NMN793_v1.1 is a fine-tuned model based on Chenkin Noob XL (CKXL), aiming to address two main problems I encountered with CKXL: poorly rendered hands and a somewhat rough image quality.

However, the real solution to these problems will likely come with the official release of the model.

NMN793_v1.1 is not a universally compatible model; it inherits the poor LoRA compatibility issues of CKXL. However, since CKXL has been trained on a considerable number of artists' works and has a significant learning capacity, combining it with artist strings can generate many possibilities, which is quite fun! Of course, NMN793_v1.1's understanding of artistic styles differs somewhat from CKXL, but I don't think this affects its playability.

Usage:

Similar to other Illustrious models, but CLIP Set Last Layer must be 2. If you are using ComfyUI, it will recognize itself and does not require CLIPSetLastLayer nodes.

You can choose between quality words and negative words at your own discretion.

Positive quality words: masterpiece, best quality, high resolution, aesthetic, excellent, newest

Negative quality words: low resolution, worst quality, bad quality, e621, furry

The quality terms, negative terms, and resolution can all be handled according to NAI's guidelines; they're all pretty similar. You can check the example images.

Chenkin Noob XL (CKXL) has a new tag system; those interested can check out its page.

Here are my parameters (you can definitely try other parameters, especially different samplers):

+ Sampler: euler a or dpmpp 2m

+ Schedule: Beta

+ Steps: 20~25

+ CFG: 5~6

You can freely share this model, but you need to indicate the source and bear all the consequences caused by it yourself!