【AI绘画】Midjourney进阶：中心点构图详解

-  从非共性的角度来看，在Midjourney中使用构图相关的提示词可以显著提升生成画面的质量。这是因为Midjourney的模型在训练时，不仅使用了图像，还结合了与这些图像相关的提示词。通常来说，这些图像来源于专业摄影或高质量渠道，具有较高的视觉标准。当模型学习了这些图片与提示词之间的关联后，用户在生成图像时如果使用了构图相关的提示词，系统会优先选择类似图库中的优质素材，从而生成更为精致的画面。因此，通过使用构图提示词，可以明显地提高生成图像的质量。

-  从共性的角度来说，构图本身就是一种极为有效的手段，可以用于引导观众的视线、突出画面的主题、增强视觉吸引力，并传递创作者的情感和观点。构图所带来的这些视觉上的提升效果，是不受具体工具限制的，无论是在摄影、绘画还是AI生成图像中都是通用的。因此，当在Midjourney中使用构图提示词时，画面的质量提升几乎是必然的结果。这些道理或许听起来有些抽象，但在实际操作和练习时，大家会切实地感受到构图带来的效果提升。

- 在人像图中，中心点构图能够有效突出人物的面部表情和整体特点，使得观众的视线集中在被摄人物的核心部分。

- 在静物图中，这种构图方式可以很好地突出静物的细节和特征，吸引观众对静物的质感和形状进行深入观察。

- 在风景图中，中心点构图有助于突出风景中的主要元素，使画面的主题更加明确，便于观众理解景象的重点所在。

lavish dining table set with a variety of gourmet dishes, wine glasses, candles, and floral centerpiece, warm ambient light, luxurious atmosphere, 8k, ultra details

lavish dining table with a single ornate dish as the centerpiece, Center Point Composition, surrounded by wine glasses and candles, warm ambient light, luxurious atmosphere, 8k, ultra details

import torch, torchvision.transforms as transforms; from torchvision.models import vgg19; import torch.nn.functional as F; from PIL import Image; import matplotlib.pyplot as plt; class StyleTransferModel(torch.nn.Module): def __init__(self): super(StyleTransferModel, self).__init__(); self.vgg = vgg19(pretrained=True).features; for param in self.vgg.parameters(): param.requires_grad_(False); def forward(self, x): layers = {'0': 'conv1_1', '5': 'conv2_1', '10': 'conv3_1', '19': 'conv4_1', '21': 'conv4_2', '28': 'conv5_1'}; features = {}; for name, layer in self.vgg._modules.items(): x = layer(x); if name in layers: features[layers[name]] = x; return features; def load_image(img_path, max_size=400, shape=None): image = Image.open(img_path).convert('RGB'); if max(image.size) > max_size: size = max_size; else: size = max(image.size); if shape is not None: size = shape; in_transform = transforms.Compose([transforms.Resize((size, size)), transforms.ToTensor(), transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))]); image = in_transform(image)[:3, :, :].unsqueeze(0); return image; def im_convert(tensor): image = tensor.to('cpu').clone().detach(); image = image.numpy().squeeze(); image = image.transpose(1, 2, 0); image = image * (0.229, 0.224, 0.225) + (0.485, 0.456, 0.406); image = image.clip(0, 1); return image; def gram_matrix(tensor): _, d, h, w = tensor.size(); tensor = tensor.view(d, h * w); gram = torch.mm(tensor, tensor.t()); return gram; content = load_image('content.jpg').to('cuda'); style = load_image('style.jpg', shape=content.shape[-2:]).to('cuda'); model = StyleTransferModel().to('cuda'); style_features = model(style); content_features = model(content); style_grams = {layer: gram_matrix(style_features[layer]) for layer in style_features}; target = content.clone().requires_grad_(True).to('cuda'); style_weights = {'conv1_1': 1.0, 'conv2_1': 0.8, 'conv3_1': 0.5, 'conv4_1': 0.3, 'conv5_1': 0.1}; content_weight = 1e4; style_weight = 1e2; optimizer = torch.optim.Adam([target], lr=0.003); for i in range(1, 3001): target_features = model(target); content_loss = F.mse_loss(target_features['conv4_2'], content_features['conv4_2']); style_loss = 0; for layer in style_weights: target_feature = target_features[layer]; target_gram = gram_matrix(target_feature); style_gram = style_grams[layer]; layer_style_loss = style_weights[layer] * F.mse_loss(target_gram, style_gram); b, c, h, w = target_feature.shape; style_loss += layer_style_loss / (c * h * w); total_loss = content_weight * content_loss + style_weight * style_loss; optimizer.zero_grad(); total_loss.backward(); optimizer.step(); if i % 500 == 0: print('Iteration {}, Total loss: {}'.format(i, total_loss.item())); plt.imshow(im_convert(target)); plt.axis('off'); plt.show()