JAV Reiko_Kobayakawa

Imitation model. Modelled after Japanese model and AV star Reiko Kobayakawa (小早川怜子).

Almost impossible for those that are into more mature,milf-themed JAV to not know this one. Arguably one of the high networth AV star in the industry as well. I am surprise there isnt already a rendition of her in civit. Figure wise, slightly taller than average (5'2) with a slim yet composed body figure. So here it is.

~~~~~~~~~~~~~~~~~~~~~~~~~~
Recommended LoRA strength: 0.8 (0.6-1.0 are plausible range dependent on checkpoints; higher values >1 lead to aged facial features and increase 0ppai size unintentionally)

Main Prompt:

reikokobayakawa_jav, breasts, long hair, eyeliner, perky breasts

variations: tied hair

facial variations: mature makeup, parted lips, lipstick, eyeshadow,

*Use of facial prompts depend on checkpoints, use it if your checkpoint have a bias towards minimal makeup, OR dont use it if your checkpoint over-react with "mature makeup" tags

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Checkpoints that work well for this LoRA:

1) Realisian or RealisticMix. Both checkpoints have fairly similar performance, with Realisian controlling her biological age better.

2) PerfectWorld, UltimateHentai Realistic or other photoRealistic Hentai models Works adequately well with 0.6-0.8 LoRA weight, assuming you use the right makeup prompts (see below).

3) MagicMix realistic. Not perfect, as it tend to make her look much younger and very korean-uzzlang like, but can sometimes give a better image (Reiko used to be young+hot ;) ) overall.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Note for achieving good resemblance:

This LoRA requires some synergy with makeup prompts, to make it work. Because some of her training imgaes included pictures of her without much makeup (and I think moving forward thats much better for any integration into checkpoint than letting the activation tag soak it all up), you will need to manually prompt the makeup back for a realistic representation of her face.

For this, you need to check just 3 very simple things:

1) Her eyeliner

2) Her eyeshadow

3) Lipstick

For image generation, I tend to double dip on eyeshadow, as the generation can vary based on whether ornot "eyeshadow" is prompted in the main prompt.

So a good example prompt will be something like:
"1girl, reikokobayakawa_jav, (long hair), eyeliner, eyeshadow, lipstick:0.3, milf, (mature), medium breasts, <lora:JAV reiko_kobayakawa v1_epoch_9:0.8>"

accompanied by a facial AD prompt:

"face of reikokobayakawa_jav, (mature woman), makeup, eyeshadow, eyeliner, lipstick, <lora:JAV reiko_kobayakawa v1_epoch_9:0.8>"

Take note of the "mature woman" tag as some checkpoints are very sensitive towards age-scaling, and can make her look way too "mature" (aka. old).

From here, you can control the makeup intensity purely from facial AD, set at ~0.4-0.45 denoise, without worrying about facialAD not blending well with the main image assuming you do not change the size of the AD window.

It is still better to use AD at baseline generation (512,768 etc) and then upscale it later.

I would also suggest using img2img instead of HiRes. But thats personal preference.

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:

Nothing that needs immediate fixing, badHands, and occasional bad hand-body ratio. Just re-seed.

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:

Minor inpainting on some Badhands.

Happy prompting. And do share [SFW] creations~