### RunningHub Online Workflow / RunningHub 在线工作流
Ready-to-use workflow (CN): https://www.runninghub.cn/post/2070419601577570306/?inviteCode=rh-v0990
Ready-to-use workflow (Overseas): https://www.runninghub.ai/post/2070419601577570306/?inviteCode=rh-v0990
Fan benefit: use invite code rh-v0990 to claim 1000 RH coins, plus 100 RH coins every daily login. If the coins do not appear automatically, click the avatar in the upper right and bind/activate invite code rh-v0990.
国内站工作流:https://www.runninghub.cn/post/2070419601577570306/?inviteCode=rh-v0990
海外站工作流:https://www.runninghub.ai/post/2070419601577570306/?inviteCode=rh-v0990
粉丝福利:点击右上角头像 - 邀请码,填入 rh-v0990,可领取 1000 RH 币;每天登录还可领 100 RH 币。若没有自动到账,请在右上角绑定/激活邀请码 rh-v0990。
### Training Notes The source material for this model was processed with Tutu's Image & Video Prompt Reverse Tool. For clothing LoRA training, the captioning logic is not to describe the target outfit in excessive detail. It is the opposite: the target outfit itself should be described as little as possible, while pose, scene, camera angle, lighting, face, background, and other non-target variables should be separated clearly. This helps the LoRA learn the outfit as the subject feature instead of treating the written clothing description as something that must be repeated every time. The training was completed with Tutu Trainer. Tutu Trainer can handle LoRA training, model resource management, parameter presets, and training checks. It does not require manual environment setup or repeated parameter tuning, so training can start quickly while still keeping strong results. It uses the proprietary Tutu Timebook algorithm. In my own tests, under similar conditions, it can produce better LoRA results than most other trainers I have tried. It is free to use, and you are welcome to test it. ### Official Website And Contact Official website: https://zhaotutu.xyz/ Tutu Super Model Trainer: https://zhaotutu.xyz/downloads/tututrainer/ Image & Video Prompt Reverse Tool: https://zhaotutu.xyz/downloads/tutuannotator/ Bilibili: https://space.bilibili.com/431046154 YouTube: https://www.youtube.com/@zhaotutu/videos Telegram: https://t.me/zhaotutu TikTok: https://www.tiktok.com/@zhaotututu GitHub: https://github.com/zhaotututu Email: [email protected] QQ: 331506796 ### Model Introduction This is a long qipao-style lace dress LoRA for Z-Image-Turbo. The outfit direction is a black lace long dress with a sleeveless halter cut and traditional Chinese knot buttons at the neckline. It has a modern fashion editorial feeling while still keeping a clear qipao-inspired silhouette. The model was trained for the outfit concept rather than for a fixed face or a fixed person. In actual tests, when the LoRA weight is around 0.75, the long dress shape, black lace surface, vine-like decorative pattern, halter neckline, and black Chinese knot button structure can all appear quite stably. The outfit can be used in full-body shots, upper-body portraits, seated poses, half-body compositions, and material close-ups. This model has relatively low influence on unrelated image elements. You can change the environment, background, lighting, camera distance, indoor space, street scene, exhibition hall, stage, rainy night, snow scene, or other visual conditions, while the main outfit concept can still remain recognizable. In testing, shorter outfit prompts were more stable than overly long clothing descriptions. ### Model Info Base model: Z-Image-Turbo / ZImageTurbo Type: LoRA Trigger word: tutulianyiqun Recommended weight: 0.75 Loading method: use a Z-Image-Turbo compatible workflow and load it with a normal LoRA loader. ### Recommended Prompt ```text tutulianyiqun. She wears a unique long qipao-style dress. The main body of the dress is made of black lace fabric with delicate and intricate vine patterns. The dress uses a sleeveless halter-neck cut, and the neckline is decorated with traditional black Chinese knot buttons. A 25-year-old adult woman with a modern fashion editorial look, long black hair naturally falling over her shoulders, complete clear face, delicate soft facial features, healthy translucent skin, clean modern makeup, calm confident eyes, relaxed natural expression, realistic and detailed image.
Full-body supplement:
from the top of her hair to her feet fully visible in frame, wearing black patent pointed stiletto heels, glossy shoe surface, shoe tips, thin heels, and the ground under her feet clearly visible训练说明

素材使用图图的图片视频提示词反推器进行处理。对于服装 LoRA 训练来说,标注逻辑不是把目标服装细节写得越满越好,而是反过来:目标服装本身要尽量少描述,把姿态、环境、镜头、光照、脸部、背景等非目标变量区分清楚。这样模型才会把服装当成需要学习的主体特征,而不是把文本里的服装描述当成每次都必须复述的条件。
训练使用图图训练器完成。图图训练器可以用于 LoRA 训练、模型资源管理、参数预设和训练检查。不需要手动配置环境,也不需要反复调整复杂参数,就可以快速开始训练,并尽量获得稳定的训练结果。它采用独家的图图时间簿算法,我自己的实测里,在接近条件下,它的 LoRA 训练效果优于我试过的大多数训练器。工具完全免费,欢迎尝试。
官网与联系方式
图图的超级模型训练器:https://zhaotutu.xyz/downloads/tututrainer/
图片视频提示词反推器:https://zhaotutu.xyz/downloads/tutuannotator/
Bilibili:https://space.bilibili.com/431046154
YouTube:https://www.youtube.com/@zhaotutu/videos
Telegram:https://t.me/zhaotutu
TikTok:https://www.tiktok.com/@zhaotututu
GitHub:https://github.com/zhaotututu
QQ:331506796
模型介绍
这是一个用于 Z-Image-Turbo 的长款旗袍风蕾丝连衣裙 LoRA。服装方向是黑色蕾丝长裙、无袖挂脖剪裁、领口黑色中式盘扣,整体有现代时尚杂志感,同时保留明确的旗袍风轮廓。
这个模型训练的是服装概念,不是固定脸,也不是固定人物。实际测试里,LoRA 权重在 0.75 左右时,长裙轮廓、黑色蕾丝面料、蔓藤花纹、挂脖领口和黑色中式盘扣结构都能比较稳定地出现。它可以用于全身图、上半身肖像、坐姿、半身构图和材质近景。
这个模型对无关元素的影响相对较低。你可以更换环境、背景、光照、镜头距离、室内空间、街景、展厅、舞台、雨夜、雪景等画面条件,服装概念仍然能保持较高识别度。测试中发现,短版服装提示词比过长的服装描述更稳定。
模型信息
基础模型:Z-Image-Turbo / ZImageTurbo
类型:LoRA
触发词:tutulianyiqun
正常推荐权重:0.75
加载方式:使用兼容 Z-Image-Turbo 的工作流,用普通 LoRA 加载器加载即可。
建议提示词
tutulianyiqun, 身着一件设计独特的长款旗袍风格连衣裙,该连衣裙主体采用黑色蕾丝面料,饰有精致繁复的蔓藤花纹,裙装采用无袖挂脖式剪裁,领口位置饰有传统的黑色中式盘扣,一位25岁左右的成年女性,年轻现代的时尚杂志模特气质,黑色长发自然披散,脸部完整清晰,五官精致柔和,皮肤状态健康通透,清透现代淡妆,眼神平静自信,表情自然放松,画面真实细腻完整全身图补充:
头顶发丝到脚下完整进入画面,脚穿黑色漆皮尖头细跟高跟鞋,鞋面有清晰反光,鞋尖、细跟和脚下地面完整可见学習について

素材は Tutu の画像・動画プロンプト逆推定ツールで処理しました。衣装 LoRA の学習では、対象となる衣装の細部をできるだけ多く書くことが目的ではありません。むしろ逆で、対象衣装そのものの説明はできるだけ少なくし、ポーズ、環境、カメラ、ライティング、顔、背景など、衣装以外の変数を分けて記述します。これにより、LoRA が衣装を学習対象の主体特徴として覚えやすくなります。
学習は Tutu Trainer で行いました。Tutu Trainer は LoRA 学習、モデルリソース管理、パラメータプリセット、学習チェックに使えるツールです。手動の環境構築や複雑なパラメータ調整をしなくても、すぐに学習を始められ、安定した結果を目指せます。独自の Tutu Timebook アルゴリズムを採用しており、私の実測では、近い条件下で試した多くのトレーナーより良い LoRA 結果が出ています。無料で使えるので、ぜひ試してみてください。
公式サイトと連絡先
公式サイト:https://zhaotutu.xyz/
Tutu Super Model Trainer:https://zhaotutu.xyz/downloads/tututrainer/
画像・動画プロンプト逆推定ツール:https://zhaotutu.xyz/downloads/tutuannotator/
Bilibili:https://space.bilibili.com/431046154
YouTube:https://www.youtube.com/@zhaotutu/videos
Telegram:https://t.me/zhaotutu
TikTok:https://www.tiktok.com/@zhaotututu
GitHub:https://github.com/zhaotututu
QQ:331506796
モデル紹介
これは Z-Image-Turbo 向けのロングチャイナ風レースドレス LoRA です。衣装の方向性は、黒いレースのロングドレス、ノースリーブのホルターネック、襟元の黒いチャイナボタンです。現代的なファッション誌風の雰囲気を持ちながら、チャイナドレス風のシルエットも分かりやすく残ります。
このモデルは、固定の顔や特定人物ではなく、衣装コンセプトを学習したモデルです。実際のテストでは、LoRA weight を 0.75 前後にしたとき、ロングドレスの形、黒いレース生地、蔓のような装飾模様、ホルターネック、黒いチャイナボタンの構造が比較的安定して出ます。全身、上半身ポートレート、座りポーズ、半身構図、素材が分かる近景にも使えます。
このモデルは、無関係な要素への影響が比較的少ないです。環境、背景、ライティング、カメラ距離、室内空間、街中、展示室、ステージ、雨の夜、雪景色などを変えても、衣装コンセプトは認識しやすい状態で残ります。テストでは、長すぎる衣装説明より、短い衣装プロンプトのほうが安定しました。
モデル情報
Base model: Z-Image-Turbo / ZImageTurbo
Type: LoRA
Trigger word: tutulianyiqun
通常推奨 weight:0.75
読み込み方法:Z-Image-Turbo 互換ワークフローで、通常の LoRA loader を使って読み込んでください。
推奨プロンプト
tutulianyiqun. She wears a unique long qipao-style dress. The main body of the dress is made of black lace fabric with delicate and intricate vine patterns. The dress uses a sleeveless halter-neck cut, and the neckline is decorated with traditional black Chinese knot buttons. A 25-year-old adult woman with a modern fashion editorial look, long black hair naturally falling over her shoulders, complete clear face, delicate soft facial features, healthy translucent skin, clean modern makeup, calm confident eyes, relaxed natural expression, realistic and detailed image.全身画像用の追加:
from the top of her hair to her feet fully visible in frame, wearing black patent pointed stiletto heels, glossy shoe surface, shoe tips, thin heels, and the ground under her feet clearly visible
C站如果有单独的“Version description / About this version”,填这个:
```text
V1.0 release. Recommended weight: 0.75. Trigger word: tutulianyiqun. Use with a Z-Image-Turbo compatible workflow and a normal LoRA loader.
V1.0 版本。推荐权重:0.75。触发词:tutulianyiqun。请使用兼容 Z-Image-Turbo 的工作流,并通过普通 LoRA 加载器加载。
V1.0 リリース。推奨 weight:0.75。Trigger word:tutulianyiqun。Z-Image-Turbo 互換ワークフローで、通常の LoRA loader を使ってください。








