EchoNai3-XL

319.14MB的文件是VAE。

1.
此模型的数据集由Nai3生成，我使用Nai3生成了近3000张图片，之后从中选取了400张左右高质量的同画风图片，然后基于AnimagineV3.1训练了一个Lora，之后我将这个训练好的lora融合进了AnimagineV3.1中，得到了现在的EchoNai3-XL。**此模型旨在借助Nai3某些高质量的画风，改善目前SD动漫大模型"AI味儿"过浓的问题。**此模型尚处试验阶段，远不是最终版本，之后我会把后续版本上传到这个页面，喜欢的可以关注一下这个模型，有任何问题也可以在评论区指出~
This model is built upon images generated by Nai3. I created nearly 3,000 images using Nai3 and then handpicked around 400 high-quality ones with a consistent art style. These were used to train a Lora based on AnimagineV3.1, which I then merged back into AnimagineV3.1 to create EchoNai3-XL. The goal is to reduce the 'AI traces' often found in current SD anime models by incorporating some Nai3's high-quality art styles. This is still a work in progress, not the final version. I'll post updates here, so feel free to follow along. Any feedback is welcome in the comments!

2.
它更适合创作背景简洁、风格唯美的肖像，而不是宏大复杂的场景。
相比制作宏大、绚丽的场景画，它更适合制作背景较为简单、风格偏唯美的肖像画

3.
请务必添加触发词"nai3_for_sdxl"
Don't forget to use the trigger word "nai3_for_sdxl"

4.
正面提示词模板：
Positive prompts template:

extremely detailed texture, hires, 8K, wallpaper, detailed, best quality, good quality, extremely detailed background, extremely detailed texture, amazing quality, very aesthetic, absurdres

负面提示词模板：
Negative prompts template:

nsfw, lowres, {bad}, error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract], bad anatomy, bad hands, @_@, mismatched pupils, heart-shaped pupils, glowing eyes

5.
生成参数建议
Recommended settings

DPM++ SDE(The best)/Euler a/DPM++ 2M

CFG~5 (CFG~7 for Euler a/DPM++ 2M)

1024x1024/832x1216/1216x832

6.
推荐使用Danbooru标签格式，而不是自然语言
Recommend to use Danbooru tags instead of natural language.