我报名参加金石计划1期挑战——瓜分10万奖池,这是我的第1篇文章,wenxin.baidu.com/moduleApi/e… 能力很强,最近一段时间,各式各样的创意层出不穷,看的我是心花怒放,恰好中秋快到了,自己也打算做一个。
文心大模型制作MV《十五的月亮》_哔哩哔哩_bilibili
www.bilibili.com/video/BV1PY…
二、制作步骤
1.ak和sk申请
这一步主要就是获取官方的ak和sk
登录文心大模型官网后,在顶部选择
右上角“创建API key”,记住这两个key,在之后会用到。当然,你也不要和别人分享你的key,好好保存。
2.编辑Promot
以歌词为基准,进行描述性修改
三、运行项目
1.fork项目
直接fork磊哥哥项目,即可
2.添加风格描述
可以参考 文心大模型-产业级知识增强大模型Promot指南,如比较喜欢像素艺术,可以在生成时,text后添加像素艺术即可。有多重艺术风格可选:
promot修改:
for index, txt in enumerate(self.lrcs):
print(index, txt)
if index >= startindex:
textlist = txt.split("@:")
if len(textlist) > 1 and textlist[-1] is not '':
txt = textlist[-1]
txt = txt.replace("n", "")
txt = txt + "中国风,像素艺术"
print("real text:", txt)
idict = self.getImageDict(txt)
self.readAndSave(idict, text=txt, ind=index)
3.参考指南
美学相关的词汇: aesthetics.fandom.com/wiki/List_o…
DALL-E 2 的 Prompt 技巧资料: docs.google.com/document/d/…
DiscoDiffusion Prompt 技巧资料: docs.google.com/document/d/…
四、生成图概略
五、代码
import wenxin_api
from wenxin_api.tasks.text_to_image import TextToImage
import requests
import argparse
import sys
import os
styleDict = ["油画", "水彩", "卡通", "粉笔画", "儿童画", "蜡笔画"]
class MV:
def __init__(self):
pass
def getImageDict(self, text):
input_dict = {
"text": text,
"style": self.style,
}
return TextToImage.create(**input_dict)
def readAndSave(self, imgDict, text=None, ind=0):
for index, url in enumerate(imgDict['imgUrls']):
r = requests.get(url)
imgName = ""
if text is None:
imgName = imgName + url.split('/')[-1] + '.png'
else:
imgName = imgName + str(ind) + "_" + text + "_" + str(index) + '.png'
out_ii = os.path.join(self.songName, imgName)
with open(out_ii, 'wb') as f:
f.write(r.content)
f.close()
def make(self, lrc=None, songName=None, style=0, startindex=0):
if style < 6 and style >= 0:
self.style = styleDict[style]
else:
try:
sys.exit(0)
except:
print("Wrong Style")
self.lrcs = []
if os.path.exists(lrc):
with open(lrc, 'r', encoding='UTF-8') as fp:
lines = fp.readlines()
for line in lines:
self.lrcs.append(line)
else:
self.lrcs.append(lrc)
print(self.lrcs)
if songName is not None:
self.songName = songName
else:
if os.path.exists(lrc):
self.songName = os.path.basename(lrc).split(".")[0]
else:
self.songName = lrc
os.makedirs(self.songName, exist_ok=True)
for index, txt in enumerate(self.lrcs):
print(index, txt)
if index >= startindex:
textlist = txt.split("@:")
if len(textlist) > 1 and textlist[-1] is not '':
txt = textlist[-1]
txt = txt.replace("n", "")
txt = txt + "中国风,像素艺术"
print("real text:", txt)
idict = self.getImageDict(txt)
self.readAndSave(idict, text=txt, ind=index)
wenxin_api.ak = "xxxxxxxx"
wenxin_api.sk = "xxxxxxxxx"
mv = MV()
def main(args):
mv.make(args.lrc, args.songName, args.style, args.startindex)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--lrc', type=str, required=True)
parser.add_argument('--songName', type=str, required=False)
parser.add_argument('--style', type=int, default=0, required=False)
parser.add_argument('--startindex', type=int, default=0, required=False)
args = parser.parse_args()
main(args)