这是一个以Pony为底模训练的兰斯10画风的莉亚·帕拉帕拉·利萨斯的LoRA，可以在保持比较好的兰斯10原作画风的基础上召唤莉亚。————————————<h3 id="tags-g1phd3u9z">Tags</h3>Pony底模常用(Commonly used with the Pony base model)：<pre><code>score_9,score_8_up,score_7_up,score_6_up,score_5_up,score_4_up</code></pre>画风 (Art style)：<pre><code>Rance_10_art</code></pre>角色 (Character)：<pre><code>Ria_Parapara_Leazas, 1girl, blue hair, blue eyes, long hair, medium breast</code></pre>服饰 (Outfit)：<pre><code>white dress,capelet,black pantyhose,bare shoulders,purple skirt, hair ornament</code></pre>————————————<h3 id="(recommended-base-models)-pomyx6oay">推荐底模(Recommended Base Models)：</h3>Hassaku XL (Pony)<a target="_blank" rel="ugc" href="https://civarchive.com/models/376031/hassaku-xl-pony">https://civarchive.com/models/376031/hassaku-xl-pony</a>Mikoshi [Pony]<a target="_blank" rel="ugc" href="https://civarchive.com/models/543369/mikoshi-pony">https://civarchive.com/models/543369/mikoshi-pony</a>————————————<h3 id="pony-wjdeuu3qm">注意！Pony底模相关：</h3>需要使用SDXL VAE。本人使用的VAE下载于PonyV6的附件：Pony Diffusion V6 XL<a target="_blank" rel="ugc" href="https://civarchive.com/models/257749/pony-diffusion-v6-xl">https://civarchive.com/models/257749/pony-diffusion-v6-xl</a>需要将CLIP终止层数设置为2。<h3 id="important!-regarding-pony-based-models:-orbl8w5lo">Important! Regarding Pony-based models:</h3>Requires SDXL VAE.The VAE I use is from the Pony V6 attachments:Pony Diffusion V6 XL<a target="_blank" rel="ugc" href="https://civarchive.com/models/257749/pony-diffusion-v6-xl">https://civarchive.com/models/257749/pony-diffusion-v6-xl</a>Set CLIP Skip to 2.————————————<h3 id="(recommended-parameters)-03uv3uj3z">推荐参数(Recommended Parameters)：</h3>采样方法(Sampler)：Eular aSteps： 24~40Width：1024 (768~1440 is fine)Hight：1024 (768~1440 is fine)CFG Scale：3~8————————————<h3 id="(hires.-fix)-(recommended-use)-9noiem6rk">推荐使用高分辨率修复(Hires. fix) (Recommended Use)：</h3>放大算法 (Upscaler)：R-ESRGAN 4x+ Anime6BSteps：4~15重绘幅度 (Denoising Strength)：0.3~0.4Hires Distilled CFG Scale：0Hires CFG Scale：3 (0~5 is fine)————————————由于本人是新手，对于学习率的把控并不好，因此这次吸取了过拟合的教训之后降低了一定的学习率，并且没有像之前一样将角色和服装全部分别训练进一个tag，而是像一般的LoRA一样保留了特征词，换取了更高的自由度，但额外的代价是服饰表现并不稳定。同时由于莉亚在兰斯10中的立绘过于单一，倾向于出现莉亚立绘中的常见的手势。但这次画风的训练更加成功，为未来训练更多兰斯10中的角色留下了基础。————————————案例：<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e24b3c32-8d5f-412f-9f77-82d77cd804b6/width=525/e24b3c32-8d5f-412f-9f77-82d77cd804b6.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/efbe3eaf-de46-4589-a0ba-d17006ece02b/width=525/efbe3eaf-de46-4589-a0ba-d17006ece02b.jpeg" />

Files with SHA256: 1a3a4d710f422ff9fe7c1c6e71517d57e623feb53e831e5079e7ba8ebf2177ff