165 lines
6.2 KiB
Python
165 lines
6.2 KiB
Python
# -*- coding: utf-8 -*-
|
||
"""
|
||
主模块
|
||
"""
|
||
import warnings
|
||
|
||
warnings.filterwarnings(
|
||
action="ignore", category=UserWarning, module="volcenginesdkarkruntime.*"
|
||
)
|
||
|
||
import json
|
||
from pathlib import Path
|
||
from typing import Any, Dict, List
|
||
from uuid import uuid4
|
||
|
||
from volcenginesdkarkruntime import Ark
|
||
from volcenginesdkarkruntime.types.responses import (
|
||
ResponseOutputMessage,
|
||
ResponseOutputText,
|
||
)
|
||
|
||
|
||
# 初始化火山引擎 AsyncArk 客户端
|
||
ark_client = Ark(
|
||
base_url="https://ark.cn-beijing.volces.com/api/v3",
|
||
api_key="2c28ab07-888c-45be-84a2-fc4b2cb5f3f2",
|
||
) # 本人火山引擎账密
|
||
|
||
|
||
def generate_task_id() -> str:
|
||
"""
|
||
生成任务标识
|
||
:return: 任务ID
|
||
"""
|
||
return uuid4().hex.upper().replace("-", "")
|
||
|
||
|
||
def get_brand_words() -> List[str]:
|
||
"""
|
||
获取品牌词
|
||
:return: 品牌词
|
||
"""
|
||
try:
|
||
with open(
|
||
file=Path(__file__).parent / "brand_words.txt", mode="r", encoding="utf-8"
|
||
) as file: # Trae IDE 需要指定文件路径（和 PyCharm 不同）
|
||
brand_words = [line.strip() for line in file.readlines() if line.strip()]
|
||
assert brand_words, "品牌词为空"
|
||
return brand_words
|
||
except FileNotFoundError:
|
||
raise FileNotFoundError("未找到品牌词文件")
|
||
except Exception as exception:
|
||
raise exception
|
||
|
||
|
||
def refactor_storyboard_prompt(brand_word: str) -> str:
|
||
"""
|
||
重构分镜脚本的提示词
|
||
:param brand_word: 品牌词
|
||
:return: 重构后分镜脚本的提示词
|
||
"""
|
||
try:
|
||
with open(
|
||
file=Path(__file__).parent / "storyboard_prompt_template.txt",
|
||
mode="r",
|
||
encoding="utf-8",
|
||
) as file:
|
||
storyboard_prompt = file.read()
|
||
return storyboard_prompt.replace("{{品牌词}}", brand_word)
|
||
except FileNotFoundError:
|
||
raise FileNotFoundError("未找到分镜脚本的提示词模板文件")
|
||
except Exception as exception:
|
||
raise exception
|
||
|
||
|
||
def get_storyboard(brand_word: str) -> Dict[str, Any]:
|
||
"""
|
||
获取分镜脚本
|
||
:param brand_word: 品牌词
|
||
:return: 分镜脚本
|
||
"""
|
||
# 重构分镜脚本的提示词
|
||
storyboard_prompt = refactor_storyboard_prompt(brand_word)
|
||
|
||
retries = 0 # 重试次数
|
||
while True:
|
||
try:
|
||
# 调用豆包Seed语言大模型，基于分镜脚本的提示词生成分镜脚本
|
||
response = ark_client.responses.create(
|
||
model="doubao-seed-2-0-pro-260215",
|
||
input=[
|
||
{
|
||
"role": "user",
|
||
"content": [{"type": "input_text", "text": storyboard_prompt}],
|
||
}
|
||
],
|
||
)
|
||
# 解析响应并反序列化以此作为分镜脚本
|
||
storyboard = json.loads(
|
||
[item for item in [item for item in response.output if isinstance(item, ResponseOutputMessage)][0].content if isinstance(item, ResponseOutputText)][0].text # type: ignore
|
||
)
|
||
return storyboard
|
||
except Exception as exception:
|
||
retries += 1
|
||
if retries > 2:
|
||
raise Exception(f"获取分镜脚本发生异常，{str(exception)}")
|
||
continue
|
||
|
||
|
||
storyboard = {
|
||
"核心营销要点": "淘宝闪购每日上新全品类正品好物，限时超低价开抢，购物省心又划算",
|
||
"分镜脚本": [
|
||
{
|
||
"阶段": "前段",
|
||
"口播": "买好物想省钱看这里！",
|
||
"TTS情绪": "energetic",
|
||
"首帧提示词": "橙色淘宝闪购logo 画面中央偏上 半透明小尺寸、特效未展开 浅橙色渐变背景 亮橙+米白 柔和光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"尾帧提示词": "橙色立体淘宝闪购logo 画面正中央 清晰显示、周围带细碎金光闪效 浅橙色渐变背景 亮橙+暖白 明亮光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"运镜提示词": "slight zoom in",
|
||
},
|
||
{
|
||
"阶段": "中段",
|
||
"口播": "淘宝闪购全是官方正品，服饰美妆数码家居全品类，每天限时开抢价格超划算",
|
||
"TTS情绪": "happy",
|
||
"首帧提示词": "服饰、美妆、数码、家居四类商品缩略图围绕淘宝闪购logo 画面中央 半透明小尺寸、特效未展开 暖白色背景 多彩明快 柔和光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"尾帧提示词": "服饰、美妆、数码 、家居爆款商品清晰展示，旁附「直降50%」「限时抢」标签，淘宝闪购logo居上方中央 高亮饱满、爆闪优惠光效完全展开 暖白色背景 多彩明亮 充足光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"运镜提示词": "slow push in",
|
||
},
|
||
{
|
||
"阶段": "后段",
|
||
"口播": "快点击下方链接抢！",
|
||
"TTS情绪": "cheer",
|
||
"首帧提示词": "黄色「点击下方链接」按钮+淘宝闪购logo 画面中央 半透明小尺寸、特效未展开 亮橙色背景 亮黄+橙红 柔和光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"尾帧提示词": "高亮立体「点击下方链接」按钮+闪烁箭头指向屏幕下方，淘宝闪购logo居按钮上方 完全展示、高亮饱满、脉冲光效完整 亮橙色背景 亮黄+橙红 明亮耀眼光影 9:16竖屏 1080P 高清干净 无水印",
|
||
"运镜提示词": "slight zoom in",
|
||
},
|
||
],
|
||
}
|
||
|
||
|
||
def generate_frame(frame_prompt: str) -> str:
|
||
"""
|
||
生成视频帧
|
||
:param frame_prompt: 视频帧提示词
|
||
:return: 视频帧
|
||
"""
|
||
# 调用豆包Seed图像大模型，基于视频帧提示词生成视频帧
|
||
response = ark_client.responses.create(
|
||
model="doubao-seed-2-0-pro-260215",
|
||
input=[
|
||
{
|
||
"role": "user",
|
||
"content": [{"type": "input_text", "text": frame_prompt}],
|
||
}
|
||
],
|
||
)
|
||
# 解析响应并反序列化，以此作为视频帧
|
||
frame = json.loads(
|
||
[item for item in [item for item in response.output if isinstance(item, ResponseOutputMessage)][0].content if isinstance(item, ResponseOutputText)][0].text # type: ignore
|
||
)
|
||
return frame
|
||
|
||
|
||
print(storyboard)
|