|
"description": "<p><span style=\"color:rgb(250, 82, 82)\">UPD 09.08: </span>V2 is out.<br />What new:<br />1) I've trained the model for additional ~26 535 steps on a completely new set of pictures.<br />2) Better style variety, way better natural language prompt understanding, better prompt <span style=\"color:rgb(193, 194, 197)\">adherence, better SFW. </span><br /><span style=\"color:rgb(193, 194, 197)\">3) </span><strong><span style=\"color:#fa5252\">BUT!</span></strong><span style=\"color:rgb(193, 194, 197)\"> Hardcore scenes are now less consistent (you can still get them right, but it will take more rerolls). I wanted to fix that, but honestly I got too exhausted from working on V2. Sorry guys, probably not the update you expected, but it's definitely the update the checkpoint needs.</span><br /><br /><strong><span style=\"color:rgb(76, 110, 245)\">LUSTIFY!</span></strong> <br /><br />A photoreal checkpoint, capable of creating decent porn scenes without any loras (although, with just a little help of your favourite <strong>NSFW </strong>concept lora it can get even more consistent). But that's not the only use - it can also create gorgeous <strong>SFW </strong>pics of objects/animals/fantasy creatures/you name it (you'll have to convince it tho - the model is horny).<br /><br />\"<strong><span style=\"color:rgb(76, 110, 245)\">LUSTIFY!</span></strong>\" understands both danbooru tags and natural language prompting, because it was trained like that. For V1 (not needed for V2) I recomend putting \"score_9\" tag into the begining of your prompt. Yes, the tag from the pony models. Yes, I am sure.<br /><br />A couple of things I need to make clear <strong><span style=\"color:rgb(76, 110, 245)\">before you download the checkpoint</span></strong>:<br />1) All of my stuff is very opinionated. I create models mostly to suit my needs, my workflows, my prompting, etc. <br />2) Sometimes you can get bad hands and bad faces at distant shots. These are hard to fix with finetuning, but really easy with inpainting/adetailer. So, please, don't bother me with the complaints about this. Either adapt to the needs of this checkpoint, or use something else (spoiler, 99% of checkpoints suffer from that because of the limitations of this tech).<br />3) Everything I do is experimental, and I do a lot of dumb-crazy stuff that in theory shouldn't work. So it's to be expected that <em>something, </em>what used to work with other checkpoint for you, might not work with this one. Let me know if there are such things, and I'd see if I can do something about it.<br /><br /><strong><span style=\"color:rgb(76, 110, 245)\">Recomended parameters:</span></strong><br />Sampler: DPM++ 2M SDE/DPM++ 3M SDE, <br />Scheduler: Exponential/Karras<br />Steps: 30<br />Cfg: 4-7.<br />Highres.fix: upscale by 1.4-1.5, denoising ~0.4<br />For the distant shots of people you probably wanna <strong>use</strong> highres.fix and/or adetailer.</p><p>Also, use reasonable positive/negative prompts, shizo prompting does more harm than good (with any SDXL model).<br /><br /><strong><span style=\"color:rgb(76, 110, 245)\">FAQ:</span></strong><br />1. Q: Output is broken is Fooocus. A: Turn off \"Fooocus Enhance\" style.<br /><br /><strong><span style=\"color:rgb(76, 110, 245)\">Some tags that have a big visual impact with this model:</span></strong></p><ol><li><p>Camera type tags: \"shot on Polaroid SX-70\", \"shot on Kodak Funsaver\", \"shot on GoPro Hero\", \"shot on Canon EOS 5D\", \"shot on Leica T\".</p></li><li><p>Photography styles: \"analog photo\", \"glamour photography\", \"street fashion photography\".</p></li><li><p>Lighting types: cinematic lighting, \"neon lighting\", \"soft lighting\", \"dramatic lighting\", \"low key lighting\", \"bright flash photography\", \"warm golden hour lighting\", \"radiant god rays\".</p></li><li><p>Film types: \"Ilford HP5 Plus\", \"Lomochrome color film\", \"Fujicolor Pro\".</p></li><li><p>Photographers: Alessio Albi, Martin Schoeller, Miles Aldridge, Oleg Oprisco, Tim Walker.</p></li><li><p>Others: \"bokeh\", \"dreamy haze\", \"technicolor\", \"underexposed\".</p></li></ol><p>Alternatively, you can give a read to <a target=\"_blank\" rel=\"ugc\" href=\"https://civitai.com/models/118213/creating-photorealistic-images-with-ai-using-stable-diffusion\">this </a>wonderful prompting guide by <a target=\"_blank\" rel=\"ugc\" href=\"https://civitai.com/user/PromptGeek\">PromptGeek</a>. It has A LOT of useful tips I didn't know about even after 2 years of using SD.<br /><br /><strong><span style=\"color:#4c6ef5\">Work In progress:</span></strong><br />Initially it started out as a mix of my personal unreleased SDXL finetune with such wonderful checkpoints as <a target=\"_blank\" rel=\"ugc\" href=\"https://civitai.com/models/502468?modelVersionId=558487\">bigASP </a>and <a target=\"_blank\" rel=\"ugc\" href=\"https://civitai.com/models/133808?modelVersionId=324958\">Pyro's NSFW</a>. Then I've finetuned this mix on a curated dataset of 1000 pics for aesthetic tweak.<br />The recipe was something like: 0.55*(my unreleased SDXL finetune)+0.45*(0.5*(bigASP)+0.5*(Pyro's NSFW)). <br />For V2 I've merged V1 with <a target=\"_blank\" rel=\"ugc\" href=\"https://civitai.com/models/617652?modelVersionId=690475\">NatVis </a>(20%) and then trained on new 1700 pics for additional 26 535 steps.<br /></p>", |
