Type | |
Stats | 2,807 |
Reviews | (476) |
Published | Nov 19, 2023 |
Base Model | |
Training | Steps: 10 Epochs: 20 |
Usage Tips | Clip Skip: 2 Strength: 0.8 |
Trigger Words | blue archive |
Hash | AutoV2 76C4B9CF4A |
大家好,我是chosen鲷,这次带来的是碧蓝档案平涂风格lora。
训练集全是碧蓝档案的cg图,我一直非常喜欢这游戏的美术风格,和碧蓝航线那种吃百家饭的立绘不同,它的画风非常统一,而且明亮又温馨,看起来非常舒服。虽然是个韩国游戏,但是画风却比日系游戏还日系,感觉很像日本番剧里的截图的感觉。后面有时间的话,我还会训练ba背景风格的lora。
lora有点过拟合,建议权重0.8。触发词:blue archive
推荐的tag:
(masterpiece:1.2),best quality,PIXIV,blue archive
这里写想要的内容
<lora:blue archive-000018:0.8>,
Negative prompt: EasyNegative,badhandsv5-neg,Subtitles,word,
Steps: 32, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 3089360094, Size: 768x512, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.5, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.11.0, Hires upscale: 2, Hires steps: 12, Hires upscaler: R-ESRGAN 4x+ Anime6B, Lora hashes: "blue archive-000018: 3fc8d3abb551", TI hashes: "EasyNegative: 66a7279a88dd, badhandsv5-neg: aa7651be154c", Version: v1.6.0-2-g4afaaf8a
大模型我推荐用我的ghost或者chosen irises以达到最佳效果(下方推荐),其他的常见常见二次元模型也均可,推荐使用偏二次元的模型。
感谢大家的下载使用,如果喜欢的话欢迎关注我,我的bilibili pixiv ID均为chosen鲷,欢迎大家关注我的频道https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0,https://www.pixiv.net/users/40911695 谢谢。
The following content is from Google Translate
Hello everyone, I am chosen鲷. This time I bring you the Blue Archive style Lora .
The training dataset consists of cg images from Blue Archive. I have always loved the art style of this game. Unlike the character illustrations in Azur Lane that feature various artists, Blue Archive's art style is consistent, bright, and warm, which makes it very comfortable to look at. Although it is a Korean game, its art style feels even more Japanese than some Japanese games. It gives me the feeling of screenshots from Japanese anime. If I have time later, I will also train a background style for Lora.
Lora tends to overfit, so I recommend using a weight of 0.8. Trigger phrase: blue archive.
Recommended tags:
(masterpiece:1.2), best quality, PIXIV, blue archive
Write the desired content here
<lora:blue archive-000018:0.8>,
Negative prompt: EasyNegative, badhandsv5-neg, Subtitles, word, Steps: 32, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 3089360094, Size: 768x512, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, Denoising strength: 0.5, Clip skip: 2, ADetailer model: face_yolov8n.pt, ADetailer confidence: 0.3, ADetailer dilate erode: 4, ADetailer mask blur: 4, ADetailer denoising strength: 0.4, ADetailer inpaint only masked: True, ADetailer inpaint padding: 32, ADetailer version: 23.11.0, Hires upscale: 2, Hires steps: 12, Hires upscaler: R-ESRGAN 4x+ Anime6B, Lora hashes: "blue archive-000018: 3fc8d3abb551", TI hashes: "EasyNegative: 66a7279a88dd, badhandsv5-neg: aa7651be154c", Version: v1.6.0-2-g4afaaf8a
For the big model, I recommend using my ghost or chosen irises to achieve the best results (recommended below). Other common second-dimensional models can also be used, and models with a leaning towards the second dimension are recommended.
Thank you all for downloading and using it. If you like it, please follow me. My bilibili and pixiv IDs are both chosen鲷. Welcome to follow my channels: https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0, https://www.pixiv.net/users/40911695. Thank you.