The 319.14MB file is a VAE.

319.14MB的文件是VAE

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!

It’s better suited for creating aesthetically pleasing portraits with simple backgrounds, rather than for producing grand and elaborate scenes.

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

请务必添加触发词"nai3_for_sdxl"

Don't forget to use the trigger word "nai3_for_sdxl"

因为此模型基于AnimagineV3.1，所以推荐与AnimagineV3.1使用同样的正面提示词和负面提示词

This model is based on AnimagineV3.1, so I recommend using the same positive and negative prompts:

正面提示词模板：

Positive prompt template:

masterpiece, best quality, very aesthetic, absurdres

负面提示词模板：

Negative prompt template:

nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]

生成参数建议： Recommended settings:

DPM++ SDE/DPM++ 2M/Euler + Normal

CFG~5

1024x1024/832x1216/1216x832

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.

由于我是用comfy ui生成的示例图片，所以tensor art可能无法解析图片的元数据。下面是一套示例模板，仿照这套模板编写提示词就可以
Since the example images are generated using comfy UI, tensor art might not be able to parse the image metadata. Here is a sample template, you can follow this template to write your prompts

Positive prompts:

(masterpiece, best quality, very aesthetic, absurdres), ((nai3_for_sdxl)), 1girl, demon horns, power (chainsaw man), solo, impasto, oil painting, best quality, white background, simple background, upper body, red horns, hair between eyes, cross-shaped pupils, long hair, blonde hair, pink hair, small breasts, medium breasts, collarbone, sports bra, bare shoulders, shadow, crop top, midriff, bare arms

Negative prompts:

nsfw, (happy, joyful, smile, laughing, looking at viewer), lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]

EchoNai3-XL - v1.0