https://discord.gg/CXKvxTSfkm is the link to discord, there are two channels, Chinese and English

在线生成网址 online generation link: (340) T-illunai3 - v4 | Stable Diffusion Model - Checkpoint | Tensor.Art (tusiart.com)

ps: For some unknown reasons, my model cannot run online on Civitai, even after multiple applications, it is useless. Therefore, I uploaded it to this platform synchronously, and online generation of the model can also be done on it

出于一些未知的原因，我的模型无法在civitai上在线运行，即使多次申请都无济于事，所以我同步上传了这个平台，在上面同样可以进行该模型的在线生成

QQ群：451450868（英灵殿）

929721518（被封的备用群

v1只是一个初步进行微调的版本，但是效果好的惊人,在动漫绘画的领域比pony的微调效果要好上不少，肢体的正确率也十分可观,希望各位能多多测试一下这个微调,

基于illustrious-xl Illustrious-XL - v0.1 | Stable Diffusion XL Checkpoint | Civitai

V1 just a preliminary version with fine-tuning, but the results are surprisingly good. In the field of anime painting, the fine-tuning effect is much better than Pony's, and the accuracy of body movements is also considerable. I hope everyone can test this fine-tuning and give some comments about it.

base on illustrious-xl Illustrious-XL - v0.1 | Stable Diffusion XL Checkpoint | Civitai

训练须知：

v1版本使用了100张经过筛选的高质量的使用nai3生成的图片，以及画风相近的一些公开素材作为姿势的补充，训练使用了prodigy算法进行训练,illustrious模型的女性素材很多，导致了模型会比较难以生成男性的图像，而且某些时候需要一些比较复杂的提示词才能生成目标图像，这将在之后的更多训练集的版本尝试进行一些改善,

Training instructions:

The v1 version used 100 high-quality images generated using Nai3 that had been filtered, as well as some publicly available materials with similar art styles as pose supplements. The training was conducted using the "prodigy" to train. The illustrious model has a lot of female materials, which makes it difficult for the model to generate male images, and sometimes requires some complex prompt words to generate target images. This will be improved in more training set versions in the future.

在v4中，我使用了600张由nai3生成的风格图片作为默认画风的训练，这一步是基于illustrious模型的，然后我用模型融合的方法在这一fine-tuning后的模型上加入了一些noob的模型成分，以来平衡illu的一些缺失的语义和识别prompt的能力,

In v4, I used 600 style images generated by nai3 as the default art style for training. This step was based on the illustrious model, and then I added some noob model components to this fine tuned model using model fusion to balance some of Illu's missing semantics and prompt recognition ability.

推荐使用质量词Recommend quality words

正面提示词

best quality,amazing quality,very aesthetic,absurdres,beast quality,masterpiece,artist:xxx

负面提示词

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry,4-koma, sketch, (bad:1.05), error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, abstract, (blurry, lowres, low quality:1.1), ugly, (big head, bad anatomy:1.05), artist name, (watermark:1.1)

Recommend using quality words

Positive prompt words

best quality,amazing quality,very aesthetic,absurdres,masterpiece,artist:xxx

Negative prompt words

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry,4-koma, sketch, (bad:1.05), error, fewer, extra, missing, worst quality, jpeg artifacts, bad quality, watermark, unfinished, displeasing, chromatic aberration, signature, extra digits, artistic error, username, scan, abstract, (blurry, lowres, low quality:1.1), ugly, (big head, bad anatomy:1.05), artist name, (watermark:1.1)

描述:

这个版本是基于noobai的early access版本进行训练的一个版本，使用了多样的经过筛选的522图进行额外画风以及脚部的训练，强化了部分noobai的画风问题，以及一部分画面过于扁平的问题，并且相较于v1，nai3的画风效果没有那么固化，因为训练集足够大能让我放进更多适合中和画风的内容，这个版本训练了200个epoch，在3090显卡上训练了36个小时

This version is based on the early access version of NOOBAI for training, using a variety of filtered 522 images for additional art style and foot training, strengthening some of NOOBAI's art style issues and some of the flat images. Compared to v1, the art style effect of Nai3 is not as rigid because the training set is large enough for me to include more content suitable for neutral art styles. This version was trained for 200 epochs and 36 hours on a 3090 graphics card

训练词语: