大家好，我是chosen鲷，这次带来的是巧克力lora。
这是一个风格类lora，训练的图片全部来自于nai3，之所以训练这个，是因为我用nai3出的很多图想要通过sd放大，但是发现不管用什么模型放大，都会对原有图片风格有所改变，为了克服这一点，我用我nai3出的一百多张图训练了这个lora，这样我用sd放大nai3图片的时候，只要挂着这个lora，再找一个类似于ghost这种兼容性好的大模型，就可以不改变原有风格放大图片了。
当然了，直接把这个lora用于出图也是可以的，例图中的图片都是用sd1.5直出的。画风非常细腻厚重，所以我给它取名为巧克力。
lora不会过拟合，建议权重0.9。触发词：chocolate
推荐的tag：
Chocolate,
这里写你想要的内容
<lora:Chocolate-000018:0.8>,
Negative prompt: EasyNegative,badhandsv5-neg,Subtitles,word,
Steps: 32, Sampler: DPM adaptive, CFG scale: 7, Seed: 611174500, Size: 1088×1984, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, VAE hash: ab33ad55c6, VAE: YOZORA.vae.pt, Denoising strength: 0.15, Clip skip: 2, ENSD: 31337, SD upscale overlap: 64, SD upscale upscaler: R-ESRGAN 4x+ Anime6B, Lora hashes: “Chocolate-000018: 544834e09a89”, TI hashes: “EasyNegative: 66a7279a88dd, badhandsv5-neg: aa7651be154c”, Version: v1.6.0-2-g4afaaf8a
大模型我推荐用ghost加上YOZORA的vae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。
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’m chosen鲷, and this time I bring you Chocolate lora.
This is a style-based lora model trained on images exclusively from nai3. The reason I trained this model is because I wanted to upscale my images generated using nai3 with SD upscaling. However, I found that no matter which model I used for upscaling, it would alter the original image style. To overcome this, I trained this lora model with over a hundred images generated by nai3. Now, when I upscale my nai3 images with SD upscaling, as long as I use this lora model alongside a compatible large model like ghost, I can enlarge the images without changing their original style.
Of course, you can also use this lora model directly for generating images. The example images provided in the prompt were all generated using SD 1.5.
Lora doesn’t overfit, so I recommend a weight of 0.9. Trigger phrase: chocolate
Recommended tag:
Chocolate,
[Here you can write your desired content]
<lora:Chocolate-000018:0.8>,
Negative prompt: EasyNegative, badhandsv5-neg, Subtitles, word, Steps: 32, Sampler: DPM adaptive, CFG scale: 7, Seed: 611174500, Size: 1088×1984, Model hash: e3edb8a26f, Model: ghostmix_v20Bakedvae, VAE hash: ab33ad55c6, VAE: YOZORA.vae.pt, Denoising strength: 0.15, Clip skip: 2, ENSD: 31337, SD upscale overlap: 64, SD upscale upscaler: R-ESRGAN 4x+ Anime6B, Lora hashes: “Chocolate-000018: 544834e09a89”, TI hashes: “EasyNegative: 66a7279a88dd, badhandsv5-neg: aa7651be154c”, Version: v1.6.0-2-g4afaaf8a
For the large model, I recommend using ghost with YOZORA’s VAE for optimal results (recommended below). Other commonly used anime-style models should work as well.
Thank you for downloading and using it. If you like it, please feel free to follow me on Bilibili and Pixiv. My IDs are both chosen鲷. You can also check out my channel on Bilibili: https://space.bilibili.com/3867754?spm_id_from=333.1007.0.0, and my Pixiv profile: https://www.pixiv.net/users/40911695. Thank you.