Arcueid Brunestud (Tsukihime Remake style)

A DoRA (like LoRA, but better?) of the best girl from Tsukihime, the White Princess of the True Ancestors, Arcueid Brunestud - in Remake style!

Recommended generation values:

A distilled CFG/flux guidance scale of 5 and a lora weight of 0.8 is a good starting point, but I have found it to be quite flexible. Feel free to experiment with larger ranges like 0.6-1.1 lora weight or 2.5-10 distilled CFG. Results can change a lot per prompt and seed.

Trigger words:

You may want to have these keywords in the majority of your generations (you can change them of course):

Arcueid, 1girl, solo, short hair, bangs, blonde hair, red eyes, slit pupils, ahoge, hair intakes, antenna hair

In addition to these keywords, add the following if you want a particular trained outfit:

• Normal outfit (white turtleneck sweater and short blue skirt)

normalarc, long sleeves, blue miniskirt, black pantyhose, brown knee boats, necklace, white sweater, turtleneck, turtleneck sweater, silvery necklace

• Old outfit (purple skirt)

purplearc, long sleeves, long purple skirt, black pantyhose, black high heels, necklace, white sweater, turtleneck, turtleneck sweater, silvery necklace

• Date outfit (white coat with big black bow)

coatarc, long sleeves, white coat, white fur capelet, big black bow, black buttons, black thighhighs boots, necklace, turtleneck, hair intakes, silvery necklace

• Clad in glaciers/Anti-heat dress

glacierarc, blue flower, hair ornament, white dress with black edges, white elbow gloves, cleavage, halterneck, bare shoulders, detached sleeves, grey pantyhose, black footwear, high heels

After those, you should add in natural language the rest of the prompt. Do not include parts of outfits that you do not want to appear in the image (i.e don't prompt 'high heels' if you want a portrait, obv). Again, feel free to change them and experiment (or use completely different outfits altogether). Archetype-Earth is not included sadly, might do a specific lora/dora in the future

The sample images were generated using flux1-dev-Q4_0 and t5-v1_1-xxl-encoder-Q5_K_M

Hope you like it!