JAV Yuka_Tachibana (橘優花)版本JAV_yuka_tachibana_v1 (ID: 456474)

Imitation model. Modelled after retired(?) Japanese AV actress Yuka Tachibana (橘優花; alias Airi Hinata 日向爱理).

Now back to big 0ppai JAV celebrities.

Yuka's work in the JAV industry spanned from 2013 - 2016, she announced her retirement in the beginning of 2017, but eventually came back with an alias under Airi Hinata. It wasn't clear since then whether she remains active in the industry, or had moved on with her career. Within the JAV fanbase, she hasn't been among the favourites despite a splendid figure - many has commented on her acting been rather stiff (abit like Anzai's case), and her expression is fairly - standard. Still, I think her features and vibes are probably quite unique, to be considered for a celebrity LoRA.

~~~~~~~~~~~~~~~~~~~~~~~~~~
Recommended LoRA strength: 0.8 (0.6-1.0 are plausible range dependent on checkpoints; higher values >1 dont do much, apart from some background-bleeding; fairly stable LoRA in this case but dont go overboard with this..)

Main Prompt:

yukatachibana_jav, breasts, long hair, parted lips, teeth,

variations: single hair bun, high ponytail, tied hair,

facial variations: parted lips, lipstick, makeup, blush, teeth, smile

*Use of facial prompts depend on checkpoints, I didnt actually bother, since the main prompts works fine.

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

Checkpoints that work well for this LoRA:

1) Realisian. Not perfect, but currently the closest I can get (esp for her tied hair apperances)

2) RealisticMix. Has comparable performance with Realisian in facial resemblance. But my major issue with this is the exaggerated breast size, but thats ok, she is almost that type.

3) Other realistic checkpoints. Tried a random set of 5. Does OK. require more negative prompting adjustments to remove "ugly" flags.

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

Note for achieving good resemblance:

Very straightforward to use. If you are lazy, you can just use the main prompt to control for everything and apply Adetailer to fix the faces.

Else, you can separate the prompts into 2 stages.

1. Generate the main prompt using a LoRA strength of 0.6-0.8. I did not run into any weird body proportion issues with 0.8. so its probably safe to say that this weight is stable for use.

2. Now depending on what you use for (1), turn on facial Adetailer, and put something like

"face of yukatachibana_jav, long hair, parted lips, teeth, <lora:JAV_yuka_tachibana_v1_epoch_10:0.9>"

Do remember that if you want her with her ponytail, make sure you set the prompts on both ends to reflect that. As you can see, I tend to set the LoRA weight slightly higher for Adetailer, to correct her facial features.

set Adetailer denoise strength to 0.45-0.5. for Full body, you can try to set this higher to 0.55.

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 0.20 -0.25 (anything higher will overcorrect face during the process), and turn off Adetailer. I prefer 0.2 since I use foolhardy-reMarci.

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

Issues:

Nothing too noticeable, apart from NSFW checkpoints getting too excited with "breasts". Easy problem to fix with negative prompts and re-seeds.

Post-production edits/Reproducibility:

Note that these are native 512x768 images. I forgot where I left the highres images after migrating folders... But the prompts are embedded in the images, feel free to scale them yourself.

Happy prompting. And do share [SFW] creations~

描述:

v1. Initial Release. Trained on 182 images. Epoch 10 chosen due to consistency on "from back" and "from side"

训练词语: yukatachibana_jav,breasts,long hair,parted lips,high ponytail