Vladilena Milizé from Eighty Six. Also comes with her alt outfit, Bloody Regina (as a separate lora, more on that below).NOTE: If you're using tag auto completion, the Booru's misspell Lena's name as "vladilena millize" with 2 L's. Make sure you don't generate with the wrong prompt!NOTE 2: "Bloody Regina" is Lena's post time skip design. It's extremely similar to her original design with only the colour being different, so I split it into its own separate lora since I couldn't really train it. The dataset is also a bit limited since most of her art comes from her pre-timeskip design.Bloody Regina is ONLY trained to generate her outfit, it does not handle other clothes/nudes well.<h2 id="heading-561">Vladilena Milize v3.0</h2>Use vladilena milize to generate Lena. Without any other tags it'll probably try to default to her military uniform, but you'll also want to include the uniform tags as well if that's what you're generating:<ul><li>military uniform</li><li>military hat</li><li>garter straps, white thighhighs</li><li>blue boots</li></ul>Only use the tags that you want visible in the generation. i.e. don't add the blue shoes prompt if you're generating an upper body shot. You can also use descriptor tags such as antenna hair for more accurate generations. I'd suggest including holding in the negative prompts since there's a slight bias for it.For NSFW generations, add large breasts to your prompt. I find that the AI is usually horny, but apparently for this model it tries to nerf Lena's bust.<h2 id="heading-3">Bloody Regina v1.0</h2>Use vladilena milize, military uniform to generate Lena. Specifying the level of detail in her outfit will make it more consistent :<ul><li>black jacket</li><li>black skirt</li><li>multiple belts</li><li>garter straps, white thighhighs</li><li>black boots</li><li>(optional) military hat</li></ul>If you don't want her hat to generate, you should put it in the negative prompts for higher consistency. Like I mentioned before, the dataset was somewhat limited so I couldn't get this outfit to train as perfectly as I wanted.

Vladilena Milizé (86 -Eighty Six-)

<ul><li>trained on 768 instead of 512</li><li>trained at 16dim/8alpha and then resized to 8dim/4alpha</li></ul>Overall, this version should provide details better and should be easier to change styles with.

Files with SHA256: 096a2ab7a98f2dc1e64cfe17cbfb6b6ab5e95da93facbb8b47f609864797ee09