📝 发布说明 (Description)
✨ 穿越云层,追寻那颗坠落的彗星。 本 LoRA 致力于完美复刻新海诚导演在《你的名字》中那令人惊叹的唯美光影、绚丽天空与标志性的色彩美学。
✨ Beyond the clouds, chasing the falling comet. This LoRA is dedicated to perfectly replicating the breathtaking cinematic lighting, gorgeous skies, and iconic color aesthetics from Makoto Shinkai's masterpiece, Your Name.
🔑 核心触发词 / Core Trigger Word: yourname style
🛠️ 硬核训练与极简打标 / Hardcore Training & Minimalist Tagging
降维打击的画质基底: 本次训练集堪称极致。我逐帧筛选了《你的名字》电影原片,最终收集了 1509 张最顶级的原画截帧。所有素材均来源于 4K 超清片源,并高质量压缩至 1024 分辨率进行深度训练,确保模型在学习宏大场景时,连一片云彩的边缘都清晰可见。
A 4K Down-Sampled Foundation: The dataset for this model is pushed to the extreme. I meticulously went through the original movie frame by frame and gathered 1,509 top-tier original screencaps. All source materials were extracted from a 4K resolution source and high-quality compressed to 1024 for deep training, ensuring the model captures everything down to the crisp edges of the clouds.
全新“暴力删标”逻辑,彻底解放提示词: 在打标阶段,我采用了完全不同于以往的策略——大量删除画面描述标签。 通过这种“暴力清洗”,我成功将新海诚那极具辨识度的灵魂画风(如:绝美云层、高反差明暗、丁达尔光束、唯美黄昏)强制绑定在了触发词 yourname style 上。
New "Aggressive Tag Pruning" Logic, Freeing Your Prompts: During the tagging phase, I adopted a completely different strategy—aggressively deleting descriptive tags. Through this "violent cleaning," I successfully baked Makoto Shinkai's highly recognizable soul (e.g., breathtaking cloudscapes, high-contrast lighting, Tyndall god rays, beautiful sunsets) directly into the trigger word yourname style.
💡 这意味着什么?(What does this mean for you?) 你不再需要堆砌许多诸如 beautiful sky, detailed clouds, cinematic lighting, lens flare... 这样繁琐的光影词汇。只需要极其简单的几句自然语言,加上触发词,模型就能瞬间为你召唤出原汁原味的新海诚世界!
You no longer need to stack hundreds of complex lighting tags like beautiful sky, detailed clouds, cinematic lighting, lens flare... Just write a few simple natural language descriptions along with the trigger word, and the model will instantly summon an authentic Makoto Shinkai world for you!
💡 推荐参数与使用建议 / Usage Tips
推荐权重 (Recommended Weight):
0.7 - 0.9(0.8 为最佳甜点位 / 0.8 is the sweet spot)提示词建议 (Prompting Advice): 保持提示词简短!把复杂的画风交给 LoRA,你只需要告诉它画面里有什么人、什么景即可。
Keep your prompts short! Let the LoRA handle the complex art style. You only need to describe who and what is in the scene.
📝 范例提示词 / Example Prompt:
yourname style, 1boy and 1girl, sitting on a bench in a park, sharing a drink, trees casting dappled shadows on the ground around them. Loose clothes, relaxed postures, casual conversation implied. Late afternoon, warm but not harsh light.
📸 求返图! / Share Your Art!
期待看到大家用极简的提示词,召唤出最震撼的新海诚天空!欢迎在评论区多多返图交流!
I'm looking forward to seeing you summon the most stunning Makoto Shinkai skies with minimalist prompts! Please share your generated images and creations in the reviews!
⚠️ 免责声明 / Disclaimer
本模型仅供 AI 绘画爱好者学习、交流与同好分享使用,严禁用于任何形式的商业用途。本模型旨在向新海诚(Makoto Shinkai)导演及其团队极致的美术风格致敬,无任何恶意侵权意图。《你的名字》的所有相关设计、美术风格及知识产权均归 新海诚 及 CoMix Wave Films 所有。使用者由此产生的一切商业纠纷或法律责任由使用者自行承担,模型作者概不负责。 This model is strictly for educational purposes, personal communication, and sharing among fans. Any form of commercial use is strictly prohibited. This model is created as a tribute to the exquisitely beautiful art style of Makoto Shinkai and his team, with absolutely no malicious intent to infringe upon copyrights. All designs, art styles, and related intellectual properties of Your Name belong to Makoto Shinkai and CoMix Wave Films. The creator of this model assumes no responsibility for any commercial disputes or legal issues arising from its use.
Description
FAQ
Comments (6)
i'll never forget this anime because a facebook post
guy1: what's the anime called?
guy2: your name
guy1: Andrew
感觉连亮色调的图都带有一种阴暗的基调。可能是数据集里暗色图太多了,又或者是因为标签没打好,导致 LoRA 把白天和黑夜的场景混在一起平均化了,出图就成了这种默认的折中风格?
我用缩略图看了一下训练集,电影整体色调偏暗,过于明亮的图片不多,白天黑夜配比大概6:4或者7:3
打标错误或者训练集白天黑夜的配比不均,不会导致融合变阴暗,而是在不提及明度的情况下,默认出黑夜。
所以最大的可能就是本身训练集中的白天的图不够明亮,学习到的画风,让ai认为,白天就是这个亮度。
有时间我再检查一下打标,看看有没有办法优化一下不
@570538629941 我觉得最好的方法是把亮色和暗色的图片分开存放到两个文件夹里,并增加亮色文件夹的循环次数(repeats),例如命名为 "3_light_scenario" 和 "1_dark_scenario"。同时,使用清晰的标签如 "day, daytime, lightrays, sunlight | dusk, dawn, evening | night, night sky, moonlight, stars, milkyway, galaxy" 并结合基于人工智能的基础自然语言描述,这样应该能让模型学到更清晰的训练意图和概念隔离
Very useful.
I use it on medium weights to get quality screencap look.












