<h2 id="heading-146">[Requested] Imitation model. Modelled after Japanese AV star, influencer and youtube celebrity Konan Koyoi (小宵こなん).</h2>This JAV actress shouldn't be unfamiliar to many. Debuting only back in 2021, Konan is one of S1's top performer for quite some time last year, and gained huge popularity in the JAV and general audience. She havn't gotten the luxury of time, but S1 had been pushing it, averaging about 16 films per year (as of 2023), and more to come. Physique wise, slightly mature and developed body, alittle bit of thicc, and a very cute idol face. PS Note:I usually don't rework on JAV LoRAs that have been previously published by others, since I think they would be fairly similar (source images, gravure photoshoots, wallpapers, insta-grab). But I figured maybe I will do this one (and see how it goes), and just for self-learning purposes, compare and see how this revised LoRA work relative to the other published models. This LoRA is trained using 327 images - from close-up face-only shots, to full body. Supports both landscape and portrait generations.LoRA requested by Riokotaro, author for RealisticMix (<a target="_blank" rel="ugc" href="https://civarchive.com/models/272697/realisticmix">RealisticMix - II v1 | Stable Diffusion Checkpoint | Civitai</a>) one of my most used checkpoint for testing JAV LoRAs :D <h3 id="heading-108">~~~~~~~~~~~~~~~~~~~~~~~~~~ Recommended LoRA strength: 0.8 (0.6-1.0 are plausible range dependent on checkpoints; higher values changes the shape of the face and elongate the face..; I recommend just sticking around 0.8 for most cases)</h3><h3 id="heading-109"></h3><h2 id="heading-110">Main Prompt:</h2><h3 id="heading-147">konankoyoi_jav, breasts, long hair, brown eyes, bangs, </h3>[special] variations: medium hair, tied hair, hair bun, ponytail, double buns, hair bow, hair ornamentfacial variations: parted lips, lipstick, makeup, blush, teeth*Use of facial prompts depend on checkpoints, use it if your checkpoint have a bias towards minimal makeup. For Konan, you dont need this, really.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~Checkpoints that work well for this LoRA:1) Realisian. Superior for this. 2) RealisticMix. Slightly more exaggerated sexyness. slight large 0ppai bias. 3) All other models. I did not try this extensively, only PerfectWorld and MagicMix Realistic. Works fine, body ratios are decent, but if you are looking for high resemblence, the models on top are better. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~Note for achieving good resemblance:Because of the ridiculously large training set - This LoRA is easy when it comes to reproducing good resemblance. You can literally just use the same main prompt for ADetailer. 1. Generate the main prompt using a LoRA strength of 0.8, try to control for similar weight or slightly decrease the 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, konankoyoi_jav, breasts, long hair, brown eyes, bangs, &lt;lora:JAV_konan_koyoi_v1_epoch_4:0.8&gt; , looking at viewer, shot using canon DSLR, turtleneck shirt, skirt, highwaist, cowboy shot,..." If you want to add makeup promtps you can do it there. And set Adetailer denoise strength to 0.4. 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 0.25 -0.3 (anything higher will overcorrect face during the process), and turn off Adetailer.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<h3 id="heading-446">(Additional) Differences with other existing Konan Koyoi LoRAs:</h3>Guess some of you might be interested to see which LoRA to use, and for what purposes (face merging, LoRA mixing, body resemblence etc). So I have prepared a grid image to show some quick comparisons. Compared to the previous models (averaging ~60-120 images per model), this model uses 320+ images with minimum pixels 768 (scaled to 768 max). TBH, all the models are equally impressive, and does the job - it falls down to personal preference, and what you want to achieve with this LoRA. (see preview for the grid comparison) When to use this model:1) Since I trained my models using WD14, you can get better control over customized prompting. If you want to access variations of hairstyle, postures, realistic expressions, use this model. 2) If you want high realism, and resemblence (even if that means her body is less refined, and less beautified), use this model.3) If you want to use this model for face-merging, and have better control over Konan's individual feature, use this model. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/9af072c2-20e7-483e-9f1e-608178ee1c27/width=525/9af072c2-20e7-483e-9f1e-608178ee1c27.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ec1f896b-4010-46f5-890d-699182fe939c/width=525/ec1f896b-4010-46f5-890d-699182fe939c.jpeg" />###Generic prompt: 1girl, konankoyoi_jav, breasts, long hair, brown eyes, bangs, &lt;lora:JAV_konan_koyoi_v1_epoch_4:0.8&gt; , looking at viewer, shot using canon DSLR, turtleneck shirt, skirt, highwaist, cowboy shot, detailed body, attractive body, perfect human body, realistic face, (ultimate quality, masterpiece, highres:1.0), realistic:1.8, photorealistic,####~~~~~~~~~~~~~~~~~~~~~~~~~~~~Issues:back view (from behind) is sometimes a hit and miss. One way to remedy this is to use "looking at viewer" to avoid Adetailer failing on 67degree face angles. Post-production edits/Reproducibility:NoneHappy prompting. And do share [SFW] creations~

JAV Konan_Koyoi (小宵こなん)

v1. Initial release. Trained on 320+ images. Overfitting occured at Epoch 10, rolling back to Epoch #4. Verified and tested for all different camera angles (full body, cowboy shot, profile, upper body), and seems to be stable for most. Retaining epoch 8 as 2nd best alternative.

Files with SHA256: c7b268d2db672007829ec96cabab99c5ab4695a37acd8bc01c402fbb6eab84e6