Yara Flor from something unlimited by SunsetRiders7 and Gunsmokegames. If you interested in his work check out their Patreon and discord, links are below. I just made this kinda for fun to see how it would look like.
Using the strength of 1 will give you problems if you are trying to prompt the character in different clothing. I recommend using the strength of 0.8 as it keeps the character looking almost if not the same style and look. Preview is shown below.
Anima
Trigger words: @suyaraflor
optional - black hair, long hair, bangs, brown eyes, mole under eye
Heroine Outfit: - tiara, leotard, highleg leotard, two-tone leotard, breastplate, gold armor, cleavage cutout, bracer, thigh strap, gold belt, knee boots, gold boots
Recommended strength: 0.85
Illustrious
Trigger words: Yara Flor, 1girl, tiara, black hair, brown eyes, dark-skinned female, lips, long hair
optional - "mole under eye", I didn't train it well to keep it on the same side so I usually just leave it out of the positive and put mole in the negative
Heroine Outfit: - bracer, leotard, cutout, thigh strap,
I used WAI-NSFW-illustrious-SDXL as my checkpoint to generate these images.
https://civarchive.com/models/827184?modelVersionId=1283437
Pateron
https://www.patreon.com/gunsmokegames
Discord
http://discord.gg/38pEx3d
You are able to play the public download.
Upscaler
I used R-ESRGAN 4x+
Adetailer
https://civarchive.com/models/207406/adetailer-face-finder-furry-anime-realistic-ddetailer
Lora Strength Preview
Keeping the strength at 1 wanted to try to keep some of the character heroine costume if you want more flexibility without changing too much of the character style of drawing. I recommend using anywhere from 0.7 - 0.9. I personally like 0.8.
Description
Yara Flor (Something Unlimited), Anima version
Yara Flor in the Something Unlimited art style, trained on Anima Base v1.0, so it works on any Anima checkpoint. Previews were made on dustVision Anima.
Trigger: @suyaraflor
Her hair, eyes, mole, skin and lips come with the trigger. For the most reliable likeness, add:black hair, long hair, bangs, brown eyes, mole under eye
Main costume: tiara, leotard, highleg leotard, two-tone leotard, breastplate, gold armor, cleavage cutout, bracer, thigh strap, gold belt, knee boots, gold boots
Every costume piece was tagged in training, so you can swap it for any outfit or leave it out.
Strength
0.85: best for most scenes (preview 20 compares strengths).
1.0: strongest likeness; best for close-ups and the costume.
0.7: use when the outfit must come out exactly as prompted (her hair can come out shorter, so add
long hair).
Tips
She has muscles and abs like in the SU art. They're tagged, so put
muscular female, absin the negative to slim her down.In back views or small-face shots, keep the five feature tags above in the prompt so her hair and eyes don't drift.
She's already in the SU style, so no style LoRA is needed. Stacking Something Unlimited Style v3.0 at 1.0 + this LoRA at 0.85 also works (preview 19) and gives stronger SU lips and eyeliner.
You can change her hair and eyes for cosplay (previews 17 and 18).
Settings used for previews: dustVisionAnima v1.0, Euler a, Normal, 28 steps, CFG 4, 832x1216 / 1216x832, quality tags masterpiece, best quality, score_9, score_8, no Hires Fix.
Training (Kohya sd-scripts via kohya_ss v26, local)
Data: 32 images, all cropped from the two official SU Yara Flor sheets: 8 figures, each at 4 zoom levels (thigh-up, upper body, portrait, face). There's very little source art, so cropping was used to give the trainer more to learn from.
Network: LoRA, dim 32 / alpha 16, network dropout 0.1.
Optimizer: LR 1e-4, AdamW8bit, cosine.
Length: 5 repeats × 10 epochs = 1,600 steps, batch 1. Epoch 8 released (1,280 steps).
Timesteps and noise: sigmoid timestep sampling, noise offset 0.1.
Captions: shuffled, keep tokens 1, tag dropout 0.1.
Resolution: 1024 with buckets, bf16.