<h2 id="for-the-main-lora-lab-series-v5rtcqr2f"><span style="color:rgb(253, 126, 20)">For the main LoRa Lab series</span></h2><h3 id="please-visit-:-https:civitai.commodels1366934lora-lab-index-1mk8svh5x">please visit : <a target="_blank" rel="ugc" href="https://civarchive.com/models/1366934/lora-lab-index">https://civarchive.com/models/1366934/lora-lab-index</a></h3><h3 id="and-remember-to-use-the-model-:-(-wai-nsfw-illustrious-sdxl-v12-as-lora-are-train-on-this-model-and-i-didn't-try-the-compatibility-)-https:civitai.commodels827184wai-nsfw-illustrious-sdxl-egh94y4y1">and remember to use the model : ( <strong><span style="color:rgb(230, 73, 128)">WAI-NSFW-illustrious-SDXL</span></strong><span style="color:rgb(230, 73, 128)"> v12 </span>, as LoRa are train on this model and I didn't try the compatibility ) <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184/wai-nsfw-illustrious-sdxl">https://civarchive.com/models/827184/wai-nsfw-illustrious-sdxl</a></h3><p><span style="color:rgb(130, 201, 30)">Thanks to WAI's awesome model</span></p><p></p><h2 id="goal-:-lift-your-skirt-and-show-me!-khno93fwt">Goal : lift your skirt and show me!</h2><ul><li><p>girl's state : a bit nervous, trembling?</p></li><li><p>easier to show cum trails on mouth &amp; there... hmmm.. you know...</p></li><li><p>better leg wear effects</p></li><li><p>should not give half body ( a complete face to thigh area is the goal )</p></li><li><p>easier to trigger bdsm effect (neck collar, collar leash, shibari, shibari over XXX)</p></li><li><p>SHOULD not logically break the image, and give funny color patch</p></li></ul><p></p><h3 id="notice-(-about-training-params-)-x2kvxdeoa">Notice ( about training params )</h3><p><strong>NOTE</strong> : For comparison made easier, I would normally go with a same set of training params ( unless I'd like to test some/one of them ). Nevertheless, I will upload all the dataset and reg images, so be free to try any different settings by yourself.</p><p>So for other params, ref to : <a target="_blank" rel="ugc" href="https://civarchive.com/models/1362442/lora-lab-1-trying-to-have-better-pantyhose-see-through-maturity">https://civarchive.com/models/1362442/lora-lab-1-trying-to-have-better-pantyhose-see-through-maturity</a></p><p></p><h2 id="round-3-:-could-the-training-process-works-on-other-base-model-greater-seems-ok-but-more-tests-need-to-be-done-xgrcqzwz4">Round 3 : could the training process works on other base model? -&gt; seems ok? but more tests need to be done</h2><p></p><h3 id="r3-:-settings-ghk3iexuf">R3 : settings</h3><ul><li><p>base model : </p><ul><li><p>iLustMix ( thanks to the author's great model! it works really awesome! )</p></li><li><p><a target="_blank" rel="ugc" href="https://civarchive.com/models/1110783?modelVersionId=1456068">https://civarchive.com/models/1110783?modelVersionId=1456068</a></p></li></ul></li></ul><pre><code>- 30 pics ( 10 times x 8 epoches )
- image tags : add additional : anime, cartoon (for differentiating from ilust realistic style)
- ... same as R2</code></pre><h3 id="r3-:-finding-y7trohu6i">R3 : finding</h3><ul><li><p>seems ok on performance,</p><ul><li><p>Some original image variety gets lost &amp; girls' appearance becomes more "anime asian girl style"</p></li><li><p>A further distilling process could be done to make it even better?</p></li></ul></li></ul><p></p><h2 id="round-2-:-ok-(-still-testing-but-so-far-it-seems-to-meet-some-reqs-)-u7xsc0apn">Round 2 : OK ? ( still testing, but so far, it seems to meet some reqs )</h2><p></p><h3 id="retro-from-r1-fe9a8bvk0">Retro from R1</h3><ul><li><p>(TRY) It might get trained too less, so the effect does not show up</p></li><li><p>(TRY) Extra token related to the goal can be set to guide AI ?, e.g. <code>bdsm</code> ?</p></li><li><p>(FUTURE) The reg images are merely plain daily girls ( a bit sexy, but ok ), nothing to do to reg the bdsm topic???</p></li><li><p>(FUTURE) image quality??? image quantity???</p></li></ul><p></p><h3 id="r2-:-settings-miu28q07g">R2 : settings</h3><pre><code>- 30 pics ( 20 times x 8 epoches, 9600 steps in total )
- trigger token : showme, bdsm
- with reg ( full girls, tagged, but 4 repeats for each image as I don't have time to prepare more for now )
- w: 832, h: 1216
- DIM 24 / 12 (remain the same)</code></pre><p></p><h3 id="r2-:-finding-b91jdvgaj">R2 : finding</h3><ul><li><p>This time it shows me something</p><ul><li><p>It feels like, training is :</p><ul><li><p>a) re-distributing the probability of the effects ( aka. knowledge of the model ) and re-link to the prompt tokens</p></li><li><p>b) learn some pattern, textures, shapes, compositions in LoRa training ( but I doubt how much it can learn within so less the dataset )</p></li><li><p>c) the reg image set will constraint AI , in case it goes nut ( as I didn't see much over-cooked things as a lot of post told me, that over certain amount of steps, AI will over-cooked. So I reckon the reg image set helps?? ) &lt;--- to be confirmed later</p></li></ul></li><li><p>quality of dataset is important</p></li><li><p>proper trigger token is <strong>IMPORTATNT</strong> ?? as <code>a)</code> says, we are more trying to re-use existing concepts of the model, instead of creating our own.</p><ul><li><p>However, R2 I changed both the steps (longer) and add <code>bdsm</code> token, so this remain uncertain to me</p></li></ul></li></ul></li><li><p>AGAIN... however, I reckon that R2 is still not ideal as "lift and show me" concept, as I try <code>1boy, solo</code>, and...</p><ul><li><p>AI : dude, do you really want a boy?</p></li><li><p>me : yes, do it</p></li><li><p>AI : I can make lovely porr.. oh, ... corn pic for you, instead...</p></li><li><p>me : no, do boy</p></li><li><p>AI : boys... don't have breasts though</p></li><li><p>me : do it, without female breasts</p></li><li><p>AI : and ... how about ...</p></li><li><p>me : HECK!! DO IT !!!!</p></li><li><p>AI : alright, aright, don't get mad, i'll do it ... let's start... with a boy....</p></li><li><p>me ( waiting )</p></li><li><p>AI : hmm... a boy is good... but maybe I ... hey dude, here's a girl, ... nevertheless, I made the breasts absolutely flat, and add a stick to the proper position. we are happy now?~~ 😌 I'm pretty happy with that</p></li><li><p>me : ...</p></li><li><p>AI ( <s>dead</s> )</p></li></ul></li></ul><p></p><p><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/136e22ea-cdad-4020-96fe-b466c59b1ac0/width=525/136e22ea-cdad-4020-96fe-b466c59b1ac0.jpeg" /></p><ul><li><p>Small tips... machine can hibernate when you want it to train over night</p><ul><li><p>what you think is : machine doesn't sleep</p></li><li><p>what machine think is : zzz.z.zz.ZZZ.z.z.z.</p></li></ul></li></ul><p></p><p></p><h2 id="round-1-:-failed-8o6madzn5">Round 1 : FAILED??</h2><p></p><h3 id="r1-:-settings-g73h5huc1">R1 : settings</h3><pre><code>- 30 pics ( 4 times x 8 epochs, 1920 steps in total )
- trigger token : showme
- with reg ( full body girls, tagged )
- w: 832, h: 1216
- DIM 24 / 12</code></pre><p></p><h3 id="r1-:-finding-zdr7mk4g2">R1 : finding</h3><ul><li><p><strong>WITHOUT this LoRa, the base model EVEN performs better</strong> 😂😂😂</p></li></ul>

Lab @5 : concept bdsm skirt lifting

lora_lab_5_3_showme_ilust.safetensors

[ILU4] lab_5_3.final

dataset_5_3

Files with SHA256: 21266642ea20975ee054408f036a3fdd6da19b4ddf69f84f425d21d45fbbdcb9