蓝耘怎么调用通义万相-通义万相2.1生成视频详解及代码

class BlueFarming:  def __init__(self):      self.supported_gpus = ["NVIDIA A100", "NVIDIA V100", "NVIDIA H100"]      self.software_technologies = ["Kubernetes", "Docker"] def describe_hardware(self):      gpu_list = ", ".join(self.supported_gpus)      print(f"蓝耘平台的硬件部分支持以下高端 GPU: {gpu_list}。") def describe_software(self):      software_list = ", ".join(self.software_technologies)      print(f"蓝耘平台在软件方面集成了 {software_list} 技术，为计算提供高效环境。") def calculate(self):      print("蓝耘平台像超级计算器一样，能快速准确地完成复杂计算任务。")


if __name__ == "__main__":  blue_farming = BlueFarming()  blue_farming.describe_hardware()  blue_farming.describe_software()  blue_farming.calculate()

import time


class TongyiWanxiang2_1:  def __init__(self):      # 模拟模型采用的架构和技术      self.architectures = ["VAE 架构", "DiT 架构"]      self.technologies = ["IC - LoRA 技术"]      print("通义万相 2.1 已就绪。") def generate_content(self, text_description, output_type="图片"):      print(f"接收到描述: {text_description}，请求生成 {output_type}。")      print("正在使用 VAE 架构、DiT 架构以及 IC - LoRA 技术进行处理...")      # 模拟处理时间      time.sleep(2)      if output_type == "图片":          print(f"已生成符合描述的图片。")      elif output_type == "视频":          print(f"已生成符合描述的视频，该视频质量高、细节丰富。")      elif output_type == "3D 内容":          print(f"已生成符合描述的 3D 内容。")      else:          print("不支持的输出类型。")      return f"生成的 {output_type}"


# 使用示例
if __name__ == "__main__":  model = TongyiWanxiang2_1()  description = "我想要一个美丽的海边日落场景，有金色的沙滩和蓝色的大海"  result = model.generate_content(description, output_type="视频")  print(result)

pip install requests  # 用于发送网络请求
pip install torch  # PyTorch 深度学习框架
pip install torchvision  # 用于处理图像和视频
pip install opencv-python  # 用于计算机视觉任务

import requests
import json

# 通义万相 2.1 API 地址，这里需要替换成实际的地址
api_url = "https://your-tongyiwanxiang2.1-api-url"
# 我的 API Key，就像前面说的钥匙
headers = {  "Content-Type": "application/json",  "Authorization": "Bearer YOUR_ACCESS_TOKEN"
}

# 我输入的文本提示，描述我想要的视频场景
prompt = {  "场景描述": "一个宁静的公园，有绿树和湖泊，阳光洒在地面上",  "风格": "清新写实风格",  "时长": 10,  "帧率": 24
}

data = {  "prompt": prompt,  "parameters": {      "width": 1280,      "height": 720  }
}

try:  # 发送请求给通义万相 2.1  response = requests.post(api_url, headers=headers, data=json.dumps(data))  response.raise_for_status()  result = response.json()  frame_urls = result["frame_urls"]  print("成功获取视频帧 URL 列表")
except requests.RequestException as e:  print(f"请求出错: {e}")
except KeyError:  print("返回结果格式有误，未找到预期的视频帧 URL 列表。")

import requests
import cv2
import numpy as np
from torchvision import transforms
import torch

# 假设蓝耘平台有 GPU 可用，这样计算会更快
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 定义图像预处理函数，把图像转换为适合模型处理的格式
preprocess = transforms.Compose([  transforms.ToTensor(),  transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

frames = []
for url in frame_urls:  try:      # 下载视频帧      frame = cv2.imdecode(np.frombuffer(requests.get(url).content, np.uint8), cv2.IMREAD_COLOR)      frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)      frame_tensor = preprocess(frame).unsqueeze(0).to(device)      frames.append(frame_tensor)  except Exception as e:      print(f"下载或处理视频帧时出错: {e}")

print("视频帧下载并预处理完成")

import torch.nn.functional as F

# 定义视频帧增强函数，添加光影效果
def enhance_frame(frame):  # 应用锐化卷积核，让画面更清晰  kernel = torch.tensor([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]], dtype=torch.float32).unsqueeze(0).unsqueeze(0).to(device)  enhanced_frame = F.conv2d(frame, kernel, padding=1) # 添加光影效果，模拟阳光的明暗变化  light_mask = torch.randn_like(frame) * 0.1  enhanced_frame = enhanced_frame + light_mask  enhanced_frame = torch.clamp(enhanced_frame, 0, 1)  return enhanced_frame

enhanced_frames = []
for frame in frames:  enhanced_frame = enhance_frame(frame)  enhanced_frames.append(enhanced_frame)

print("视频帧增强处理完成")

from torchvision.utils import make_grid
import torchvision.io as io

# 将处理后的视频帧转换为适合保存的格式
output_frames = []
for frame in enhanced_frames:  frame = frame.squeeze(0).cpu()  frame = make_grid(frame, normalize=True, scale_each=True)  frame = (frame * 255).byte().permute(1, 2, 0).numpy()  output_frames.append(frame)

import cv2

# 保存处理后的视频
fps = prompt["帧率"]
height, width, _ = output_frames[0].shape
fourcc = cv2.VideoWriter_fourcc(*'XVID')
out = cv2.VideoWriter('output_video.avi', fourcc, fps, (width, height))
for frame in output_frames:  frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)  out.write(frame)
out.release()

print("处理后的视频已保存为 output_video.avi")