FLUX.2 [klein] 9B 场景保留换装 LoRA

中文介绍

这是一个基于 FLUX.2 [klein] 9B 训练的双参考图换装 LoRA。

它可以将 Image 2 中的服装迁移到 Image 1 的人物身上，同时尽可能保留 Image 1 原有的人物身份、面部特征、发型、姿势、动作、表情、构图、背景和光线。

本模型主要用于服装迁移和虚拟换装，不是人物替换或换脸模型。

参考图片的作用

• Image 1： 提供需要保留的人物身份、面部特征、发型、体型、姿势、动作、表情、镜头角度、构图、背景和光线。

• Image 2： 提供需要迁移的服装款式、剪裁、颜色、材质、图案、轮廓和服装配饰。

主要特点

• 将 Image 2 中的服装迁移到 Image 1 的人物身上

• 尽量保留 Image 1 的人物身份和面部特征

• 尽量保留 Image 1 的发型、姿势、动作和表情

• 保留 Image 1 的镜头角度、构图、背景和光线

• 减少 Image 2 中人物、发型、姿势和背景的泄漏

• 改善服装颜色、材质、图案、剪裁和整体轮廓的迁移效果

使用方法

1. 将需要换装的人物图片加载为 Image 1。

2. 将目标服装参考图片加载为 Image 2。

3. 加载本 LoRA。

4. 输入自然语言换装提示词。

5. 根据服装还原度和人物保留效果调整 LoRA 权重。

原始测试提示词

Have the girl in image1 wear the clothes in image2, while keeping the background, movements and lighting unchanged.

推荐提示词

Have the adult woman in image 1 wear the clothes from image 2, while keeping the background, pose, action, facial expression, composition, and lighting of image 1 unchanged.

详细版本：

Have the adult woman in image 1 wear the clothes from image 2. Transfer the clothing design, colors, materials, patterns, cut, and accessories from image 2, while keeping the identity, face, hairstyle, body shape, pose, action, facial expression, camera angle, composition, background, and lighting of image 1 unchanged.

推荐设置

• Base model： FLUX.2 [klein] 9B

• Model type： LoRA

• Reference images： 2

• Image 1： 需要保留的人物和场景

• Image 2： 目标服装参考

• Trigger word： 无需触发词

• Workflow： ComfyUI

• LoRA strength： 使用展示图中的实际权重

如果展示图使用的是 1.0 权重：

• 推荐权重： 1.0

• 建议范围： 0.8–1.2

较低权重通常更容易保留原人物和原场景，但服装迁移效果可能较弱。较高权重可能增强服装迁移效果，但也可能增加服装变形、人物身份偏移或参考图元素泄漏。

适用场景

• 写实人物换装

• 虚拟试衣

• 连衣裙、衬衫、外套和传统服饰迁移

• 服装颜色、材质和图案迁移

• 保留原人物身份的服装编辑

• 使用人物穿搭图片作为服装参考

• 使用单独的服装商品图片作为参考

展示图说明

每组对比图片包含：

• Image 1： 需要换装的人物和需要保留的原始场景

• Image 2： 需要迁移的目标服装参考

• No LoRA： 仅使用 FLUX.2 [klein] 9B 底模生成

• + LoRA： 在相同条件下加载本 LoRA 后生成

No LoRA 和 + LoRA 使用相同的输入图片、提示词、种子、分辨率和工作流，仅改变是否加载本 LoRA。

部分展示测试素材来自 Z-Image-Turbo 返图区或社区展示区。相关图片均为 AI 生成内容，仅用于模型功能测试、技术比较和效果展示。

English Description

This is a two-reference outfit-transfer LoRA trained for FLUX.2 [klein] 9B.

It transfers the clothing from Image 2 onto the person in Image 1, while preserving as much of the original identity, facial features, hairstyle, pose, action, expression, composition, background, and lighting from Image 1 as possible.

This model is designed for outfit transfer and virtual try-on. It is not a character-replacement or face-swap model.

Reference Image Roles

• Image 1: Provides the identity, facial features, hairstyle, body shape, pose, action, expression, camera angle, composition, background, and lighting that should be preserved.

• Image 2: Provides the target clothing design, cut, colors, materials, patterns, silhouette, and clothing accessories.

Main Features

• Transfers the outfit from Image 2 onto the person in Image 1

• Preserves the identity and facial features from Image 1

• Preserves the original hairstyle, pose, action, and expression

• Preserves the camera angle, composition, background, and lighting

• Reduces identity, hairstyle, pose, and background leakage from Image 2

• Improves clothing color, material, pattern, cut, and silhouette transfer

How to Use

1. Load the person to be dressed as Image 1.

2. Load the target clothing reference as Image 2.

3. Enable this LoRA.

4. Enter a natural-language outfit-transfer prompt.

5. Adjust the LoRA strength according to clothing accuracy and identity preservation.

Original Test Prompt

Have the girl in image1 wear the clothes in image2, while keeping the background, movements and lighting unchanged.

Recommended Prompt

Have the adult woman in image 1 wear the clothes from image 2, while keeping the background, pose, action, facial expression, composition, and lighting of image 1 unchanged.

Detailed version:

Have the adult woman in image 1 wear the clothes from image 2. Transfer the clothing design, colors, materials, patterns, cut, and accessories from image 2, while keeping the identity, face, hairstyle, body shape, pose, action, facial expression, camera angle, composition, background, and lighting of image 1 unchanged.

Recommended Settings

• Base model: FLUX.2 [klein] 9B

• Model type: LoRA

• Reference images: 2

• Image 1: Original person and scene to preserve

• Image 2: Target clothing reference

• Trigger word: No trigger word required

• Workflow: ComfyUI

• LoRA strength: Use the actual value shown in the showcase examples

If the showcase images were generated at strength 1.0:

• Recommended strength: 1.0

• Suggested range: 0.8–1.2

Lower strengths may preserve more of the original subject and scene, but the clothing transfer may be weaker. Higher strengths may improve the outfit transfer but can also increase garment distortion, identity drift, or reference-image leakage.

Suitable Use Cases

• Photorealistic outfit transfer

• Virtual try-on

• Dresses, shirts, jackets, and traditional clothing

• Clothing color, material, and pattern transfer

• Clothing edits that preserve the original identity

• References showing a person wearing the target outfit

• Standalone clothing product references

Showcase Format

Each comparison contains:

• Image 1: The original person and scene to preserve

• Image 2: The target clothing reference

• No LoRA: Generated with the FLUX.2 [klein] 9B base model only

• + LoRA: Generated under the same conditions with this LoRA enabled

The No LoRA and + LoRA comparisons use the same input images, prompt, seed, resolution, and workflow. The only difference is whether this LoRA is enabled.

Some showcase references were obtained from the Z-Image-Turbo community return-image or gallery section. All displayed references are AI-generated and are used only for model evaluation, technical comparison, and demonstration.

免责声明 / Disclaimer

本资源仅用于 AI 图像编辑研究、学习、测试和合法创作。

使用者应确保自己对所有输入人物图片、服装参考、服装设计、品牌标志、图案和生成结果拥有必要的权利、许可或授权。

不得将本模型用于欺骗、诈骗、冒充、骚扰、诽谤、身份盗用、侵犯隐私、非自愿私密内容、涉及未成年人的不当内容或其他违法用途。

本模型不会自动向使用者授予参考服装、品牌、标志、角色设计、图案或其他第三方素材的版权、商标权或商业使用权。

生成结果可能包含服装结构错误、身体轮廓变化、人物身份偏差、图案错误、标志变形、手部瑕疵或其他非预期结果。公开发布或商业使用前，请进行人工检查。

展示素材仅用于非商业技术测试和模型效果比较。上传者不对原始展示素材主张所有权，相关权益归原上传者或相应权利人所有。

使用者还应遵守 FLUX.2 [klein] 9B 的底模许可证、Civitai 平台规则以及所在地适用法律。

This resource is intended only for AI image-editing research, learning, testing, and lawful creative use.

Users must possess the necessary rights, permissions, or authorization for all input portraits, clothing references, clothing designs, trademarks, patterns, and generated outputs.

The model must not be used for deception, fraud, impersonation, harassment, defamation, identity theft, privacy violations, non-consensual intimate content, inappropriate content involving minors, or any unlawful activity.

This model does not grant copyright, trademark, merchandising, or commercial rights to clothing designs, brands, logos, character designs, patterns, or other third-party materials shown in the references.

Generated results may contain incorrect garment structures, altered body silhouettes, identity drift, inaccurate patterns, distorted logos, hand artifacts, or other unintended results. All outputs should be manually reviewed before publication or commercial use.

The showcase materials are used only for non-commercial technical testing and model comparison. The uploader does not claim ownership of the original showcase materials. Any applicable rights remain with the original uploaders or respective rights holders.

Users must also comply with the FLUX.2 [klein] 9B base-model license, Civitai platform rules, and all applicable laws.