Arcane & League of Legends (LoL) Artstyle (双城之战/英雄联盟风格) | 全角色收容 (All Characters Included) | Fortiche Production

📌 请务必花两分钟看完简介与说明，以获得最佳的使用体验！(Please take two minutes to read the description and guide for the best experience!)[CN] 是的，这是一个风格 LoRA，但在某种意义上，它也算是一个“全家桶”？一个更比十个强？哈哈哈，开个玩笑啦！不过还是请您至少耐心看完下方的使用说明哦~~~[EN] Yes, this is an art style LoRA, but in a way, it's also a massive "All-in-One" character package.Buy one, get ten free? Hahaha, just kidding!But please do at least skim through the usage instructions below~~~ 💬 作者碎碎念 (A Note from the Creator)[CN] 顺便提一句，本人的提示词功底相对薄弱，上方展示的所有例图，仅仅是为了验证本 LoRA 的基础泛化功能与画面质感。为了展现模型最原始、最纯粹的状态，所有的测试图除了本 LoRA 之外，均未加载任何其他的人物或画风 LoRA，也未在提示词中引用任何特定的画师风格。更多的进阶玩法与绝美构图，就全权交由各位“艺术家”们自行开发与体验了。我非常期待能在评论区看到大家的惊艳返图！[EN] As a quick side note, my prompting skills are relatively basic. All the showcase images provided above are purely to verify the baseline functionality, generalization, and visual texture of this LoRA.To demonstrate the model in its most raw and unadulterated state, no other character or style LoRAs were loaded for these test images, nor were any specific artist names referenced in the prompts.I leave the exploration of advanced techniques and breathtaking compositions entirely in the capable hands of you brilliant artists. I absolutely cannot wait to see the stunning artwork you share in the reviews! 📖 简介 (About This LoRA)[CN] 极致的 Fortiche 工业美学，献给所有沉醉于底城霓虹与皮城荣光的创作者。本 LoRA 旨在尽可能复现《双城之战》（第一部与第二部）中极具张力的 2.5D 厚涂与令人窒息的电影级光影美学。无论你是想为心爱的角色重塑硬核的蒸汽朋克质感，还是想生成史诗般的宏大场景，这个模型都能为你提供降维打击般的视觉体验。为了保证画风的绝对纯净度与质感，本模型的训练集 100% 来源于原剧的高清精选剧照，未掺杂任何同人或杂图。强烈推荐使用横图（Landscape）进行生成，以获得最佳的电影构图体验！[EN] The ultimate Fortiche industrial aesthetics, dedicated to all creators captivated by the neon lights of Zaun and the glory of Piltover.This LoRA is a humble attempt to replicate the highly dynamic 2.5D thick-painting art style and breathtaking cinematic lighting of Arcane (Seasons 1 & 2).Whether you want to clothe your favorite characters in a hardcore steampunk texture or generate epic, monumental scenes, this model delivers a visually stunning experience.To ensure absolute stylistic purity, the training dataset consists 100% of carefully curated high-resolution screenshots from the original series, with zero fanart or mixed styles. It is highly recommended to use landscape orientation for generation to achieve the best cinematic composition! ⚙️ 核心使用说明 (Usage Guide)全局画风触发词 (Global Trigger): arcane artstyle[CN] 强烈建议置于提示词最开头。[EN] Highly recommended to place at the very beginning of the prompt.人物/场景触发词 (Secondary Triggers):[CN] 请见下方各角色与场景的专属英文触发词。[EN] Please refer to the exclusive English trigger words for each character/scene below.推荐画幅 (Aspect Ratio):[CN] 强烈推荐横图（如 16:9, 3:2 等），以完美契合大片级别的电影构图。[EN] Horizontal aspect ratios (e.g., 16:9, 3:2) are strongly recommended to perfectly match the cinematic framing.✨ 稳定触发人物咒语 (Character Prompts)💡 提示与用法 (Tips & Usage)[CN] 提示： 角色名旁括号内的英文即为该角色的“二级触发词 (Secondary Triggers)”。下方列出的角色提示词已经过精简，仅保留了最核心的外貌特征。您直接复制粘贴即可精准“召唤”该角色，并且可以根据自己的创作需求，在提示词末尾自由添加动作描述或更换背景环境。当然，如果您需要一个更完善的起手式，也非常欢迎直接复制我上方展示的例图中的完整提示词，来进行修改和“抽卡”。[EN] Note: The English text in parentheses next to the character's name serves as their specific "Secondary Trigger".The prompt snippets listed below have been streamlined to include only their core visual features. You can copy and paste them directly to accurately "summon" the character. Feel free to append any action descriptions or change the background settings at the end of the prompt according to your creative needs.Alternatively, if you'd like a solid starting point, you are more than welcome to copy the full prompts from my uploaded showcase images and modify them as you see fit! 爆爆 (Powder): Powder, young child with short blue hair and freckles, wearing a green shirt, indoor blurry background.爆爆 - IF线 (DreamPowder): DreamPowder, young girl with blue hair and freckles, wearing a choker and necklace, standing outdoors near a tree, portrait.金克丝 (Jinx): Jinx, young woman with long braided blue hair, freckles, shoulder tattoo, brown choker, blurry background.蔚奥莱 - 幼年 (Violet): Violet, young woman with short pink and red hair, wearing an open red jacket over a brown vest, outdoors during the day.蔚 (ArcaneVi): ArcaneVi, woman with short asymmetrical red hair, facial tattoos, piercings, wearing a red jacket, blurry brown background.凯特琳 (Caitlyn): Caitlyn, woman with long black hair, wearing a blue jacket, beret, and white ascot, black background.梅尔 (MelMedarda): MelMedarda, dark-skinned woman with black dreadlocks bun, glowing facial tattoo, gold armor and jewelry, green background.安蓓撒 (Ambessa): Ambessa, dark-skinned woman with short black dreadlocks, wearing gold armor with shoulder plates, portrait shot.希尔科 (ArcaneSilco): ArcaneSilco, man with short slicked black hair, heterochromia red eye, facial scar, wearing a red shirt and jacket, black background.塞薇卡 (Sevika): Sevika, dark-skinned woman with short black hair, mechanical prosthetic arm, wearing a jacket, graffiti alley at night.杰斯 (Jayce): Jayce, mature man with short black hair and facial hair, wearing a white jacket over a black shirt, black background.维克托 (Viktor): Viktor, dark-skinned man with short brown hair, wearing a brown collared shirt and vest, indoor setting.黑默丁格 (Heimerdinger): Heimerdinger, yordle male with pointy ears and blonde hair, wearing a blue jacket, indoor brown background.辛吉德 (ArcaneSinged): ArcaneSinged, bald man with facial scar and colored sclera, wearing a scarf, outdoors with blurry green background.艾克 - 野火帮首领 (FirelightEkko): FirelightEkko, dark-skinned man with dreadlocks, wearing a brown jacket, cinematic lighting, portrait.艾克 - IF线 (DreamEkko): DreamEkko, dark-skinned man with white dreadlocks tied up, wearing a green jacket over a white shirt, blurry background.🎬 场景与特效指令 (Environments & VFX)[CN] 提示： 已提炼最核心的视觉标签，可直接搭配全局画风词使用。[EN] Note: Core visual tags have been extracted. Use them directly alongside the global style trigger.祖安底城 (Zaun Scenery): Zaun_scenery, cinematic night scene, weathered iron bridge, dense industrial buildings, flickering neon lights, wet cobblestone street, toxic emerald smog.皮尔特沃夫 (Piltover Scenery): Piltover cityscape, outdoors during the day, large airship flying above fantasy-style buildings and towers, bright blue sky, no human.双城爆炸特效 (Explosion VFX): arcane vfx, explosion, no human.[CN] 温馨提示：本 LoRA 的 2600+ 张训练集全部提取自电影级原生宽银幕素材。虽然上方展示的例图为了适配网页排版全部采用了竖图，但本模型在生成【横图（Landscape/Cinematic aspect ratio）】时会表现得更加游刃有余、质感拉满。强烈推荐各位艺术家尝试横图构图！[EN] Pro-tip: This LoRA was trained on over 2,600 frames of native, cinematic widescreen material. While the showcase gallery uses portrait images optimized for web scrolling, this model is inherently more comfortable and visually stunning when rendering in [Widescreen/Landscape aspect ratios]. I highly encourage you brilliant artists to unleash its full potential in cinematic widescreen formats! 🧪 炼丹纪实与理论探索 (Training Insights & Theoretical Attempts)[CN] 万里挑一的地狱级收集： 这次《双城之战》模型的炼制，对我而言是一场极度消耗精力的持久战。初始素材来源于两部剧集中每两秒一次的自动截图，总计 20,000 多张原始帧。我从中手动初筛出 5,000 张进行细致分类，最终以极其苛刻的标准，精选出 2,639 张顶级剧照作为核心训练集。说句心里话，如果没有极大的决心和耐心，请千万不要轻易尝试这种大型训练集的精细化收集与打标。我必须承认，我严重低估了构建《双城之战》训练集的工作量。它的收集与分类难度远超我之前处理的《紫罗兰永恒花园》。这不仅是因为剧中出场且有名字的角色繁多，更因为皮尔特沃夫与祖安在建筑风格和画面色调上的巨大差异，需要进行极其繁琐的细分。最让人崩溃的是，第一部的整体画面风格偏暗。在电脑的文件夹缩略图里，根本看不清图片的内容与细节，很大一部分图片必须单张点开放大才能辨认，分类工作堪称地狱级考验。当我意识到这个“天坑”的时候，我已经分好几百张图了——总不能半途而废吧！最后完全是靠着“来都来了，干都干了”的执念，硬生生熬完了整个训练集的构建。单单是前期的收集与分类，实际工作时长就远超 40 个小时。直到昨天，模型才被炼制出来，并且进行了测试，挑选出发挥比较稳定的其中一个。此外，为了防止最终出图“偏科”、导致模型生成的画面永远是一团死黑，我又咬紧牙关，补充收集了画面光影稍亮一些的第二部剧照，来平衡整体的数据分布。[EN] A Hellish, One-in-a-Million Curation Process: Creating this Arcane model was an absolute marathon. The initial source material came from automated screenshots taken every two seconds across both seasons, yielding over 20,000 raw frames. I manually filtered this down to 5,000 for detailed categorization, and finally, with exceedingly strict standards, handpicked a core dataset of 2,639 top-tier cinematic shots.To be completely honest, unless you have immense determination and patience, I strongly advise against attempting the meticulous collection and tagging of a large-scale dataset like this. I must admit, I severely underestimated the workload. Building the Arcane dataset was exponentially harder than my previous work on Violet Evergarden. Not only is the cast of named characters much larger, but the stark architectural and visual contrasts between Piltover and Zaun required incredibly tedious sub-categorization.The most soul-crushing part? Season 1's notoriously dark art style. In standard folder thumbnails, it was practically impossible to make out the contents or details. A huge portion of the images had to be opened and zoomed in individually just to identify them, turning the sorting process into a hellish ordeal.By the time I realized the sheer scale of this nightmare, I had already sorted hundreds of images. I couldn't just give up halfway! Ultimately, it was pure sunk-cost stubbornness—the "I'm already in too deep, might as well finish it" mentality—that forced me to power through.The initial gathering and sorting phase alone consumed well over 40 hours of actual labor. It was not until yesterday that the model was trained and tested, and one that performed consistently was selected.Furthermore, to prevent the model from overfitting to dark tones and constantly producing muddy, pitch-black generations, I bit the bullet and gathered brighter shots from Season 2 to balance out the overall lighting data.🎯 针对“二级触发词”的底层验证与创新尝试 (Validating "Secondary Triggers" & Innovative Attempts)[CN] 一次“画风与人物融合”的创新探索： 我每次炼制 LoRA 时，都会尽量给自己找一点“创新点”。这次《双城》模型的核心探索，就是验证在庞大的全局画风触发词统摄下，能否通过“多重二级触发词”来稳定召唤特定人物的不同形态。 从底层逻辑来看，在 AI 的视野里并没有绝对的“画风 LoRA”与“人物 LoRA”之分。当我们用海量图片绑定一个全局词时，AI 学习到的是共性的笔触和光影；而当我们在此基础上，为特定人物打上专属标签时，AI 理论上就能在全局画风的框架内提取并记住这个人物。这就是在画风库里“顺便”练出人物的基础。[EN] An Innovative Exploration of Merging Style and Characters: Every time I train a LoRA, I try to find a little "innovation point" for myself.The core exploration of this Arcane model was to verify whether specific characters in their various forms could be stably summoned using "multiple secondary trigger words" under a massive global style trigger. From an underlying logic perspective, AI doesn't strictly distinguish between a "Style LoRA" and a "Character LoRA."When we bind massive amounts o