Yoimiya. The dress is difficult to make appear correctly on entirely realistic models. It works best on semi-realistic and anime models, because the dataset is in anime format. All images are generated with A1111 on PC. Work better with simple negative prompt (3 embeddings) and detailed positive prompt. Tested with my checkpoint Realistic2epic. For SD, not for Pony or XL.
Weight 0.60 - 0.80
If you like the character, always leave a rating and a picture, and if you give some buzz you will allow me to train other LORAs or MODELS. Downloading only is not useful. For model visibility, remember to rate the images in my gallery. Thank you.
Negative prompt : (CyberRealistic_Negative-neg), (worst quality:2), (low quality:2), (bad-hands-5, lowres, bad anatomy, deformed, missing fingers, extra digit, fewer digits, cropped,watermark,username),(blurry:1.4) ,easynegative, unreal engine, (nipples, navel, nsfw)
Delete ((nipples, navel, nsfw)) in negative prompt for NSFW and add them in positive prompt.
Positive prompt : LORA, realistic, masterpiece, detailed, 8k, beautiful photography of yoimiya \(genshin impact\), 1girl, blonde hair, bandage, collar, bandage arm,hair accessory, Japanese clothes, bangs, blue eyes, kimono, sarashi, bandaged leg, bandaged arm, bare shoulders, massive cleavage, arm tattoo, (orange kimono:1.2), floreal print, big breasts, upper body shot, upper body shot...
Add for better pic : realistic skin detail, hourglass body shape, photorealistic,...
Use suggested resources for perfect results. Noise weight 0.2, detail 0.6, with cyberrealistic neg in negative prompt.
Recommended for a better result: ADETAILER mediapipe face mesh with impaint 0.55
Setting : Sampling DPM Karras, HiresFix on, ITF-skinDiffDetail-Lite at 2x, Denoising 0.38, Upscale 2x, Sampling steps 35+, Resolution 512x768, Upscale 2x ITF-skinDiffDetail-Lite next passage.