JAV Azusa_Tani版本JAV_azusa_tani_v1 (ID: 415846)

Imitation model. Modelled after Japanese AV star Azusa Tani (谷あづさ , previously a.k.a Eren Fujisaki 藤咲エレン).

This actress first debuted under the alias of Eren Fujisaki back in 2016, and kept that alias for 2 years til 2018. Filmography wise, she did decent amount of films under both aliases, with her newcomer debut under Azusa Tani with E-BODY being one of the beslseller on DMM for quite a while back in 2018-2019. There's something to that unique facial feature of her, that she gives off quite a unique stare with the japanese crooked teeth, making her very recognisable despite the Alias change. Rather than a full-out mature and tonned body, Azusa's body built is alittle bit more relatable and average, still a H-cup though, but petite and with a less puffed-up ass (unlike Reiko K. or Hikari Sena), and I was quite surprise that the LoRA picked that up without any additional prompting. Now lets get to the fun part.

~~~~~~~~~~~~~~~~~~~~~~~~~~
Recommended LoRA strength: 0.8 (0.6-1.0 are plausible range dependent on checkpoints; higher values changes the shape of the face and increase the sharpness of the eyes..; I recommend just sticking around 0.8 for most cases)

Main Prompt:

azusatani_jav, breasts, brown eyes, black hair, brown hair, bangs

variations: black eyes, medium hair, tied hair, hair bun, ponytail

facial variations: parted lips, lipstick, blush, teeth

*Use of facial prompts depend on checkpoints, use it if your checkpoint have a bias towards minimal makeup, most of the time you wont need any of these

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

Checkpoints that work well for this LoRA:

1) Realisian. Almost perfect. No doubt, check the previews.

2) RealisticMix. Slightly trading accuracy for more refined facial features. Not a fan of those corrections, since my aim is to preserve her "appearance".

3) PerfectWorld and UltimateHentai Hit and miss on certain generations. and bad 0ppai size projection (too big tbh)

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

Note for achieving good resemblance:

This LoRA is easy when it comes to reproducing good resemblance, since Azusa's facial feature and body shape is unique enough for SD to render them without too much problem.

1. Generate the main prompt using a LoRA strength of 0.8, try to control for similar weight on full body modes.

2. Just turn on face Adetailer, without any additional prompt.

So a typical prompt for your main prompt would be something like:
"1girl, azusatani_jav, breasts, brown eyes, black hair, brown hair, bangs <lora:JAV_azusa_tani_v1_epoch_10:0.8> looking at viewer, shot using canon DSLR, cowboy shot, tshirt, detailed body, attractive body, perfect human body, realistic face"

If you want to add makeup promtps you can do it there. And set Adetailer denoise strength to 0.4. Thats all. Easy right?

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

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:

BadHands here and there. Nothing major. bad head to body ratio for some generations using other checkpoints, but nothing major with the recommended checkpoints.

Post-production edits/Reproducibility:

None

Happy prompting. And do share [SFW] creations~

描述:

v1. Release at Epoch #10. No difference between 9,10. some difference between 8,10. But 8 dont work well for some checkpoints, so 10 selected as best global epoch.

训练词语: azusatani_jav,breasts,black hair,brown hair,bangs,ponytail