<h2 id="mamimi-(current-style)-lora-for-anima">Mamimi (Current Style) — LoRA for Anima</h2>This LoRA captures the current art style of Mamimi <code>mamimi_(mamamimi)</code> on Danbooru) for Anima!I'm quite proud of this model — it's my first time training a LoRA and I'm very satisfied with the results. This release is the product of two training runs, merging the best epochs from each one. &gt; 🖼️ All sample images are pure Text-to-Image. No inpainting, no img2img, no Hires fix — completely standard generation. <h3 id="trigger-words">🔑 Trigger Words</h3>- Main trigger: <code>mamimi</code>- Boost: add <code>@mamimi</code> to also use Anima's internal artist style, which strengthens this LoRA nicely.Recommended: use both together → <code>mamimi, @mamimi</code><h3 id="about-the-style"> 🎨 About the Style</h3>This LoRA leans toward Mamimi's current style, after they shifted to a more rendered look. That said, not the entire dataset consists of recent illustrations.It can also achieve interesting compositions beyond simple <code>1girl</code> prompts — one of the reasons I'm happy with the result.<h3 id="things-to-keep-in-mind">⚠️ Things to Keep in Mind</h3>- Bikini bias: The LoRA tends to generate bikinis (or bikini-like outfits) even when you prompt for other clothing. - Fix: add <code>bikini</code> to your **negative prompt**.- Exposed chest: It may generate exposed breasts even when not prompted. - Fix: add negatives like <code>bare breast, exposed breast</code>.- Censor bars on genitals: Black censor bars may sometimes appear on penises. - Fix: greatly reduce them with negatives like <code>censoring, bar censor, black bar censor, black lines on penis, black bars on penis</code>. That said, they aren't always going to disappear, at least with my prompting skills.

Files with SHA256: 326039671837a293ed96d5c8a0aa3f0b47168c6d9789bd9ca1a6550e6ded5d0b