Mixed_style

PulenKompot-Voluptuous_Animagine.safetensors

1683954_training_data.zip

v4.0

mix_cartoon_v9.0

mix_cartoon_v8.0

v7.0

v6.0

v5.0

v3.0

v2.0

v1.0

A LoRA that is meant to help generate<h1 id="thick-(or-thicc-apparently.-i'll-do-you-one-better-zakrugleni)-hwr8d4jd3">thick (or thiCC, apparently. I'll do you one better -- закръглени)</h1>and<h1 id="voluptuous-women.-uu059z46t">voluptuous women.</h1><h2 id="****update-4-l5fdu7w1b">****UPDATE 4</h2>Not a direct upgrade, but rather a spin-off, the version "Venus" should be pretty self-explanatory. If not, make sure to read its "About this version".As for other updates, I've noticed that I forgot to upload the Pony version trained on the expanded dataset (same one for the <a target="_blank" rel="ugc" href="https://civarchive.com/models/272841?modelVersionId=747822">Animagine-trained</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/272841?modelVersionId=749506">SDXL-base</a>), so stay tuned for that.<hr /><h2 id="***update-3-1fstz2ucj">***UPDATE 3</h2>I know nothing can beat my <a target="_blank" rel="ugc" href="https://civarchive.com/models/662040/pulenkompot-voluptuous-flux">Flux</a> version of Voluptuous, so much prompt accuracy, such great nsfw results, much wow.Having tried training my <a target="_blank" rel="ugc" href="https://civarchive.com/models/272841/pulenkompot-voluptuous-ponyxl-and-sd-15">Voluptuous</a> dataset on a different base checkpoint, I've decided to 1) expand and readjust the dataset, and 2) not just stop with the Flux one (which is definitely not not a failure and I would absolutely continue to create Flux Loras). The Flux checkpoint is a marvel to work with -- easy, quick to set up, light on the GPU, and most importantly - WORDS!On top of the SD1 version, the Pony version, I have one trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/260267/animagine-xl-v31">Animagine XL</a> (great results, and works on Pony, too), and one trained on the base SDXL checkpoint (fits with every SDXL and PonyXL models, and fits well with realistic checkpoints)<hr /><h2 id="**update-2-82fqfb9u1">**UPDATE 2</h2><h3 id="the-ponyxl-version-is-here!-gg721rwhk">The PonyXL version is here! 🥳</h3>It was a pain to train. It took several weeks, and burned down my house. I really need a 4090 (and a new home now)...Considering the capabilities of Pony Diffusion and all its merges, I don't really think a thicc/voluptuous/curvy LoRA is necessary for those models, but for consistency's sake here it is.Prompting the Pony is a bit different then SD1.5 based models. If you want really thick thighs for example don't do ((thick thighs:1.3)) or something similar, simply put huge thighs.Likewise, gigantic breasts, gigantic hips, narrow waist, sagging breasts, etc. are tags that actually work now, so use accordingly (and per what your desired outcome is) within the models and with my LoRA.Use<pre><code>3d</code></pre>in the negative prompt if you want less/no realistic looking images.Happy generating! 🐎🐎🐎<hr /><h3 id="*update-rnj6azr7b">*UPDATE</h3>We are going the Star Wars route, previous versions appear later 😆If you are confused read the "About this version" section.TLDR - previous version is also good, so I thought I could publish it as well.<pre><code>this is a section brake, because I'm dumb and can't figure out how to do it
*UPDATE section end</code></pre>I haven't tested it on too many checkpoints, but it should work fine on most, if not all, of the anime and semi-realistic style models.Works fine with photorealistic models, too, especially if you use <a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a> for the faces.Overall, it should be good on pretty much every famous checkpoint you are probably used to using, and also my checkpoint mixes, including <a target="_blank" rel="ugc" href="https://civarchive.com/models/270793/pulenkompot-kiwi">the Kiwi</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/286841/pulenkompot-kapina">the Kapina</a>. I've used it to generate the examples here.<h1 id="useful-information-(relevant-to-sd1):-4jdpxj3j3">Useful information (relevant to SD1):</h1>As usual, use Hires. fix is pretty much mandatory (at least 1.5 times, but 2 times is what I recommend when it comes to 512x768, or 512x512). Download free VRAM here: <a target="_blank" rel="ugc" href="https://th.bing.com/th/id/OIP.CykaFcF-5V34WbsJfnoOVgHaEK?rs=1&amp;pid=ImgDetMain">click</a>As mentioned, ADetailer to fine tune the face and eyes. The eldritch hands/fingers we just gonna have to live with 🥲.For the other options such as CFG, denoising strength, etc. refer to what is recommended by the model you're using. Or experiment and see what works for your individual case.voluptuous (for v1.2) or goddess (for v1.1) as a trigger tag should be enough, but if you want to make sure, I'd recommend tags such as: wide hips, thick thighs, huge ass, huge breasts, narrow waist, gigantic breasts, sagging breasts - in any combination that you deem fitting and necessary.Make sure to use both hands when typing the prompt."Never get yourself in a sticky situation or get caught with your pants down." - Plato, probablyAnd don't forget, your GPU loves you, but it also needs some rest now and again. Don't abuse it too much.Oh, and, if you're on this page, I'd assume you'd need lots of hydration. 💧💧💧<a target="_blank" rel="ugc" href="https://tenor.com/view/wry-smile-dog-gif-20022420">:)</a>

Voluptuous and Venus [IL] | [SDXL] | [PonyXL] | [SD1]

Trained on the Animagine XL V3I quite like the results the LoRA produces. It can be used with Pony and any of its merges, too, so for variety's sake there's that.Q: How does an anime-trained LoRA produce such realistic results? :O A: Use the "realistic" tag

Files with SHA256: d30c1ea72b16f57b5a3fb6b046fda367adbd6e25aa251f4010ea32a6ee48222c