Rumiko Takahashi Style LoRA [PonySDXL]

First of all, a shout-out to the proflific user Lykon who created the excellent SD1.5 Rumiko Takahashi Style LoRA.

This is my attempt at a version for Pony, trained on a large high quality dataset that spans the artist's career and features multiple mediums. I've provided some guidelines and examples for getting the purest recreation of the style but I encourage you to experiment with different parameters, prompts and other LoRAs.

Guidelines:

• Requires the prompt "in ru_ta artstyle"

• Trials so far gave best results with a weight of 0.8-1

• The usual Pony prompts ("score_9, score_8_up", etc.) can be omitted. They don't give the purest reproduction of the artist's style. They may offer more stability to image generation but at the cost of dilluting style identity. In particular, "source_anime" will produce a much more generic anime image, albeit with a Takahashi flavor. Personal preference applies.

• 24% of the dataset is made up of beautiful, hires black and white manga panels. To get that style, use "monochrome, greyscale".

• The tags "retro artstyle, 1980s, 1990s" are NOT useful and I find they produce a squashed, caricaturised version of the style. "Watercolour" is a very helpful tag.

• CFG scale of 6 or 7 really allows the style to shine through. I've had good results with Euler a and DPM++ 2M so I can't speak for any sampler in particular. Have a go.

Limitations:

• This LoRA is focused on reproducing the artist's style and it has a limited knowledge of Takahashi's character names. However, it can produce very accurate likenesses of Inuyasha, Lum, et al, when prompted with specific character traits.

• Very occasionally produces borders, text, or chibi characters without being asked. But this happens so rarely that there's no need to safeguard against it with negative prompts.