Tired of the dull, basic, and default ANIMA results? Have you tried a thousand style LoRAs with none of them satisfying your original ideas and imagination?Try this out!This is a basic command text file that guides large language models to transform any prompts you give them and write multiple versions of them with style and vibe references that Anima models support.You can see the huge improvement in the example images! (they're basically the same ideas on the same checkpoints but with different refrences)Just copy paste it to a chat and then send your prompt or describe a mood, idea and theme and ask them to write in those frames.<h3 id="which-llms-to-use">Which LLMs to use?</h3><ul><li>Grok (Uncensored): It's uncensored, and you can use it for converting any NSFW prompts. Plus, if you're on the free plan and it stops answering you, you can just go to Twitter (X if you're nasty) and use it there anytime, you don't have to wait for its limit to expire.</li><li>Claude (censored): It's the most clever model, exceptional at generating long and detailed prompts in huge quantities. You can even ask it to write 10 styles at a time without it reducing and butchering them for you.</li><li>Gemini (censored): I like it because of its creativity. I don't know what's different about it, but the prompts it returns to me generally have better visual impact. I tried reading and comparing the prompts to other language models, but I can't figure it out, there's something unique about it.</li><li>ChatGPT (censored?): Eh... it's okay. The only unique thing about it is that you can easily trick it into bypassing its filter and getting it to write NSFW prompts. Oh well, solo NSFW prompts at least.</li><li>MiniMax (censored): basically claude more and less.</li></ul><h3 id="checkpoints-suggestion-:">Checkpoints suggestion :</h3>For the best results, I suggest using one of these checkpoints, since some checkpoints are way too stylized on their own and no longer understand the base Anima styles and references.<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2544636/wai-anima?modelVersionId=2983680">WAI-ANIMA</a> (stays very close to the base Anima while also supporting NSFW work)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2642932/janima?modelVersionId=2967640">JANIMA</a> (excellent at concept art and various artistic illustrations not good at NSFW)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2578175/ramthrusts-nsfw-pink-alchemy-anima?modelVersionId=3053107">RAMTHRUST</a> (very coherent at prompt understanding and executing artistic ideas)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/941345/hoseki-lustrousmix-anima-v10?modelVersionId=2982979">Hoseki</a> (very similar to JANIMA, but sometimes even better)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2641326/hassaku-anima?modelVersionId=3195694">Hassaku</a> (one of the best, gives totally different results even on the same seed compared to other models; if most models are returning nonsense results, try switching to Hassaku from time to time)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/2695493/one-obsessionanima?modelVersionId=3190113">One-Obsession</a> (a weird one, to be honest, sometimes it nails it, sometimes it totally misses. Its best feature is prompt understanding; it executes complex poses and ideas better than most models but still lacks in delivering some requested style references and tends to switch to its default style. V1 is better than V2, and I wasn't rich enough to try V3 yet)</li></ul>Add any others you've tried and that worked for you in the comments, for me and others to try out!<h3 id="avoid-these-models:">Avoid these models:</h3>These are the stylized checkpoints that, no matter the prompt or style references, always result in their own style:<ul><li>Diving-Anima</li><li>Miao-Miao</li><li>Unholy-Desire</li><li>riMixlustrious</li><li>ultraFineTune</li><li>Real-Dream (and any realism/semi-realism promises)</li></ul>These are by no means bad checkpoints, infact, they're some of my favorites but they're not built for executing various styles; they're great at their own unique style.

Files with SHA256: 168518b6f76320a37e13c9c409373c04c35fb9adacde148d50c5c3b395234c95