<h2 id="heading-146">Imitation model. Modelled after Japanese AV actress and onlyfans model Mary Tachibana (橘メアリー).</h2>Mary came a long way, having debuted almost 10 years ago in 2014 - but with a totally different level of popularity. The actress got married back in 2017, and not long after joined the idol group BLACK DIAMOND (jp) and started her DJ career (DJ Mary). However, unlike JAV artists like Sora Aoi or Yuka, she remained active in the JAV industry and came back with a much more tonned body post-COVID (posting workout selfies here and there), reduced significant amount of weight, and now dominates the NTR, mature lady genre of JAV. This LoRA focuses on the latter part of the more refined and slightly slender body physique. It almost felt like she got younger and hotter. (She is likely one of my fav JAV artist moving into 2024 &lt;3 ) I almost wanted to just keep this model to myself, but sharing is &lt;3<h3 id="heading-108">~~~~~~~~~~~~~~~~~~~~~~~~~~ Recommended LoRA strength: 0.7 (0.5-1.0 are plausible range dependent on checkpoints; higher values creates a sharper facial feature and increases bust size)</h3><h3 id="heading-109"></h3><h2 id="heading-110">Main Prompt:</h2><h3 id="heading-147">marytachibana_jav, breasts, brown eyes, brown hair, jewelry, long hair</h3>variations: black eyes, medium hair, tied hair, hair bun, ponytail, earringsfacial variations: parted lips, lipstick, lip gloss, blush, thick makeup, smile, teeth*Use of facial prompts depend on checkpoints, but you can experiment freely with this LoRA, the flexibility is much stronger than any other LoRAs I made.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Checkpoints that work well for this LoRA:1) Realisian. This time, Realisian is superior. Other models dont even come close. 2) RealisticMix and AsianBRM_BeautyRealMix. Go with RealisticMix if you want a more mature and developed look, go with BRM if you want a slightly younger look. Shall separate the examples in preview to a separate uplaod for better comparison. 3) UltimateHentai Realistic. Similar to RealisticMix, but a less refined version of its render. oh and worse BadHands problems compared to the 3 models above. 4) PerfectWorld. Technically not terrible, and gets the features correct. But its too 2.5D to my liking, and unrealistic Oppais in &gt;1/2 generations.5) Other models. I randomly tried 4 other asian realistic models. Differences not too drastic. Can probably get it to work with abit more prompting and weigth adjustment. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~Note for achieving good resemblance:To achieve good resemblance, you dont really need to do much accept prompting well. (yes, its a good, and almost brain-dead LoRA). complexity 1/5 stars.If you do struggle to get the LoRA to work properly, perhaps lower the LoRA weight slightly in the main prompt, that will avoid distorted body figures (control it somehwere between 0.65-0.8) and use the stronger weights in .If you increase the AD denoise strength &gt;0.5, you run into the risk of getting eye bags and wrinkles. Its not entirely a bad thing, if you are into that photo-realism (Mary is technically 30 this year, so oh well..). My settings for the Adetailer prompts were:face of marytachibana_jav, brown eyes, brown hair, smile, parted lips, ponytail, &lt;lora:JAV_mary_tachibana_v1_epoch_10:0.8&gt; with Adetailer Strength 0.4.If you want her look without the ponytail, just remove ponytail with "long hair" or just leave it blank. 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 wont do too much harm, but just stick to this to avoid badImages), and turn off Adetailer.~~~~~~~~~~~~~~~~~~~~~~~~~~~~Issues:BadHands very occasionally. If I want to nittpick, then the 0ppai consistency is not very high, esp with clothing prompts. Overall, nothing much to note. Post-production edits/Reproducibility:2 pictures, on bad hands. Happy prompting. And do share [SFW] creations~

JAV Mary Tachibana

v1. Initial Release. Trained using a whooping ~150 images, incl. unusual angles, predominantly focused on her post 2018 appearances. Epoch #10.

Files with SHA256: c543978aa4589a756798582b96c7fa84b4062cd778d2b9b25a4b1e5af2c1f7dc