Type | |
Stats | 1,323 352 |
Reviews | (151) |
Published | Sep 5, 2024 |
Base Model | |
Training | Steps: 6,000 Epochs: 12 |
Usage Tips | Clip Skip: 2 |
Hash | AutoV2 6DE4E6D451 |
使用了肌理感相对强的油画进行的训练,因此使用过程中不推荐对画面进行过多元素的添加或者描述,否则物体和肌理感叠加会使得画面看起来非常混乱,示例图片的树林其实有点那个意思
- 适合生成大场景 + 单个人物的图像,因为背景的部分虚化 + 人物的具象利于表达想要传达的氛围
- 人像的高光很弱;
- 人物、静物等的特写也可以生成,但未对人像训练集进行特殊的设计和考虑;
推荐权重 0.8- 1.0
The model was trained using oil paintings with relatively strong textures, so it's not recommended to add or describe too many elements in the image during use. Otherwise, the overlapping of objects and textures will make the image look very messy. The forest in the sample image somewhat illustrates this.
Suitable for generating images with large scenes + single characters, because the blurred background + the concrete character is conducive to expressing the desired atmosphere.
The highlights on portraits are very weak.
Close-ups of people, still lifes, etc. can also be generated, but no special design or consideration was given to the portrait training set.
Recommended weight: 0.8-1.0