Tutu’s sexy uniform (lace semi-transparent cheongsam)/图图的情趣制服（蕾丝半透旗袍）

Created with Tutu AI Workflow

Tagged with Tutu Annotator. Trained with Tutu Trainer.

This LoRA is part of my Tutu AI workflow. Tutu Trainer is used for LoRA training, model resource management, parameter presets, and training checks. Tutu Annotator is used for dataset annotation, prompt reverse engineering, English-friendly tags, and cleaner searchable captions.

For clothing LoRAs, annotation quality decides whether the model learns the actual outfit or pollutes the character, pose, background, and other clothing. This model is a good example: the goal is a lace semi-transparent cheongsam, not a general style filter.

• Official website: https://zhaotutu.xyz/
• Tutu Trainer: https://zhaotutu.xyz/downloads/tututrainer/
• Tutu Annotator: 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
• Reddit: https://www.reddit.com/user/zhaotututu/
• GitHub: https://github.com/zhaotututu
• Email: [email protected]
• QQ: 331506796

Tutu Sexy Uniform: Lace Semi-Transparent Cheongsam

This is the first LoRA in my Tutu sexy uniform / clothing series. It focuses on a lace semi-transparent cheongsam, trained from real photos and retouched material.

I made this model partly as an experiment: I wanted to see how well the difference extraction workflow could handle clothing. The practical result was better than expected. Compared with some hosiery LoRAs, this clothing LoRA has relatively high outfit restoration and less pollution to unrelated image content.

• Type: lace semi-transparent cheongsam / clothing LoRA
• Base model: SD 1.5
• Trigger words: tutututu, see-through, cheongsam
• Recommended start: tutututu, see-through, cheongsam
• Recommended CLIP skip: 2
• CFG Scale: start around 7

Design direction

• Lace semi-transparent cheongsam as the main outfit.
• More realistic clothing restoration instead of a loose fantasy costume style.
• Lower pollution to the character and the whole image compared with many clothing LoRAs.
• A practical start for the later Tutu clothing / uniform series.
• Unified future clothing trigger habit: tutututu, four "tu".

Training notes

The outfit prototype comes from a real clothing reference. The training material was shot from real photos and retouched, with 1024x1024 size and bf16 training details.

This was my first attempt at this specific outfit. Lace has many tiny details, and SD 1.5 resolution is still limited, so very fine lace may not always be perfectly restored. Even so, the model is useful because the outfit direction is clear and the overall pollution is low.

I also noticed that many similar clothing LoRAs are either not close enough to real clothing, do not cover current popular outfits well, or affect the character and the entire image too strongly. That is why I plan to continue building a series of clothing LoRAs.

Recommended settings

• Make sure the LoRA is loaded correctly.
• Use tutututu, see-through, and cheongsam.
• Recommended start: tutututu, see-through, cheongsam.
• Recommended CLIP skip / CLIP stop layers: 2.
• Start CFG Scale around 7.
• Do not add too many negative prompts, or the LoRA effect may be weakened.
• If you do not describe other clothing or the environment, the LoRA may affect them. Add clear clothing and background descriptions to reduce that effect.
• I did not use LoRA Block Weight in the original test, but you can try it if you need local control.

Prompt keywords

Main trigger group:

tutututu, see-through, cheongsam

• lace semi-transparent cheongsam
• see-through cheongsam
• lace cheongsam dress
• semi-transparent lace fabric
• realistic clothing texture
• cheongsam collar
• delicate lace pattern

If the outfit is not clear enough, strengthen lace semi-transparent cheongsam and see-through cheongsam. If the LoRA changes the background or other clothing, lower the LoRA weight and describe those elements more specifically.

Model compatibility

In my tests, Moyou-style and Maiju-style base models worked relatively well, and this model seems more friendly to realistic models. Different base models still need parameter adjustment.

FAQ

Why is the lace detail not sharp enough?

Lace is a very high-frequency detail, and SD 1.5 resolution has limits. Add delicate lace pattern and use a base model that handles clothing details well.

Why does the LoRA affect other clothes or the background?

Describe the other clothing and the environment clearly. If the influence is still too strong, lower the LoRA weight.

What should I do if the image shows mosaic or broken texture after increasing weight?

Lower this LoRA weight first. If several LoRAs are active, lower the other LoRA weights too, or try LoRA Block Weight for local control.

Responsible use

• You are fully responsible for any creative work made with this model.
• Do not use this model to create or share illegal, harmful, malicious, defamatory, fraudulent, or political content.
• If you use this model commercially or need custom LoRA cooperation, please contact me first. Thank you.

中文说明

这是图图情趣制服 / 服装系列里的第一个 LoRA，主题是 蕾丝半透明旗袍。素材来自实拍并经过修图处理。

我做这个模型，一部分原因是想验证差异提取流程做服装到底效果如何。实际测试下来效果比预想更好。和一些丝袜 LoRA 相比，服装类 LoRA 的还原度比较高，对画面其他内容的污染也相对更少。

• 类型：蕾丝半透明旗袍 / 服装 LoRA
• 基础模型：SD 1.5
• 触发词：tutututu, see-through, cheongsam
• 推荐起手：tutututu, see-through, cheongsam
• CLIP 终止层：2
• CFG Scale：从 7 左右开始

风格方向

• 以蕾丝半透明旗袍作为主要服装。
• 更偏真实服装还原，而不是泛泛的幻想服装风格。
• 相对减少对人物和整体画面的污染。
• 作为后续图图服装 / 制服系列的起点。
• 后续服装类 LoRA 会尽量统一触发习惯：tutututu，四个 tu。

训练思路

这件衣服的原型来自真实服装参考，训练素材使用真人实拍并经过修图处理，1024x1024 尺寸，bf16 训练细节。

这是我第一次制作这个具体服装。蕾丝本身细节非常多，SD 1.5 的分辨率仍然有限，所以非常细碎的蕾丝细节不一定每次都能完全还原。但整体来看，这个模型的服装方向清晰，画面污染也比较少。

我也发现很多类似服装 LoRA 要么不够贴近现实，要么没有覆盖当下比较流行的衣服，要么对人物和整张图的影响太大。所以我打算继续做一系列服装 LoRA。

推荐参数

• 确保正确加载 LoRA。
• 使用 tutututu、see-through、cheongsam。
• 推荐起手：tutututu, see-through, cheongsam。
• CLIP 终止层数建议设置为 2。
• CFG Scale 可以从 7 左右开始。
• 负面提示词不要堆太多，否则可能影响 LoRA 效果。
• 如果不描述其他服装和环境，LoRA 可能会对它们产生影响。把服装和背景写清楚，可以降低这种影响。
• 原始测试中我没有使用 LoRA Block Weight，但如果需要局部控制，可以尝试。

出图关键词

主触发词组：

tutututu, see-through, cheongsam

• lace semi-transparent cheongsam
• see-through cheongsam
• lace cheongsam dress
• semi-transparent lace fabric
• realistic clothing texture
• cheongsam collar
• delicate lace pattern

如果服装不够明确，请加强 lace semi-transparent cheongsam 和 see-through cheongsam。如果 LoRA 影响背景或其他衣服，请降低 LoRA 权重，并把这些元素描述得更具体。

模型适配

目前测试里，墨幽系列、麦橘系列效果相对不错，似乎对真人模型更加友好。不同大模型仍然需要根据实际效果调参数。

常见问题

为什么蕾丝细节不够锐利？

蕾丝是非常高频的细节，SD 1.5 分辨率有上限。可以补充 delicate lace pattern，并选择更擅长服装细节的底模。

为什么 LoRA 影响其他衣服或背景？

请把其他衣服和环境描述清楚。如果影响仍然太强，降低 LoRA 权重。

权重调高后出现马赛克或纹理破碎怎么办？

先降低本 LoRA 权重。如果同时使用多个 LoRA，也降低其他 LoRA 权重，或者尝试 LoRA Block Weight 做局部控制。

声明

• 你需要对使用本模型生成的任何作品承担全部责任。
• 不要使用本模型制作或传播非法、有害、恶意、诽谤、诈骗或政治用途内容。
• 如果将本模型用于商业用途，或需要定制 LoRA 合作，请提前联系我，谢谢。

日本語説明

これは Tutu sexy uniform / clothing series の最初の LoRA です。テーマは lace semi-transparent cheongsam で、実写素材をレタッチして作りました。

このモデルは、差分抽出ワークフローで clothing LoRA を作るとどうなるか試す意味もありました。実際の結果は予想より良く、hosiery LoRA と比べても outfit restoration が高く、関係ない画像要素への影響が比較的少ないです。

• Type: lace semi-transparent cheongsam / clothing LoRA
• Base model: SD 1.5
• Trigger words: tutututu, see-through, cheongsam
• Recommended start: tutututu, see-through, cheongsam
• CLIP skip: 2
• CFG Scale: start around 7

デザイン方向

• lace semi-transparent cheongsam を主役にすること。
• 現実の服に近い restoration を目指し、曖昧な fantasy costume にしないこと。
• 人物や全体構図への汚染を少なくすること。
• Tutu clothing / uniform series の最初のモデルとして使えること。
• 今後の clothing LoRA の trigger habit は tutututu、four "tu" に寄せる予定です。

制作意図

この衣装の prototype は実物の服です。素材は実写写真を使い、レタッチした上で 1024x1024 size、bf16 training detail で作りました。

この服は初めての制作でした。lace は細かい detail が多く、SD 1.5 の resolution には限界があります。そのため非常に細かい lace pattern は毎回完璧ではありません。それでも outfit direction は明確で、全体への pollution は少なめです。

似た clothing LoRA は多いですが、現実の服に近くないもの、流行りの服をあまり扱っていないもの、人物や画面全体に影響しすぎるものもあります。そのため、今後も clothing LoRA series を作る予定です。

推奨設定

• LoRA が正しく読み込まれていることを確認してください。
• tutututu, see-through, cheongsam を使ってください。
• Recommended start: tutututu, see-through, cheongsam。
• CLIP skip: 2 を推奨します。
• CFG Scale: 7 前後から試してください。
• negative prompt を増やしすぎると LoRA 効果が弱くなることがあります。
• 他の clothing や environment を書かない場合、LoRA がそれらに影響することがあります。服と背景を明確に書くと影響を減らせます。

プロンプト例

Main trigger group:

tutututu, see-through, cheongsam

• lace semi-transparent cheongsam
• see-through cheongsam
• lace cheongsam dress
• semi-transparent lace fabric
• realistic clothing texture
• cheongsam collar
• delicate lace pattern

衣装が弱い場合は lace semi-transparent cheongsam と see-through cheongsam を強めてください。背景や他の服に影響する場合は LoRA weight を下げ、それらを明確に書いてください。

FAQ

なぜ lace detail が弱いですか？

lace は高周波 detail で、SD 1.5 resolution には限界があります。delicate lace pattern を追加し、clothing detail に強い base model を使ってください。

なぜ他の服や背景に影響しますか？

他の服と environment を明確に書いてください。まだ強い場合は LoRA weight を下げてください。

weight を上げて texture が崩れる場合は？

まずこの LoRA の weight を下げてください。複数 LoRA を使う場合は他の LoRA weight も下げるか、LoRA Block Weight を試してください。