Illustrious:

I used these settings:

sampler: DPM++ 2M Karras;

cfg: 4.0;

steps: 25.

Huge thanks to @LunarSunlite13 for the advice and generous donation for future experiments!

Anima:

The LoRA works well on close-up shots and portraits. The dataset contains a lot of images with warm tones and extra details due to the artist's style, so I recommend reducing the LoRA weight to 0.8 if needed.

I still haven't managed to exactly replicate that same style. I'm posting the result from epoch 8 out of 10 here, as it turned out to be the most stable. The 10th epoch reproduces the style well, but unfortunately it has overfitted — it adds extra details and gives the image a yellowish tint. Next time I'll refine the dataset with better color correction.

No trigger word is required. Possibly in the future I'll create a LoRA on the KREA2 model, but until then I'll take a break... I spent quite a lot of time creating this LoRA.