Imitation model. Modelled after retired Japanese AV star Erika Kitagawa (北川 エリカ).
Retired back around 2019, this actress gained popularity in the Gyaru AV genre, BDSM/femdom and also traditional NTR genres. Feature-wise she is classified within the "Bijin" (beauty class) category, with a rather toned body (abs, refined biceps etc).
~~~~~~~~~~~~~~~~~~~~~~~~~~
Recommended LoRA strength: 0.8 (0.7-1.0 are plausible range dependent on checkpoints; higher values >1 lead to elongated faces, which is not very ideal)
Main Prompt:
erikakitagawa_jav, breasts, long hair, brown hair, brown eyes,
variations: tied hair
facial variations: natural makeup, teeth, parted lips, lipstick, eyeshadow, mascara
*Use of facial prompts depend on checkpoints, use it if your checkpoint have a bias towards minimal makeup
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Checkpoints that work well for this LoRA:
1) Realisian. Best when it comes to resemblance.
2) RealisticMix. Better than other models.
3) PerfectWorld, UltimateHentai Realistic or other photoRealistic Hentai models (European model works well for this LoRA). This time, PerfectWorld does very well, again.
4) ClearRealismMix-Eur, or asian-based/eurasian checkpoints. Not perfect, but still better than pure-European checkpoints.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Note for achieving good resemblance:
Use Adetailer for baseline generation (at 512, 768 etc) only, try to control LoRA strength at 0.9-1.0, adjust your denoise accordingly. For this LoRA, recommended to use Adetailer at ~0.4 denoise strength for close-up, and 0.45 to 0.55 for full body.
For this LoRA, you will need facial Adetailer. My settings were:
"face of erikakitagawa_jav, brown eyes, natural makeup <lora:JAV_erika_kitagawa_v1-000005:1.0>"
Recommended approach: is to not use HiRes, but upscale it with img2img instead.
Use a denoise strength of 0.25 -0.3 (anything higher will overcorrect face during the process), and turn off Adetailer.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Issues:
The hardest part is to get good resemblance of her face in this LoRA, without getting very elongated faces. I tried few things, changing makeup prompts, moving things back and forth facial Adetailer prompt. But found it, actually, the easiest way is to check what you have in your negative prompt - so take the advice from these checkpoint makers:
do not go overboard with negative prompts and negative embedding. In my case, removing some excessive BadImage, or "young" type of prompt, actually fixed the problem. So do check on those, if you struggle to get something good in your checkpoint.
in img2img upscaling, sometimes the shape of her eyes can be alittle off - I guess you can try running Adetailer again with very low denoise to try and fix that.
Post-production edits/Reproducibility:
Actually none. No inpainting was needed. Everything fresh from oven this time.
Happy prompting. And do share [SFW] creations~
Description
v1. Trained on 71 images. Release at Epoch #5. This is one of the very few model that experience overbaking relatively early in the training process, so took me some time to decide which Epoch is stable for use.
More testing on whether later epoches can be better after engineering the right "negative prompts" to use/exclude.
Else, I might have to retrain the model, after throwing of some suspicious outliers involving wet hair, and smeared makeup to redo this again..