LoCon trained on Fiz-Rot's artstyle for ponydiff: <a target="_blank" rel="ugc" href="https://civarchive.com/models/257749">https://civarchive.com/models/257749</a>Artist: He's back but his new stuff isn't making it into my dataset, find him yourself on PixivBecause of my lack of experience and time, the version number may be based on changes that have made certain aspects worse. As such a previous version may perform better than a later version.If you need it for some reason the (old) standard LoRA version exists here: <a target="_blank" rel="ugc" href="https://civarchive.com/models/288465">https://civarchive.com/models/288465</a>I'm still learning the ins and outs of the model and LoRA training in general, please leave reproducible feedback or tips so that I may improve over time.Check the "About this version" section for any quirks or pitfalls I'm aware of or have experienced.<h3 id="some-q's:-0exo5ys6w">Some Q's:</h3><ol><li>What's your workflow?<ul><li>Sampler from images is done with Restarts with segments of [3,2,0.06,0.30], which are values from the paper on Restarts<ul><li>21.1 The scheduler is Karras</li></ul></li><li>Initial resolution is selected from the SDXL whitepaper appendix I: <a target="_blank" rel="ugc" href="https://arxiv.org/pdf/2307.01952">https://arxiv.org/pdf/2307.01952</a></li><li>Perturbed Attention Guidance is used for the initial image</li><li>Upscaler is usually 4x_fatal_Anime_500000_G or 4x_NMKD-Siax_200k</li><li>Second sample pass is the same as the first (params are on civit's meta interface) but at ~.35 denoise</li><li>A facedetailer is used</li><li>I use ComfyUI, sorry</li></ul></li><li>What strength should I be using?<ul><li>Base Pony: .9 - 1</li><li>AutismMix: 1</li><li>21.1 I recommend Autism_Confetti at .8-1 strength, or &lt;.8 if mixing with other resources<ul><li>I absolutely recommend finding a merge at this point, Autismmix Confetti is my go-to but there are plenty of good ones like ponyFaetality or cashmoneyAnime</li></ul></li></ul></li><li>Pairs of images?<ul><li>Left image is AutismMix, Right image is base Pony</li><li>21.1 All images are AutismMix Confetti</li></ul></li><li>Why not put this as a version on the existing model?<ul><li>As far as I'm aware civit does not allow me to independently tag standard LoRA versions and LyCORIS versions on the same model page, so I've separated them to reflect this.</li></ul></li><li>What does your filename structure mean?<ul><li>It's split into 4 parts of [XX]-MAnon-[YYYY]-[ZZZZZZ]<ul><li>[XX] - Artist initials</li><li>MAnon - My moniker</li><li>[YYYY] - SD version and Model version (Ex. XLV4.1 is on SDXL [Not 1.5] and it's version 4.1 of the model)</li><li>[ZZZZZZ] - The epochs, only useful to me</li></ul></li></ul></li><li>This sucks<ul><li>God as my witness I'll make it good one day</li><li>God as my witness I think it's good now</li><li>God as my witness It's firmly mid</li></ul></li></ol>

Fiz-Rot - LyCORIS LoCon - PonyDiffusionXLV6

Second LoConTeeth continue to be a problem, but slightly less soCharacter glow also is a problem, but adding '[color] glow' or 'fog' to negative seems to helpTorsos will sometimes be longer than usual, may just be a resolution thingCertain backgrounds and objects lose a massive amount of coherency, this seems to also be a problem I missed in previous versionsIt's finally usable, in my opinionPerforms well on both base Pony and Merges in different ways, cannot recommend one model over another

Files with SHA256: 94127177224b9d70f4537a43fb670a8aa6648f174d020930527e76ce9d4de39b