V3 Instruction

Here's a new version of this model. I found that the model I had updated, V2, was veering a bit too far from the style I intended to showcase with this model. Therefore, I reworked this model and achieved a satisfactory result.

Here are the recommended settings:

Sample: DPM++ 2M Karras

Steps: 30-40 (going below 30 to 20 is feasible, but I personally found that the images looked better between 30 and 40)

VAE: vae-ft-ema-560000-ema-pruned.ckpt (this is the one I used, but experimenting with another one might potentially yield improved results)

Upscaler: 4x-Ultrasharp, as it allows for preserving the desired style in the images. R-ESRGAN 4x+ Anime6B could work, but I feel it slightly deviates from the intended style, though it still remains good overall. It will depend on your preferences.

INTRODUCTION

I created this model by merging a model that I had already been created that I had a while ago but never shared since it didn't have anything special by combining several existing Anime models . So, I merged my Lora with this model called Nier Character Style, and I found the images generated by this model interesting and quite different from what I've seen before. Perhaps this model resembles other models, but it was not my intention to plagiarize anyone.

INSTRUCTIONS:

Let me explain how to use this model; I apologize if I omit some information as I am a beginner in model creation.

This model allows you to generate anime images based on the style of the Nier series illustrations. I created this model out of curiosity to see the result of merging a model with a Lora. I was happy with the result, so I decided to share it with you.

I made this model just for fun, and I hope you will enjoy it!

I tested this model with different parameters:

I used this model with the following VAEs:

• vae-ft-ema-560000-ema-pruned

• kl-f8-anime2

I didn't notice any significant differences in the results.

I also tried using the following samplers:

• Euler a

• DPM++ 2M Karras

• DPM++ SDE Karras

All of these samplers can give you satisfying results; I didn't observe one sampler being superior to the others. Personally, I use the DPM++ 2M Karras sampler for all my image generations.

I recommend using the ADetailer extension to save time and achieve more successful facial results.

As for upscaling, you can use the R-ESRGAN 4x+ Anime6B model.

Since I merged my Lora, you may use the keyword 'nier_artstyle' if you find it enhances the results, but it is not mandatory

Regarding the steps, I would recommend setting it between 30 and 40. However, you can still achieve satisfactory results below this range: