex) > 0:
return selectedIndex[0].row()
return -1
def insertRow(self, index: int) -> None:
"""
增加一行空白行
:index: 增加的位置索引
:return: None
"""
self.model.insertRow(index)
def getSubtitle(self) -> str:
"""
获取当前编辑的字幕信息
:return: 前编辑的字幕
"""
return self.singleText.text()
def getSearchText(self) -> str:
"""
获取搜索框文本信息
:return: 搜索框文本信息
"""
return self.searchText.text()
def setVideoText(self, text: str) -> None:
"""
设置视频预览区字幕信息
:param text: 字幕信息
:return: None
"""
self.videoText.setText(text)
def setFileName(self, fileName: str) -> None:
"""
设置文件名
:param fileName: 文件名
:return: None
"""
self.filenName.setText(fileName)
def windowIsVisible(self) -> bool:
"""
当前窗口是否可见,即是否被关闭
:return: 返回当前窗口是否可见,即是否被关闭
"""
return self.centralwidget.isVisible()
def setSubTileText(self, text: str) -> None:
"""
设置字幕编辑框内容
:param text: 字幕
:return: None
"""
self.singleText.setText(text)
self.singleText.home(False)
def setSearchText(self, text: str) -> None:
"""
设置搜索框内容
:param text: 搜索文字
:return: None
"""
self.searchText.setText(text)
self.searchText.home(False)
def getFileName(self) -> str:
"""
获取文件名
:return: 文件名
"""
return self.filenName.text()
def setRowText(self, row: int, text: str) -> None:
"""
设置某一行的文案信息
:param row: 行索引
:param text: 文本
:return:
"""
item = self.model.item(row, 0)
if item is not None:
item.setText(text)
def subtitleHasFocus(self) -> bool:
"""
单句字幕编辑框是否有焦点
:return: 单句字幕编辑框是否有焦点
"""
return self.singleText.hasFocus()
def setGenVideoText(self, text: str) -> None:
"""
设置生成视频按钮的文本
:param text: 文本
:return: None
"""
self.genVideo.setText(text)
================================================
FILE: preprocess.py
================================================
import os
import pickle
import paddlenlp
# 加载tokeniezer
from conf import MODELNAME
tokenizer = paddlenlp.transformers.GPTChineseTokenizer.from_pretrained(MODELNAME)
trainData = []
# 处理所有的公众号文章
for index, path in enumerate(os.listdir('banfoText')):
if not path.endswith('.txt'):
continue
print(index, path)
with open(os.path.join('banfoText', path), 'r+', encoding='utf-8') as f:
data = f.read()
data = tokenizer(text=data, return_token_type_ids=False)
data = data['input_ids']
start = -30
lenght = 100
step = 30
if len(data) <= 2 * lenght:
continue
# 滑动窗口截断获取inputData和label
while start + step + 1 < len(data) and start + step + lenght + 1 < len(data):
start = start + step
input_data = data[start: start + lenght]
label = data[start + 1: start + lenght + 1]
trainData.append([input_data, label])
trainData.append([data[-lenght-1: -1], data[-lenght:]])
if not os.path.exists('preprocessData'):
os.mkdir('preprocessData')
with open(os.path.join('preprocessData', 'trainData.pkl'), 'wb') as f:
pickle.dump(trainData, f)
print(len(trainData))
print('done!')
================================================
FILE: requirements.txt
================================================
moviepy==1.0.3
requests~=2.24.0
numpy==1.21.2
beautifulsoup4==4.10.0
PyQt5==5.15.4
system_hotkey==1.0.3
Flask~=2.0.2
bs4~=0.0.1
Pillow~=8.3.2
piexif~=1.1.3
googletrans~=3.0.0
paddlepaddle-gpu==2.1.3.post110
paddlenlp~=2.1.0
================================================
FILE: train.py
================================================
import os
import pickle
import paddle
import paddlenlp
from paddle.io import Dataset, DataLoader
import paddle.nn as nn
from conf import MODELNAME
class BanfoDataset(Dataset):
def __init__(self, data, tokenizer):
super().__init__()
self.data = data
self.tokenizer = tokenizer
def __getitem__(self, idx):
return paddle.to_tensor(self.data[idx][0], dtype='int64'), paddle.to_tensor(self.data[idx][1], dtype='int64')
def __len__(self):
return len(self.data)
if not os.path.exists('models'):
os.mkdir('models')
paddle.set_device('gpu')
gptModel = paddlenlp.transformers.GPTModel.from_pretrained(MODELNAME)
gptModel = paddlenlp.transformers.GPTForPretraining(gptModel)
tokenizer = paddlenlp.transformers.GPTChineseTokenizer.from_pretrained(MODELNAME)
# 有本地模型存在，则加载本地模型
checkpoint = os.path.join('models', 'model_state.pdparams')
if os.path.exists(checkpoint):
model_state = paddle.load(checkpoint)
gptModel.set_state_dict(model_state)
# 设置为评估模型
gptModel.eval()
# 测试效果
encodedText = tokenizer(text='前段时间我跟一个老大哥一起吃火锅。大哥的孩子，都上学了', return_token_type_ids=False)
ids, _ = gptModel.generate(input_ids=paddle.to_tensor(encodedText['input_ids'], dtype='int64').unsqueeze(0),
max_length=16, min_length=1, decode_strategy='sampling')
ids = ids[0].numpy().tolist()
# 使用tokenizer将生成的id转为文本
text = tokenizer.convert_ids_to_string(ids)
print('generation text is {}'.format(text))
# 加载训练数据
with open(os.path.join('preprocessData', 'trainData.pkl'), 'rb') as f:
data = pickle.load(f)
trainDataLoader = DataLoader(dataset=BanfoDataset(data, tokenizer), batch_size=64, shuffle=True, return_list=True)
numEpochs = 100
learningRate = 2e-5
warmupProportion = 0.1
weightDecay = 0.1
maxSteps = (len(trainDataLoader) * numEpochs)
lr_scheduler = paddle.optimizer.lr.LambdaDecay(learningRate,
lambda currentStep, numWarmupSteps=maxSteps * warmupProportion, numTrainingSteps=maxSteps: float(currentStep) / float(max(1, numWarmupSteps)) if currentStep < numWarmupSteps else max(0.0, float(numTrainingSteps - currentStep) / float(max(1, numTrainingSteps - numWarmupSteps))))
optimizer = paddle.optimizer.AdamW(learning_rate=lr_scheduler,
parameters=gptModel.parameters(),
weight_decay=weightDecay,
grad_clip=nn.ClipGradByGlobalNorm(1.0),
apply_decay_param_fun=lambda x: x in [
p.name for n, p in gptModel.named_parameters()
if not any(nd in n for nd in ["bias", "norm"])
])
globalStep = 1
save_steps = 100
criterion = paddle.nn.loss.CrossEntropyLoss()
gptModel.train()
for epoch in range(numEpochs):
for step, batch in enumerate(trainDataLoader, start=1):
ids, label = batch
logits, _ = gptModel.forward(ids, use_cache=True)
loss = criterion(logits, label)
loss.backward()
optimizer.step()
lr_scheduler.step()
optimizer.clear_gradients()
if globalStep % save_steps == 0:
print(globalStep, loss.numpy())
gptModel.save_pretrained('models')
globalStep += 1
================================================
FILE: utils.py
================================================
import base64
import hashlib
import json
import os
import random
from collections.abc import Iterable
from urllib.parse import quote
import piexif
import requests
from PIL import Image
from bs4 import BeautifulSoup
from config import APIKEY, BAIDUAPPID, BAIDUAPPKEY
proxies = None
verify = True
def getUuid() -> str:
"""
生成一个uuid
:return: uuid
"""
return str(random.randint(1, 999999)).zfill(6)
def getBaiDuAudio(text: str, filePath: str) -> [str, None]:
"""
利用百度语音合成将文字合成语音
:param text: 合成语音的文字
:param filePath: 文件保存路径
:return: 文件路径或者None
"""
# 生成文件名
md5 = hashlib.md5()
md5.update(text.encode('utf-8'))
fileName = md5.hexdigest() + '.mp3'
fileName = os.path.join(filePath, fileName)
# 文件已经存在的话直接返回
if os.path.exists(fileName):
return fileName
url = 'https://cloud.baidu.com/aidemo'
data = 'type=tns&per=4105&spd=8&pit=7&vol=5&aue=6&tex=' + quote(text)
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.81 Safari/537.36',
'Content-Type': 'application/x-www-form-urlencoded',
'Accept': '*/*',
'Accept-Language': 'zh-CN,zh;q=0.9,en-US;q=0.8,en;q=0.7'
}
res = requests.post(url, data=data, headers=headers, verify=False)
if res.status_code != 200:
return None
res = json.loads(res.text)
if res['msg'] != 'success':
return None
data = res['data'].replace('data:audio/x-mpeg;base64,', '')
if ',' in data:
data = data[:data.find(',')]
data = base64.b64decode(data)
with open(fileName, 'wb') as f:
f.write(data)
return fileName
def decodeBaiduImg(objUrl: str) -> str:
"""
百度图片地址解码函数
:param objUrl: 编码的url
:return: 解码的url
"""
res = ''
c = ['_z2C$q', '_z&e3B', 'AzdH3F']
d = {'w': 'a',
'k': 'b',
'v': 'c',
'1': 'd',
'j': 'e',
'u': 'f',
'2': 'g',
'i': 'h',
't': 'i',
'3': 'j',
'h': 'k',
's': 'l',
'4': 'm',
'g': 'n',
'5': 'o',
'r': 'p',
'q': 'q',
'6': 'r',
'f': 's',
'p': 't',
'7': 'u',
'e': 'v',
'o': 'w',
'8': '1',
'd': '2',
'n': '3',
'9': '4',
'c': '5',
'm': '6',
'0': '7',
'b': '8',
'l': '9',
'a': '0',
'_z2C$q': ':',
'_z&e3B': '.',
'AzdH3F': '/'}
for m in c:
objUrl = objUrl.replace(m, d[m])
for char in objUrl:
char = d[char] if char in d else char
res = res + char
return res
def getBaiduImgPath(text: str) -> [Iterable, None]:
"""
从百度图片接口拉取图片
:param text: 搜索关键字
:return: 图片路径迭代器或者None
"""
url = 'https://image.baidu.com/search/acjson?tn=resultjson_com&logid=8763701186511659178&ipn=rj&ct=201326592&is=' \
'&fp=result&queryWord={0}&cl=2&lm=-1&ie=utf-8&oe=utf-8&adpicid=&st=-1&z=&ic=0&hd=&latest=©right=&word={0}' \
'&s=&se=&tab=&width=&height=&face=0&istype=2&qc=&nc=1&fr=&expermode=&nojc=&acjsonfr=click&pn=0&rn=30&itg=1' \
'&gsm=3c&1634043752626='
url = url.format(text)
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.61 Safari/537.36',
'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9',
'Accept-Language': 'zh-CN,zh;q=0.9,en-US;q=0.8,en;q=0.7'}
res = requests.get(url, headers=headers, verify=False)
if res.status_code != 200:
return None
try:
jsonData = json.loads(res.text)
except:
return None
if 'data' not in jsonData:
return None
data = jsonData['data']
for img in data:
imgUrl = ''
if 'objURL' in img:
imgUrl = decodeBaiduImg(img['objURL'])
elif 'middleURL' in img:
imgUrl = img['middleURL']
elif 'thumbURL' in img:
imgUrl = img['thumbURL']
elif imgUrl == '':
return None
if 'is_gif' in img and img['is_gif'] == 1:
is_gif = 1
else:
is_gif = 0
res = requests.get(imgUrl)
if res.status_code != 200:
continue
md5 = hashlib.md5()
md5.update(res.content)
fileName = md5.hexdigest() + '.gif' if is_gif else md5.hexdigest() + '.png'
path = os.path.join('tmp', fileName)
with open(path, 'wb') as f:
f.write(res.content)
yield path
def getDoutulaImgPath(text: str) -> [Iterable, None]:
"""
从斗图啦表情包接口拉取图片
:param text: 搜索关键字
:return: 图片路径迭代器或者None
"""
url = 'https://www.doutula.com/search?keyword=' + text
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.61 Safari/537.36',
'Accept': 'text/html,application/xhtml+xml,application/xml;q=0.9,image/avif,image/webp,image/apng,*/*;q=0.8,application/signed-exchange;v=b3;q=0.9',
'Accept-Language': 'zh-CN,zh;q=0.9,en-US;q=0.8,en;q=0.7'}
res = requests.get(url, headers=headers)
if res.status_code != 200:
return None
soup = BeautifulSoup(res.text, 'html.parser')
randomPics = soup.find_all('a', attrs={'class': 'col-xs-6 col-md-2'})
for pic in randomPics:
imgUrl = pic.find('img', attrs={'referrerpolicy': 'no-referrer'})['data-original']
try:
res = requests.get(imgUrl)
except:
return None
if res.status_code != 200:
continue
path = os.path.join('tmp', imgUrl[imgUrl.rfind('/') + 1:])
with open(path, 'wb') as f:
f.write(res.content)
yield path
def resizeImg(width: int, height: int) -> (int, int):
"""
重新设置图片大小
:param width: 图片宽度
:param height: 图片高度
:return: (width, height)
"""
default_width, default_height = 440, 360
if (width < default_width and height < default_height) or (width > default_width and height > default_height):
# 如果比例大于1.5,就不强制拉伸,按照宽度进行缩放
if width / height >= 1.5:
width, height = default_width, height * default_width / width
elif height / width > 1.5:
height, width = default_height, width * default_height / height
else:
# 小于1.5的强制拉伸
width, height = default_width, default_height
elif width > default_width and height <= default_height:
width, height = default_width, height * default_width / width
if height > default_height and width <= default_width:
height, width = default_height, width * default_height / height
return width, height
def convertToRGB(path: str) -> None:
"""
如果图片不是RGB模式,则转换成RGB模式,否则生成视频会出错
:param path: 图片路径
:return: None
"""
# convert to RGB
im = Image.open(path)
if im.mode != 'RGB':
im = im.convert('RGB')
# 获取exif是否正常,不正常则添加
try:
im.getexif()
im.save(path)
except:
exif_dict = {}
exif_dat = piexif.dump(exif_dict)
im.save(path, exif=exif_dat)
def baiduTranslate(text: str, zhcn2en: bool = True) -> str:
"""
将中文翻译成英文
:param text: 中文/英文
:param zhcn2en : 是否为中文转英文, 默认为是
:return: 英文/中文
"""
url = 'https://fanyi-api.baidu.com/api/trans/vip/translate'
headers = {'Content-Type': 'application/x-www-form-urlencoded'}
data = BAIDUAPPID + text + url + BAIDUAPPKEY
md5 = hashlib.md5()
md5.update(data.encode('utf-8'))
if zhcn2en:
data = {'appid': BAIDUAPPID, 'q': text, 'from': 'zh', 'to': 'en', 'salt': url, 'sign': md5.hexdigest()}
else:
data = {'appid': BAIDUAPPID, 'q': text, 'from': 'en', 'to': 'zh', 'salt': url, 'sign': md5.hexdigest()}
res = requests.post(url, data=data, headers=headers)
if res.status_code != 200:
return ''
res = json.loads(res.text)
if 'trans_result' not in res:
return ''
result = ''
for trans in res['trans_result']:
result += trans['dst'] + '。'
return result
# TODO
# 有点儿问题, 翻译的结果为多行的时候解决比较麻烦
def googleTranslate(text: str, zhcn2en: bool = True) -> str:
"""
将中文翻译成英文
:param text: 中文/英文
:param zhcn2en : 是否为中文转英文, 默认为是
:return: 英文/中文
"""
global proxies
global verify
text = text.replace('\n', r'\\n').replace('"', r'\\\"')
url = 'https://translate.google.cn/_/TranslateWebserverUi/data/batchexecute'
if zhcn2en:
data = '[[["MkEWBc","[[\\"{}\\",\\"zh-CN\\",\\"en\\",true],[null]]",null,"generic"]]]'.format(text)
else:
data = '[[["MkEWBc","[[\\"{}\\",\\"en\\",\\"zh-CN\\",true],[null]]",null,"generic"]]]'.format(text)
data = quote(data)
data = 'f.req=' + data + '&'
headers = {
'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/94.0.4606.81 Safari/537.36',
'Content-Type': 'application/x-www-form-urlencoded;charset=UTF-8',
'Accept-Language': 'zh-CN,zh;q=0.9,en-US;q=0.8,en;q=0.7'}
res = requests.post(url=url, data=data, headers=headers, verify=verify, proxies=proxies)
if res.status_code != 200:
return ''
lIndex = res.text.find('l,[[\\"')
rIndex = res.text.find('\\"', lIndex + len('l,[[\\"'))
if lIndex != -1 and rIndex != -1:
text = res.text[lIndex + len('l,[[\\"'):rIndex]
text = text.replace(r'\\n', '\n').replace(r'\\\"', '"')
return text
return ''
def getTips(text: str) -> str:
"""
获取指定文本的生成文本信息
:param text: 线索文本
:return: 生成的文本
"""
global proxies
global verify
# 中文 -> 英文
text = baiduTranslate(text)
if text == '':
return ''
# 获取生成文本结果
url = 'https://api.deepai.org/api/text-generator'
data = {'text': text}
headers = {'api-key': APIKEY}
res = requests.post(url, files=data, headers=headers, verify=verify, proxies=proxies)
if res.status_code != 200:
return ''
res = json.loads(res.text)
if 'output' not in res:
return ''
text = res['output']
# 英文 -> 中文
text = baiduTranslate(text, zhcn2en=False).replace('\n\n', '\n')
return text
def setProxies(switch: bool=False) -> None:
"""
是否采用代理
:param switch: 是
:return: None
"""
global proxies
global verify
if switch:
proxies = {'https': '127.0.0.1:8887',
'http': '127.0.0.1:8887'}
verify = False
else:
proxies = None
verify = True