Type | |
Stats | 472 285 |
Reviews | (55) |
Published | Mar 5, 2024 |
Base Model | |
Training | Steps: 20,000 Epochs: 20 |
Usage Tips | Clip Skip: 2 Strength: 0.8 |
Trigger Words | colorful inkpainting |
Hash | AutoV2 4C81252766 |
LoRA Introduction:
This is actually a style model that started a long time ago, This LoRA 1.0 version includes a total of 383 selected materials. The main challenge is that the style wd14 does not tag well, and manual processing is required, otherwise, more materials could be used. For the subsequent XL version, I will use cogvlm or gpt4v for tagging.
The inspiration comes from a colorful loong (based on Master Wu Guanzhong's ink painting) that I saw on a social platform. If I only reproduce a dragon, it would be too boring. Therefore, LoRA 1.0 version includes some famous scenic spots in China, including the Forbidden City, the Great Wall, the Lijiang River, the Yangtze River, Guilin Scenery with Hills and Waters, Terracotta Army, Huangshan Mountain, Huangguoshu Falls, etc. At the same time, Suzhou-style gardens, ancient costumes, and more can be well restored. Of course, if the features of the scenic spots are not obvious, it is a kind of style migration itself.
In the future, I may add an XL version, incorporating more scenic spots, and so on.
Usage Instructions:
Refer to the example image, the Sampler is recommended to restart with about 20 steps, and enlargement is recommended with R-ESRGAN 4x+ Anime6B.
Train the base model AWPainting v1.2, which can be used as a base map. Other models can be tested; LoRA weight is around 0.8.
Trigger words: colorful inkpainting. Adding trigger words can enhance the color to some extent.
模型介绍：
这其实这是一个很早之前就开始做的画风模型，本LoRA 1.0版本共计383张挑选过的素材，主要是这种风格wd14打标效果不好，需要手动处理，不然可以更多。后续XL版本我会使用cogvlm或者gpt4v打标。
灵感来源于在社交平台看到的一条五彩斑斓的loong（基础是吴冠中大师的水墨画），如果只是复现一条龙那么太无趣了；所以本LoRA 1.0版本内置了些许国内的名胜古迹，包括：故宫 （The Forbidden City）、长城（The Great Wall）、漓江（Lijiang River）、长江（The Yangtze River）、桂林山水（Guilin Scenery with Hills and Waters）、兵马俑（Terracotta Army）、黄山（Huangshan Mountain）、黄果树瀑布（Huangguoshu Falls）等，同时苏式园林、古装人物等等都可以很好的还原。当然如果景点特征不明显，那么本身就是一种画风的迁移。
后续如果我应该会增加XL版本，加入更多国内的景点等。
使用说明
1、参考例图，采样器（Sampler）推荐 Restart，Steps 20左右，放大推荐R-ESRGAN 4x+ Anime6B。
2、训练底模AWPainting v1.2，可以此底模出图，其他自行测试；LoRA权重0.8左右。
3、触发词：colorful inkpainting 加了触发词能在一定程度上加强色彩。