角色 Character

云霆（Yunting / Lightning Yun）是本人的原创角色（OC）：一只成年肌肉兽人狼，黑白双色毛发，身上有橙色涡纹与斑纹，绿眼睛，橙色肉垫。

Yunting is my original character (OC) — an adult muscular anthro wolf with black-and-white two-tone fur, orange swirl markings, green eyes and orange paw pads.

版本 Versions（请先读这里）

v1 是早期试点（只有 8 张训练图），花纹与配色还原度明显偏低，仅作存档；请从 v2 / v3 / v4 里选。只有 v1 → v2 是明确的提升。

v2、v3、v4 是三位不同的训练者各自独立训练出来的三个模型，数据集、底模、标注方式都不一样。它们之间是“效果取向的差别”，不是代际升级；v3 并不因为版本号大就比 v2 好。建议三个都试一次，看你自己的底模和画风跟哪个更合。

v1, v2, v3 and v4 are not a single iteration line: v2 / v3 / v4 are three independent training runs by three different trainers on the same character. They differ in result, not in generation. Try all three with your own checkpoint.

版本一览 Versions

Yunting v1 (Illustrious, 10ep / 15ep) — STBZ，8 图试点，存档用
Yunting v1 (Pony) — STBZ，只能配 Pony Diffusion V6 底模
Yunting v2 (Illustrious) — STBZ，27 图，第一版真正还原花纹与配色的模型，dim 32 / alpha 32
Yunting v3 (Illustrious) — 无铭，39 图，底模 WAI-NSFW-Illustrious v14，dim 32 / alpha 16，epoch 8
Yunting v4 (Illustrious) — Not Saint Dog，26 图，底模 Illustrious-XL v0.1，dim 32 / alpha 16，epoch 15，训练集里带了颜色/斑纹标签

触发词 Trigger Words

所有版本通用：yunting（v4 的示例提示词里写成 Yunting，两者等价）

v2

yunting, furry male, dog boy, white fur, green eyes, muscular, bara, yellow knee

注意：v2 的训练集把膝部橙色花纹标成了 “yellow knee”，如果膝盖偏黄，请在提示词里补 orange markings。

v3

yunting, wolf boy, furry male, two-tone fur, white hair, multicolored fur, colored sclera, bara

v3 的训练集里没有 orange knee / black ankel / white hands / white feet 这些词，描述配色请用 two-tone fur + multicolored fur。

v4

yunting, orange knee, black ankel, white hands, white feet, white fur, black fur, orange fur, green eyes, canid, canine

注意：“black ankel” 是训练集里的原始拼写，请照用；改成 “black ankle” 不会生效。

使用建议 How to use

权重：从 0.8 开始，常用区间 0.6–1.0（v2 可到 1.2）
底模：v1 (Pony) 只用 Pony Diffusion V6；其余版本用 Illustrious 系（Illustrious-XL / WAI-NSFW-Illustrious / NoobAI），不要用在 SDXL 1.0 上
采样：Euler a 或 DPM++ 2M Karras，25–32 steps，CFG 5–7，1024×1024 以上
与其他角色 LoRA 叠加时，把本模型降到 0.6–0.7，避免花纹被冲淡
负面提示词：常规 Illustrious 负面 + AS-Adult-Neg（本角色为成年设定）
各版本示例图的提示词已隐藏，但底模、采样器、seed 等参数保留，可参考复现

授权与署名 License & Credits

角色“云霆”的形象设定与美术版权归本人（LightningYun）所有。

模型权重按 Illustrious License（FAIPL-1.0-SD）发布，这是 Illustrious 系底模要求的许可证：它允许使用、修改与再分发，并且不允许我再附加更严格的条款。所以下面三条是请求，不是可强制的许可条款：

请不要把模型文件本身拿去售卖或打包分发；
请不要在未联系本人的情况下把本页内容搬运到其他平台；
发布作品时请保留角色出处链接。

硬性要求（这一条是条款）：严禁将本模型用于任何涉及未成年人、或幼态外观角色的内容。本模型只为成年角色训练，页面上也不会收录此类示例图。

The character design of Yunting belongs to me (LightningYun). The weights are released under the Illustrious License (FAIPL-1.0-SD) required by the Illustrious base models; that license allows use, modification and redistribution and does not let me add stricter terms, so the three points above are requests, not enforceable clauses. Hard rule: this model may not be used for any content involving minors or minor-coded characters.

训练署名 Credits：v1 与 v2 = STBZ（本人委托 commissioned）；v3 = 无铭（tyj3906）；v4 = Not Saint Dog。

反馈 Feedback：评论区，或我的 Discord。

Description

Base model 底模: illustriousXLPersonalMerge_v30Noob10based (Illustrious-family personal merge checkpoint)

Trained by 训练: STBZ — commissioned by me / 本人委托

Training data 数据: 27 images x5 repeats | dim 32, alpha 32 | Adafactor | 512 buckets | clip skip 1

Trigger 触发词: yunting

Recommended weight 建议权重: 0.8-1.2 (default 1.0)

Files 文件说明 (2): Yuntingv2_illuspersonal.safetensors = epoch 20 / 540 steps（主文件）| Yuntingv2_illuspersonal-000015.safetensors = epoch 15 / 405 steps（本页样图全部出自这一版）。两个权重都过了 pickle / 病毒扫描，哈希不同属正常。

This run saved a checkpoint at every epoch from 10 to 20 (11 files, 170 MB each). They are deliberately NOT all uploaded here: the download list would show eleven near-identical files with nothing but the filename to tell them apart, and Civitai treats multiple files as precision variants of the same model, so the "preferred variant" auto-pick would become meaningless. If you need a specific epoch, ask in the comments. / 这轮训练存了 epoch 10–20 共 11 个权重，故意没全部上传；需要哪个 epoch 请在评论区说。

Notes 说明: First version that reliably reproduces the two-tone fur and the orange swirl markings - a clear step up from v1. The knee markings are captioned "yellow knee" in this training set, so the knees can come out yellow; add "orange markings" to the prompt if that happens.

IMPORTANT: v2, v3 and v4 are three INDEPENDENT training runs by three different trainers (STBZ / 无铭 / Not Saint Dog). v3 and v4 came later in TIME, not later in quality - different dataset, different base checkpoint, different captioning. Try all three and keep the one that suits your checkpoint and art style.

重要：v2 / v3 / v4 是三位训练者各自独立训练的平行模型，v3、v4 只是时间上靠后，不是代际提升；只有 v1 -> v2 是明确的提升。