This is a comprehensive LoRA trained on a wide variety of Azur Line shipgirls across different nations and ship types. It includes characters from all major factions like Sakura Empire, Eagle Union, Royal Navy, Iron Blood, and more — covering battleships, carriers, cruisers, destroyers, and submarines.这是一个碧蓝航线舰娘整合型 LoRA，涵盖了所有主要阵营（如重樱、白鹰、皇家、铁血等）以及各类舰种（战列舰、航母、巡洋舰、驱逐舰、潜艇）。模型经过多样图像训练，能够生成风格各异的舰娘角色，适用于创作原角或自设舰娘。## 📌 Training Info / 训练信息<ul><li>Base model: anyloraCheckpoint_bakedvaeBlessedFp16</li><li>Image resolution: 512 × 512 (with dynamic bucket: min 512, max 1216, step 64)</li><li>Latent caching: Enabled (cache_latents, saved to disk)</li><li>Batch size: 8</li><li>Epochs: 10 (saved every 1 epoch)</li><li>LoRA rank (dim): 64</li><li>Alpha: 32</li></ul>## 🧠 Prompt Guide / 提示词使用指南This LoRA uses refined trigger tokens to distinguish between factions (nations) and ship classes in Azur Lane. Use the following structure for best results:📌 Basic Format:&lt;lora:AZUR_LINE-000008:0.8&gt;, 1girl, AZUR LINE, [FACTION], [SHIP CLASS], [attributes...]<hr />🏳️ Faction Triggers / 阵营触发词<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/e258264c-fd82-4200-9cc2-78baca4cb5b7/width=525/e258264c-fd82-4200-9cc2-78baca4cb5b7.jpeg" /><hr />🚢 Ship Class Triggers / 舰种触发词<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0fb8e38b-2007-4001-9949-ad54650fe28a/width=525/0fb8e38b-2007-4001-9949-ad54650fe28a.jpeg" /><hr />Basic Standing &amp; Camera Angles站姿与视角<ul><li>standing</li><li>full body</li><li>looking at viewer</li><li>low angle</li><li>from below</li></ul>Casual / Soft Poses动作<ul><li>sitting</li><li>lying down</li><li>smiling</li><li>blushing</li><li>brushing hair</li><li>adjusting gloves</li></ul>🎯 Prompt Usage / 提示词示例<ul><li>Dragon Empery cruiser (东煌巡洋舰)</li><li>&lt;lora:AZUR_LINE-000008:0.8&gt;, PLAN, CA, AZUR LINE, 1girl, breasts, long hair, umbrella, thighhighs, black hair, gloves, rigging, dress, solo, large breasts, hair ornament, chinese clothes, oil-paper umbrella, china dress, very long hair, high heels, turret, flower, sitting, machinery, red eyes, hair flower, holding, looking at viewer, black gloves, white dress, thighs, holding umbrella, wide sleeves, dragon, white footwear, cannon, covered navel</li><li>Iron Blood submarine（铁血潜艇）</li><li>&lt;lora:AZUR_LINE-000008:0.8&gt;, KMS, SS, AZUR LINE, 1girl, torpedo, one-piece swimsuit, ahoge, closed eyes, pink hair, solo, swimsuit, rudder footwear, iron cross, jacket, riding, bubble, air bubble, hood, gloves, smile, hairband, yellow gloves, body writing, water, small breasts, hooded jacket, tattoo, cross, watercraft, anklet, hood down, underwater, full body, blush, blue one-piece swimsuit, closed mouth, headgear, double vertical stripe, competition swimsuit, two side up, long sleeves, hair ornament, white hairband, breasts, character name, ^_^, open jacket, open clothes, water drop, uneven sleeves, v-shaped eyebrows, jewelry, submerged, arched back, asymmetrical sleeves, thighs, red footwear, machinery, splashing, leg tattoo, sidelocks, clothes writing, brown jacket, black one-piece swimsuit, high heels, covered navel, brown background, yellow background, short hair, straddling</li><li>Iris Libre battleship（自由鸢尾战列舰）</li><li>&lt;lora:AZUR_LINE-000008:0.8&gt;, FFS, BB, AZUR LINE, 1girl, blue hair, breasts, long hair, dress, white dress, holding, very long hair, looking at viewer, large breasts, blue eyes, scythe, thighhighs, wings, sky, solo, boots, cloud, see-through, holding scythe, blue sky, full body, underboob, day, robot, armored boots, non-humanoid robot, rigging, weapon, gloves, water, black footwear, arm up, absurdly long hair, see-through dress, holding weapon, robot animal, blue ribbon, outdoors, halo</li></ul><hr />⚠️ NotesMy model performs poorly when generating CVs (aircraft carriers), as the AI struggles to learn and depict carrier-based aircraft accurately. However, the carrier flight decks actually turn out quite well.btw，我的模型在航母（CV）的绘制表现上很差，因为AI非常不善于如何学习并绘制舰载机。⚙️ Recommended Generation Settings / 生图参数推荐本 LoRA 与多种常见的 anime 风格大模型兼容良好。以下是推荐使用的基础模型（safetensors 格式）及对应的 LoRA 权重范围：<ul><li>AbyssOrangeMix2_sfw：推荐权重 0.8 ~ 0.9</li><li>cetusMix3.5：推荐权重 0.6 ~ 0.8</li><li>pastelmix-fp16：推荐权重 0.8 ~ 0.9</li><li>qteamixQ_omegaFp16：推荐权重低于 0.65</li><li>tmndMixVPruned：推荐权重 0.6 ~ 1.0</li><li>darkSushiMixMix_225D：推荐权重 0.6 ~ 1.2</li></ul>请根据你想要的画风进行微调。较低的权重会保留底模风格，较高的权重则会强化 LoRA 训练特征。同时建议使用以下通用参数进行生图：<ul><li>Prompt：建议包含 azur lane、阵营（如 USS、IJN）、舰种（如 CV、BB）以及 1girl 等关键词</li><li>Negative Prompt：worst quality, lowres, blurry, mutated hands, extra limbs</li><li>采样器（Sampler）：推荐使用 Eular a</li><li>采样步数（Sampling Steps）：建议 20 ~ 30 步</li><li>CFG Scale：建议设置为 6 ~ 8</li><li>尺寸（Size）：建议不超过 512 × 768，以避免图像崩坏</li><li>Clip Skip：2（模型训练时已启用）</li></ul>⚙️ Recommended Generation SettingsThis LoRA works well with several popular anime-style base models. Below are the recommended base models (in .safetensors format) and the ideal LoRA weight range for each:<ul><li>AbyssOrangeMix2_sfw: 0.8 ~ 0.9</li><li>cetusMix3.5: 0.6 ~ 0.8</li><li>pastelmix-fp16: 0.8 ~ 0.9</li><li>qteamixQ_omegaFp16: below 0.65</li><li>tmndMixVPruned: 0.6 ~ 1.0</li><li>darkSushiMixMix_225D: 0.6 ~ 1.2</li></ul>Adjust the LoRA weight according to your preferred balance: lower values retain more of the base model's original style, while higher values emphasize the traits learned from this LoRA.General generation settings for best results:<ul><li>Prompt: Include keywords like azur lane, faction (e.g. USS, IJN), ship class (e.g. CV, BB), and 1girl</li><li>Negative Prompt: worst quality, lowres, blurry, mutated hands, extra limbs</li><li>Sampler: Recommended Eular a</li><li>Sampling Steps: 20 to 30</li><li>CFG Scale: 6 to 8</li><li>Image Size: Up to 512 × 768 recommended to avoid broken results</li><li>Clip Skip: 2 (enabled during training)</li></ul>## 🖼️ Recommended Output Resolution / 出图尺寸This model was trained on 643 in-game screenshots with the following original resolutions:<ul><li>512×512</li><li>512×576</li><li>512×640</li><li>512×704</li><li>512×768</li></ul>Recommended output resolution: up to 512×768 Going beyond this may result in distorted or broken outputs.Hires.fixRecommendations:<ul><li>Upscaling algorithms: 4x-AnimeSharp or R-ESRGAN 4x+ Anime6B</li><li>Denoising strength: 0.5 ~ 0.6</li><li>Enable: ControlNet Tile for better consistency</li></ul>## 📎 Notes / 附注说明CUC AI STUDIO 出品/Produced by CUC AI STUDIO- 本模型仅供同人创作使用- 欢迎评论区分享你的出图作品，舰娘等你召唤！Feel free to use this LoRA for your creative works — we’d love to see what you come up with! 💛 This model is for personal study and non-commercial use only. Commercial use is strictly prohibited.

Files with SHA256: 5b5ad98588904a7e5b51b8865e8a4fe83f21ca844fa720a3b2255be9d8431d7d