| import torch | |
| from torchvision.utils import save_image | |
| import os | |
| import numpy as np | |
| import tempfile | |
| from model import Generator # فرضاً مدل StyleGAN2 یا مشابه را import میکنیم | |
| # لود مدل pretrained (مثلاً unconditional anime face) | |
| device = "cpu" | |
| G = Generator() # مدل از قبل train شده | |
| G.load_state_dict(torch.load("unconditional_anime_face.pth", map_location=device)) | |
| G.to(device) | |
| G.eval() | |
| # تولید تصویر بدون پرامپت | |
| z = torch.randn(1, 512, device=device) # latent vector | |
| with torch.no_grad(): | |
| img = G(z) | |
| # ذخیره تصویر | |
| tmpdir = tempfile.mkdtemp() | |
| file_path = os.path.join(tmpdir, "output.png") | |
| save_image((img + 1) / 2, file_path) # normalize [-1,1] -> [0,1] | |
| print("Saved:", file_path) |