Type | |
Stats | 849 |
Reviews | (93) |
Published | Sep 26, 2023 |
Base Model | |
Hash | AutoV2 CC22438DC0 |
recommend to use natural language and more sytle prompts
推荐使用自然语言和足够多的质量/画风prompts
Recommend:
step:35,DPM++ 3M SDE Karras,
recommand prompts in show image.
Here are a few examples:
example 1
{ "prompt": "breathtaking,{prompt} . gorgeous,key visual, vibrant, studio anime,award-winning, professional, highly detailed,high budget, cinemascope",
"negative_prompt": "noisy, blurry, grainy,text, graphite, abstract, glitch, deformed, mutated, ugly, disfigured,black and white, low contrast"},
example 2
{"prompt": "Fairy tale {prompt} . Magical, fantastical, enchanting, storybook style, highly detailed,gorgeous, vibrant, award-winning, professional, ",
"negative_prompt": "realistic, modern, ordinary, mundane"}
Train base Kohaku-XL , this model after MBW to remove most of the contamination,while preserving the generalization ability of the base model.
Image will be closer to kohaku-XL,only retaining the style of my interference,it's normal, I do it on purpose!
基于 kohaku-XL模型训练,训练后与底模进行了MBW去除了大部分的污染,保留了底模的泛化性。出图会比较接近kohaku-XL,只保留了我干涉的画风,很正常,我故意的。