<h1 id="slima">Slima</h1>Bringing all your high quality slime girls to AnimaRefer to the <a target="_blank" rel="ugc" href="https://civarchive.com/models/2458426/anima?modelVersionId=2945208">official Anima page</a> for recommended parameters and documentation.You might be wondering why I made this a full finetune instead of a LoRA, and frankly I'm wondering that too now. But it's too late. The high quality slime girls are here to stay.The text encoder and VAE are not included so remember to grab those from the official Anima page.<h3 id="the-intention">The Intention</h3>Anima can do slime girls decently well and it does kinda understand transparency, but Slima enhances that by bringing the aesthetics of a high quality Illustrious checkpoint to the table. No more cartoonish slime girls. The transparency effect for penetration has also been enhanced by sharing a lot of the same dataset I used for <a target="_blank" rel="ugc" href="https://civarchive.com/models/2322841/translucent-penetration">Translucent Penetration</a>. Also note that this was trained specifically for NSFW so you might have a hard time getting SFW outputs if that's what you are looking for.<h3 id="the-motivation">The Motivation</h3>Anima has great prompt adherence... for the concepts it knows about. So the real reason I made this is because I wanted better prompt adherence for the Translucent Penetration V3 dataset. The fact that I could only find a single slime girl LoRA for Anima was also a tragedy so I decided to do a full finetune.<h3 id="the-result">The Result</h3>Slima V1.0 is undercooked still but it's like 90% of the way there. Most things should work and prompt weights can be used for the stubborn things. Anima base can do 1 shot generations up to ~2048x2048, but I don't have enough high quality images in my dataset for that yet. Text-to-image is pretty stable only up to ~1536x1536, but resampling at higher resolutions should still work properly. <h3 id="generation-parameters">Generation Parameters</h3>The supported samplers are the same as the base model, but euler_ancestral/beta is usually a bit more stable than er_sde. I typically generate with:<ul><li>Steps: 37</li><li>Sampler: euler_ancestral</li><li>Scheduler: beta</li><li>CFG: 4.0</li></ul>but it might be better to use 40-50 steps for now since it's still a bit undercooked.<h3 id="prompts">Prompts</h3>Normal Anima prompts are fine since the dataset was captioned with both tags and natural language. Illustrious style prompts that are mostly/fully tags also work, but if want to increase the weight of certain terms, the number will need to be higher. For example <code>(tag:1.2)</code> for Illustrious would be somewhere around <code>(tag:2)</code> for Anima, tho it's not an exact mapping.You probably want <code>slime girl</code> to be in your prompt somewhere too. The transparency will still work without it, but then you will have a normal girl with transparency and some slime traits. I recommend adding <code>missing legs, melting, deformed, different character sizes</code> to your negative prompt.<h3 id="prompt-examples">Prompt Examples</h3>V1.0 cover image:<pre><code>solo, slime girl, goo girl, monster girl, blue skin, transparent, medium breasts, round butt, slime detail, sitting on legs, arched back, from side, low angle, loving stare, beautiful woman, mature, sexy, cute, smug, dominant, narrowed eye, in heat. Her head is turned slightly to the side. She is looking at the viewer out of the corner of her eye.</code></pre>POV Cowgirl<pre><code>slime girl, goo girl, monster girl, blue skin, transparent, medium breasts, slime detail, pov, sex, vaginal, cowgirl, straddling, riding penis, upright, beautiful woman, mature, sexy, cute, smug, dominant, narrowed eyes, horny</code></pre>Oral<pre><code>slime girl, goo girl, monster girl, blue skin, transparent, medium breasts, slime detail, oral, lips around penis, penis in neck</code></pre>Doggystyle (anal)<pre><code>slime girl, goo girl, monster girl, blue skin, transparent, medium breasts, round butt, slime detail, all fours, arched back, from side, low angle, beautiful woman, mature, sexy, cute, submissive, orgasm, sex, anal, flushed, happy, penis visible inside body</code></pre><h3 id="issues">Issues</h3><ul><li>Details get a bit messed up sometimes. Increasing your steps to 50 (if you haven't already) and/or adding another sampler with a light denoise usually fixes this.</li><li>In transparency cases with penetration, the penis might be a lot thinner and/or way too long inside. The size can be mostly fixed by adding <code>distorted penis, disconnected penis</code> to the negative prompt. Sometimes <code>deep penetration</code> in the prompt causes the issue with the penis being way too long since it tries to take that literally.</li><li>Certain angles and concepts can be pretty stubborn. These kinda of issues usually have a hard time getting fixed with natural language prompting, so don't be afraid to crank up the weights for important tags.</li><li>The style is pretty strong so prompting for specific characters or artists might not work as well.</li></ul><h3 id="credits">Credits</h3>About 90% of the dataset consisted of images I generated with Illustrious over the past few months (filtered for quality and consistency). Others were from various sources or generated with WIP versions of the checkpoint. Here's a non-exhaustive list of resources I used or got inspiration from.Models:<ul><li><a rel="ugc" href="https://civarchive.com/models/2458426/anima">Anima (Official)</a> (base-v1.0 &amp; preview-3)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/1034922/dasiwa-illustrious-or-anime">DaSiWa Illustrious | Anime</a> (AlluringAbyss &amp; SweetPoison)</li><li><a rel="ugc" href="https://civarchive.com/models/1641408/nova-kemono-xl">Nova Kemono XL</a> (v8.0)</li><li><a rel="ugc" href="https://civarchive.com/models/827184/wai-illustrious-sdxl">WAI-Illustrious-SDXL</a> (v17.0 &amp; v14.0)</li></ul>LoRAs:<ul><li><a rel="ugc" href="https://civarchive.com/models/1047424/dasiwa-illustrious-slime-girl">DaSiWa-Illustrious-Slime-Girl</a> (v2.0)</li><li><a rel="ugc" href="https://civarchive.com/models/2430224/realistic-slime-girls">Realistic Slime Girls</a> (v1.0)</li><li><a rel="ugc" href="https://civarchive.com/models/344495/slime-girls-concept">Slime Girls - Concept</a> (v2.0)</li><li><a rel="ugc" href="https://civarchive.com/models/468633/slime-girl-lora">Slime Girl Lora</a> (v2.5)</li><li><a rel="ugc" href="https://civarchive.com/models/1398563/handjob-while-penetrated">handjob while penetrated</a> (illustr v0.11111)</li><li><a rel="ugc" href="https://civarchive.com/models/349436/slime-girl-concept">Slime Girl Concept</a> (v1.0 [Illustrious])</li><li>史莱姆娘 <a rel="ugc" href="https://civarchive.com/models/57112/slime-girl">slime girl</a> (slime girl XL-2)</li></ul>Inspiration:<ul><li><a rel="ugc" href="https://civarchive.com/search/images?tags=slime%20girl&amp;users=pureevilcorp&amp;sortBy=images_v6%3Astats.reactionCountAllTime%3Adesc">pureevilcorp</a></li><li><a rel="ugc" href="https://civarchive.com/search/images?tags=slime%20girl&amp;users=jdl_grmck&amp;sortBy=images_v6%3Astats.reactionCountAllTime%3Adesc">jdl_grmck</a></li><li><a rel="ugc" href="https://civarchive.com/user/GoonerGrimoire">GoonerGrimoire</a> (<a rel="ugc" href="https://civarchive.com/posts/25408174">post1</a> <a rel="ugc" href="https://civarchive.com/posts/25484739">post2</a> <a rel="ugc" href="https://civarchive.com/posts/25484763">post3</a> <a rel="ugc" href="https://civarchive.com/posts/25508649">post4</a>)</li><li>Any other users that come up in a <a rel="ugc" href="https://civarchive.com/search/images?tags=slime%20girl&amp;sortBy=images_v6%3Astats.reactionCountAllTime%3Adesc">slime girl search</a></li></ul>

Slima

Initial release trained on anime-base-v1.0

Files with SHA256: 2ba4370035930645879bbe8bc5d45c4ca71c0ac36a4f38641c8d3d09ba6f96c4