<h2 id="mesmera-is-a-lora-(lycorislocon)-trained-to-enhance-the-capabilities-of-illustrious-0.1-(and-it's-finetunes-ofc)-in-generating-hypnosismind-control-related-scenarios.-d0iuq5z4u">Mesmera is a LoRA (LyCORIS/LoCon) trained to enhance the capabilities of Illustrious 0.1 (and it's finetunes ofc), in generating hypnosis/mind control-related scenarios.</h2>Tested on these Illustious-based models: <a target="_blank" rel="ugc" href="https://civarchive.com/models/934764?modelVersionId=1346747">MiaoMiaoHarem v1.5a</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/926443/ntr-mix-or-illustrious-xl-or-noob-xl?modelVersionId=1166878">NtrMix vXII</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184?modelVersionId=1283437">WAI_NSFW v9</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/140272?modelVersionId=1240288">Hassaku v1.3A</a>.<h2 id="training-details:-2ry4880um">Training Details:</h2>Utilizing a large dataset of 22,000 images sourced from hypnohub, this LoRA focuses on a variety of hypnosis-related themes.2xT4 on Kaggle, 30gb vram in total. ~15 hrs training time, 8 epochs. "I forgot the number but it's B I G" steps.Used code from <a target="_blank" rel="ugc" href="https://github.com/hollowstrawberry/kohya-colab">Hollowstrawberry's Google Colab repo</a>, modified, and adapted to Kaggle by me. <a target="_blank" rel="ugc" href="https://github.com/klinok64/LoRA_Easy_Training_scripts_Backend">Kaggle repo</a> (code is a complete mess), find the Training SDXL Lora on Kaggle.ipynb file and import it into Kaggle.All dim alpha values was set to 24.Optimiser: ProdigyScheduler: REXBatch size: 6 (3 per gpu)Grad. acc. steps: 1Additionally used IP noise = 0.1; Prodigy 1.1.2 "slice_p=11" arg; --debiased_estimation_loss<h2 id="dataset:-8f3hc3zs8">Dataset:</h2>The training dataset consists of ~20k images scraped from hypnohub, all centered around hypnosis and related concepts. For a full list of tags used in the dataset, you can refer to the Bottom of the Tag Guide section where you can find a Gist link.Dataset was scraped from hypnohub with this prefill filtering:<pre><code>-manip,-animated,-animated_gif,-voice_acted,-animated_eyes_only,-tagme,-traditional,-sketch,-meme,-ai_art,-caption,-caption_only,-fascinum,-3d_custom_girl</code></pre>All images was downscaled to 1024 (bigger side) pixels by lanczos method in XnConvert tool.My booru scraper script also gathering tags from site. After additional filtering (see below), further tagging with JoyTag (high confidence like 0.6) in the <a target="_blank" rel="ugc" href="https://github.com/Particle1904/DatasetHelpers">Dataset Helpers</a> tool.Additionally some further filtering were applyed:<ul><li>"score:&gt;x" tag, where x value is defined by me for every main tag.</li><li>Aspect Ratio filtering: for every image were calculated Aspect Ratio value (for example value for AR 16x9 will be 1.77), and images that were too long or wide (with an AR value greater than 3) were filtered out.</li></ul>After filtering and tagging, tags in each .txt file was organized by this metod:[Concept tags] = always first in the file, sorted a-z.[Style tags] = not sorted, simply moving them after Concept tags.[Main tags] = [Concept tags] + [Style tags]. The number of main tags is calculated. This is our keep_tokens value.After these steps, .txt files looking like this: [Main tags], other tags. Then all pairs of .txt + .img were sorted by keep_tokens value. Finally, I got a folder structure whose name depended on the keep_token value (i.e. 1, 2, 3, etc.).<h2 id="tagging-guide:-mjj1jhjzy">Tagging Guide:</h2>To utilize Mesmera effectively, you can explore a range of styles and concepts. Here are list of the main tags I was focused on:Meta:<ul><li>femsub: 700 &lt;-- female submissive. use it to specify on whom chosen hypnotic effect will be applied</li><li>before_and_after: 676</li><li>pov_sub: 557 &lt;-- Point of view of submissive character</li><li>sequence: 539 &lt;-- can be used either as sequence of transformations of given character or just comic-style sequence</li><li>see-through: 468 &lt;-- contextually aware, i think...</li><li>malesub: 426 &lt;-- male submissive. same as femsub</li><li>maledom: 309 &lt;-- male dominant. use it to specify who is applying effects on fem/male sub.</li><li>gameplay_mechanics: 289 &lt;-- idk what is that, but it should help with some interfaces or level bars shit i guess</li><li>femdom: 270 &lt;-- female dominant. same as maledom</li><li>pet_play: 222</li></ul>Concepts:<ul><li>kaa_eyes: 700</li><li>happy_trance: 698</li><li>bimbofication: 683</li><li>hypnotized_hypnotist: 680</li><li>transformation: 553 &lt;-- random transformations in training data, can help can make things worse</li><li>breast_expansion: 512 &lt;-- warying breast size, depends on artist style</li><li>hypnotic_audio: 493</li><li>haigure: 480</li><li>dronification: 460 &lt;-- warying results, general non-specific tag</li><li>evil_smile: 444</li><li>hypnotic_screen: 435</li><li>bodysuit: 417</li><li>hypnotic_app: 415</li><li>tech_control: 402 &lt;-- warying results, general non-specific tag</li><li>control_indicator: 400</li><li>robotization: 365</li><li>clothed_exposure: 359</li><li>visor: 355</li><li>progress_indicator: 331</li><li>hypnotic_accessory: 329</li><li>lip_expansion: 326</li><li>zombie_walk: 316</li><li>hypnotic_music: 311</li><li>pendulum: 311</li><li>mantra: 304</li><li>altered_perception: 300 &lt;-- use with "thought bubble" tag</li><li>enemy_conversion: 291</li><li>hypnotic_eyes: 279</li><li>hypnotic_gas: 271</li><li>charm_(spell): 271</li><li>empty_eyes: 270</li><li>hypnotic_breasts: 269</li><li>electricity: 268</li><li>barcode: 266</li><li>hypnotic_penis: 262</li><li>hypnotic_light: 259</li><li>confused: 259</li><li>corruption: 242</li><li>chicken_pose: 235</li><li>spiral_eyes: 228</li><li>unhappy_trance: 212</li><li>shrunken_irises: 203</li><li>glowing_eyes: 201</li><li>hypnotic_beam: 184</li><li>standing_at_attention: 182</li><li>latex: 178</li><li>self_hypnosis: 176</li><li>housewife: 167</li><li>stage_hypnosis: 160</li><li>aura: 160</li><li>hypnotic_tentacle: 154</li><li>eye_roll: 152</li><li>hypnotic_ass: 139</li><li>harem_outfit: 132</li><li>hypnotic_feet: 129</li><li>asphyxiation: 103</li></ul>Characters:<ul><li>erika_(er-ikaa): 510</li><li>kaa: 315</li><li>crystal_(zko): 226</li><li>hypno-tan: 162</li><li>kassidy_(medrifogmatio): 134</li><li>mrs._erickson_(zko): 59</li></ul>Styles:<ul><li>sleepymaid: 624</li><li>jimryu: 512</li><li>darkhatboy: 345</li><li>oo_sebastian_oo: 331</li><li>hadant: 308</li><li>mythkaz: 267</li><li>nexus_light: 253</li><li>lairreverenteboladepelos: 243</li><li>etlabsotwe: 234</li><li>polmanning: 223</li><li>zko: 214</li><li>brellom: 200</li><li>katsiika: 199</li><li>onefeefoor: 196</li><li>eroborne: 196</li><li>zephyrgales: 177</li><li>mahoumonsterart: 177</li><li>apopop: 176</li><li>shishikasama: 174</li><li>smeef: 171</li><li>faetomi: 170</li><li>dochaunt: 167</li><li>porniky: 165</li><li>djuuicebox: 164</li><li>enetheligthingdancer: 160</li><li>ryuugu: 158</li><li>brushie_art: 152</li><li>efalabrino: 151</li><li>eshie: 151</li><li>glowhorn: 151</li><li>cursedrooks: 150</li><li>myuk: 150</li><li>psyfly: 148</li><li>4headboiii: 148</li><li>b-ginga: 144</li><li>detritus: 143</li><li>medrifogmatio: 137</li><li>artofadam: 127</li><li>batta18th: 126</li><li>ghostec: 126</li><li>gmun: 125</li><li>keeper_of_pots: 125</li><li>idpet: 123</li><li>horiizyn: 122</li><li>pstash: 122 &lt;-- not working</li><li>tomo86: 119</li><li>supercasket: 118</li><li>ameerashourdraws: 115</li><li>tunberuku: 113 &lt;-- 50/50, dirty dataset</li><li>httpwwwcom: 112</li><li>electrickronos: 112</li><li>erocoffee: 109</li><li>abarus: 109</li><li>yumiiart: 108</li><li>wrenzephyr2: 108</li><li>foolycooly: 108</li><li>lapislazuliart: 105</li><li>zelhypno: 105 &lt;-- 50/50, why? idk :(</li><li>supersatanson: 103</li><li>vahn_yourdoom: 101</li><li>zorro-zero: 100</li><li>yensh: 97</li><li>davidthewolfx10: 94</li><li>zombi62: 94</li><li>maozi_dan: 89</li><li>jostony24k0: 88</li><li>rosvo: 83</li><li>alerith: 82</li><li>reliusmax: 81</li><li>m4ns0n: 81</li><li>konaloid: 80</li><li>maynara: 80</li><li>sakurarose12: 79</li><li>sweetlittleneko: 78</li><li>malberrybush: 74</li><li>harvestman_here: 70</li><li>kronobas28: 68</li><li>orphan2: 68</li><li>syas-nomis: 63</li><li>zelamir: 62</li><li>nettleseeds: 61</li><li>4five1: 61</li><li>the_iron_mountain: 56</li><li>kibazoku: 55</li><li>cavitees: 55</li><li>sl33pyg1mp: 55</li><li>shozaya: 54</li><li>gerph: 50</li><li>blueparikeet: 49</li><li>drevod: 48</li><li>shieol1: 46</li><li>hmage: 45</li><li>konno_tohiro: 45</li><li>rnslivr: 44</li><li>deepspaceart: 42</li><li>singlesalt: 40</li><li>magukappu: 39</li><li>borvar: 38</li><li>glatu: 38</li><li>cuddlesword: 36</li><li>nez-box: 36</li><li>terasu_mc: 36</li><li>strwbrrychz: 35</li><li>thiccwithaq: 35</li><li>a_singular_fish: 33</li><li>hentai_man: 32</li><li>chien_vietnam: 32</li><li>alittleshyart: 30</li><li>rind_rin00: 28</li><li>slugbox: 24</li></ul>Full list of tags in the Mesmera's training dataset can be found in this <a target="_blank" rel="ugc" href="https://gist.github.com/klinok64/ce4d0e81cf6baa02d5625144b5a61a12">Gist</a>.<h3 id="mesmera-v1-gist-links:-k4cmmbpr1">Mesmera v1 Gist links:</h3>Tagging guide: <a target="_blank" rel="ugc" href="https://gist.github.com/klinok64/6b59f9ba9febff1d2a6c68efe19f24a1">Gist</a>.Full list of dataset tags:<a target="_blank" rel="ugc" href="https://gist.github.com/klinok64/1094d1b865945a6c9203cd3cafb1c327"> Gist</a>.<h2 id="roadmap-(kinda):-juiwrof5v">Roadmap (kinda):</h2>So... This LoRA is mostly working, but I'm still not satisfied with several quality aspects (for example, it was trained at 512 resolution). I will likely release Mesmera v2.5 this year, though I don't know exactly when. I feel slightly burnt out by this project, and I'm not going to push myself into it any further. I will focus on small-effort LoRAs for now. (27.03.25)After I release Mesmera v2.5 for Illustrious 0.1, I will likely follow this training plan:<ol><li>NoobAI Eps</li><li>NoobAI V-pred</li><li>Pony v6</li></ol>And after Pony I will mostly handling user requests:<ul><li>Training for other models on the SDXL architecture: IL 1.1, IL 2.0/3.0 (hopefully), RouWei, Hassaku, WAI, etc.</li><li>I will also try to train for Flux, Lumina, SD1.5, SD3.5, and Sana.</li><li>Additionally, I will try to experiment with realistic style fine-tunes (Illustrious, Pony).</li></ul>

[IL 0.1] Mesmera | The Biggest MultiStyle+MultiConcept Hypnosis LoRa Trained Yet

Prob last version here. This is ft on top of v2, just on higher reso dataset. Changes are minimal, slitghtly better quality.

Files with SHA256: 56608ae972fa6d63114cb40bc8f0b99215d6c8044f100f300938fa66cef3f6a1