Neon (Vision Eye) from Nikke Goddess of Victory
Aiming for highest details as accurate as possible while maintaining enough flexibility and versality within Anima capabilities.
Strength: 0.6 to 1
How to use: Prompt in natural language then include the desired tags.
Default Outfit:
neon (\vision eye\) (\nikke\), 1girl, solo, long hair, breasts, looking at viewer, shirt, skirt, long sleeves, hair ornament, gloves, hat, ribbon, jewelry, medium breasts, standing, green eyes, jacket, white shirt, full body, white hair, earrings, pleated skirt, parted lips, boots, black gloves, choker, glasses, midriff, black skirt, miniskirt, fingerless gloves, black footwear, coat, black jacket, high heels, crop top, hand on own hip, black headwear, bare legs, black choker, peaked cap, semi-rimless eyewear, black coat, red-framed eyewear, jacket on shoulders, crop top overhang, over-rim eyewear, facing left, "counters" medal, white sailor collar, high-waist skirt, chest strap, ribbon bar, black ribbon on left side of chest strap, white background,
No Jacket:
neon (\vision eye\) (\nikke\), 1girl, solo, long hair, breasts, looking at viewer, smile, open mouth, simple background, skirt, long sleeves, gloves, hat, bow, ribbon, jewelry, medium breasts, standing, green eyes, white shirt, full body, earrings, pleated skirt, boots, black gloves, choker, glasses, medium hair, black skirt, miniskirt, fingerless gloves, black footwear, high heels, crop top, hand on own hip, black headwear, bare legs, black choker, peaked cap, high-waist skirt, red-framed eyewear, crop top overhang, over-rim eyewear, facing right, white sailor collar, white background,
Description
Anima Base V1
Comments (4)
💕👍🏻
This is a really good lora. Any chance you could share like, how many images and basic training setup? I want to try training one XD but...most of the anima lora are not very good tbh. This is pretty dead on. I assume you use klein/qwen edit to make additional angles and images?
Thanks!
GPU Nvidia RTX 5090
Trainer: Kohya_SS SD Scripts
Source Config:
https://civitai.red/articles/27484/anima-lora-preset
Dataset is most important. Varies depending on subject matter.
This Dataset:
Danbooru Tagger
https://huggingface.co/SmilingWolf/wd-eva02-large-tagger-v3: this achieved best score using threshold 0.5296
(VRAM use max out at 31.4gb, reduce Batch size or resolution if needed)
1344x1344
Batch Size:2
10 Repeat Synthetic data (Nanobanana2 or GPT-Image 2)
2 Repeat on reference material
2 Repeat on miscellaneous
Repeats are balanced in the way to make the image respond well. Source material amount is 10x the amount of synthetic for example. It's up to you to balance your datasets.
Cool thank you, will look into it. Lovely lora!