<h2 id="pony-this-nai-that.i-just-want-to-grill.-bxg8kvrhr">Pony this, NAI that. I just want to grill.</h2><h3 id="this-is-a-dumping-ground-for-my-illustrious-and-noobai-based-style-mixes.-lmbaxu4ct">This is a dumping ground for my Illustrious and NoobAI-based style mixes.</h3><h3 id="tldr-generation-settings:-z7ctz2ewl">TL;DR Generation Settings:</h3><ul><li>Sampler: Restart / DPM++ 2M / DPM++ 2s a / DPM++ SDE<ul><li>Euler a will result in a much more smoothed-out look</li></ul></li><li>Scheduler: Karras / SGM Uniform</li><li>Sampling Steps: 25</li><li>I upscale to 4x using <a target="_blank" rel="ugc" href="https://civarchive.com/articles/4560/upscaling-images-using-multidiffusion">this process </a>and the upscaler below:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri/blob/main/4x_foolhardy_Remacri.pth">4x_foolhardy_Remacri</a></li></ul></li><li>Dimensions: 832x1216, 1216x832, or 1024x1024</li><li>CFG Scale: 5 - 7</li><li>One or more passes of <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a><ul><li>Some combination of the following for faces and eyes<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/293448?modelVersionId=331333">Anzhc's Face Seg Prototype</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/342514?modelVersionId=383427">Anzhc's BeautifEYEr</a></li></ul></li></ul></li></ul>I use the following positive and negative prompt:<ul><li>masterpiece, best quality, {prompt}<ul><li>oil painting \(medium\) can help enhance the look</li></ul></li><li>(lowres:1.2),(worst quality:1.2), (low quality:1.2), jpeg artifacts, bad anatomy, bad hands, text, {prompt}</li></ul>See sample images for examples.V1:Hastily made style mix, an attempt to bring over the style of my Pony-based <a target="_blank" rel="ugc" href="https://civarchive.com/models/346204/mfcg-paintjob">Paintjob</a> to <a target="_blank" rel="ugc" href="https://civarchive.com/models/795765?modelVersionId=889818">Illustrious</a>. Models were merged into <a target="_blank" rel="ugc" href="https://civarchive.com/models/818750/lucereon-or-artists-finetune-or-illustriousxl">Lucereon</a> and will likely be released standalone later on. Some models used are a bit overfit, so poses may repeat.V1.5:Merged into <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=968495">NoobAI Epsilon Pred 0.5.</a> A cleaner look, thicker outlines. DPM samplers with the Karras scheduler give a more drawn look, Restart with SGM Uniform gives a much cleaner look. Lighting seems to be inconsistent. This might be a NoobAI thing or something on my end, not yet sure. This version does not include <a target="_blank" rel="ugc" href="https://civarchive.com/models/839053/better-backgrounds-for-illustriousxl">Cham's background enhancer</a>, so use that if you find the backgrounds lacking. Lips is a particularly strong tag here.V1.55:Very similar to v1.5. Changed the NoobAI model used to v0.75 from v0.5.V2.0 - AAA Please Read:Fair warning, most models trained on Illustrious or earlier versions of NoobAI (0.5 and 0.75) probably will not play nicely with this. This is due to using the latest (as of now) version of <a rel="ugc" href="https://civarchive.com/models/353543/aaaaaaaaaaaaaaaaaaaaaaa-or-finetune-mix-on-whatever-model-i-want-at-that-point-which-is-illustrious-xl-right-nowbut-i-will-keep-pony-one-for-on-site-as-well">AAA</a>; this has custom TENCs trained on top of NoobAI v1.0. Models used were trained under NoobAI v1.0. I'm using a smaller default negative prompt:<ul><li>worst quality, low quality, bad anatomy, {prompt}</li></ul>As with most of my mixes, this is intended to produce female characters and leans towards NSFW.Big thanks to the following for their help and knowledge: @bluvoll@Anzhc@FallenIncursio@Manityro@mnemosynekoto@EDG@richyrich515@PotatCat@ChameleonAI

MFCG's Mixed Grill

<ul><li>Remade underlying models on NoobAI EpsilonPred v1.0</li><li>Remerged new models into AAA as the base model</li><li>Might not have compatibility with LoRAs made for earlier Illustrious or NoobAI versions (e.g. 0.5 or 0.75)</li></ul>

Files with SHA256: f956e8ca2a930895b9b57f4968c52bb3938b4d7ccb3cd669c693743817eb1963