English introduction
I'm Designer A Liang, a trainer dedicated to providing efficiency-enhancing models for designers.
This is a model I trained for the "24 Solar Terms" illustration, capable of generating common solar term-related illustration scenes. Based on raw images, you can integrate your own IP for a fusion. Everyone is encouraged to try and share their results.
---
Based on the example image: write the format to combine keywords (the model is quite stable, most anime-style models can be tried, recommended models can all be found on liblib)
Trigger word: Art(Illustration)
- Model weight: 0.4-0.6
- Model recommendation: Anime-style model
- Examples: Counterfeit/AWPainting/Classical Dazzling Color, etc.
- Sampling steps: 20-35
- CFG suggestion: 4-6
- Size suggestion: 512*1024, 768*1024,
- Sampler: DPM++ 3M SDE Exponential
---
The source images are all included in the example images, you can copy the parameters to produce similar pictures,
Tips: If you need the image to be more generalized, after trying 512*1024 and finding it unsatisfactory, you can try 768*1024 for more generalization.
---
Image output suggestions:
If you don't have a clear idea, you can use the following prompt structure for a "blind box draw":
Art(Illustration) + Season(Spring) + cool colors
Tips: Art(Illustration) + Season(Spring) + cool colors
Based on the Season(Spring) structure, you can replace it with:
Season(Spring)
Season(Summer)
Season(Autumn)
Season(Winter)
Based on defining the image's tone with cool colors, you can replace it with:
warm colors
cool colors
---
If you have more specific image output needs, you can compile your own reference structure:
Image control (common elements):
mountain, tea garden, rain, grass, forest, tree, maple, path
day, night, sunset, cloudy sky, reflection, etc.
Other fixed vocabulary:
outdoors, no humans,
PS: If you need to add some noise to the image, you can try adding a texture keyword.
---
Tips: Currently, some of the model's outputs have the drawback of being "greyish," requiring manual color correction to enhance the texture.
Chinese introduction
我是设计师阿良，一位致力为设计师提供增效模型的训练家。
这是我训练的一款【24节气】插画模型，可以生成常见的节气类插画场景，基于生图你添加上自家IP进行融合，具体的大家可以多多尝试返图；
——
基于示例图：书写格式，来组合关键词（模型比较稳定，大部分二次元模型均可尝试，推荐模型在liblib上均可找到）
触发词：Art\(Illustration\)
模型权重：0.4-0.6
模型推荐：二次元模型
举例：Counterfeit/AWPainting/古典炫彩等
采样步数：20-35
CFG建议：4-6
尺寸建议：512*1024,768*1024,
采样器：DPM++ 3M SDE Exponential
——
素材用图，均已放在示例图中，你可以复制参数产出类似的图片，
Tips：如果你需要画面更泛化点，尝试512*1024不佳后，可以尝试768*1024会有更多泛化
——
出图建议：
若你如果没有明确的想法，你可以按照如下的prompt的结构，进行盲盒抽卡：
Art\(Illustration\) + Season\(Spring\) + cool colors
Tips：艺术\（插画\）+季节\（春天\）+冷色
基于Season\(Spring\)的结构，你可以替换成：
Season\(Spring\)
Season\(summer\)
Season\(Autumn\)
Season\(Winter\)
基于cool colors定义画面色调，你可以替换成：
warm colors
cool colors
——
若你有更具体的出图需求，可按照参考图的结构自行编撰：
画面控制（常见元素）：
mountain，tea garden，rain，grass，forest，tree，maple，path
day，night，sunset，cloudy sky，reflection
其他固定词汇：
outdoors, no humans，
PS：如果你需要画面加点噪点，你可以尝试加入texture提示词
——
Tips：模型目前部分产出图有【画面灰】的缺点，需要人工校色一下，提升质感，具体的画面应用，你可以参考下图